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: Distributors


The Geological Society Publishing House USA
Unit 7, Brassmill Enterprise Centre AAPG Bookstore
Brassmill Lane PO Box 979
Bath BA1 3JN Tulsa
UK OK 74101-0979
(Orders: Tel. 01225 445046 USA
Fax 01225 442836) (Orders: Tel. (918) 584-2555
Fax (918) 560-2652)
First published 1998
Australia
The publishers make no representation, express or Australian Mineral Foundation
implied, with regard to the accuracy of the information 63 Conyngham Street
contained in this book and cannot accept any legal Glenside
responsibility for any errors or omissions that may be South Australia 5065
made. Australia
(Orders: Tel. (08) 379-0444
9 Geological Society 1998. All rights reserved. Fax (08) 379-4634)
No reproduction, copy or transmission of this
publication may be made without written permission. India
No paragraph of this publication may be reproduced, Affiliated East-West Press PVT Ltd
copied or transmitted save with the provisions of the G- 1/ 16 Ansari Road
Copyright Licensing Agency, 90 Tottenham Court New Delhi 110 002
Road, London WlP 9HE. Users registered with the India
Copyright Clearance Center, 27 Congress Street, (Orders: Tel. (11) 327-9113
Salem, MA 01970, USA: the item-fee code for this Fax (11) 326-0538)
publication is 0305-8719/98/$10.00.
Japan
British Library Cataloguing in Publication Data Kanda Book Trading Co.
A catalogue record for this book is available from the Cityhouse Tama 204
British Library. Tsurumaki 1-3-10
Tama-shi
ISBN1-86239-0169 Tokyo 206-0034
ISSN0305-8719 Japan
(Orders: Tel. (0423) 57-7650
Fax (0423) 57-7651)

Typeset by Bath Typesetting, Bath.

Printed by The Alden Press, Osney Mead, Oxford,


UK.
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 Dresser Atlas 1992). These measured values are
subsurface correlation and gamma ray log curve then recalculated to estimate the proportions of
shapes or signatures are often used as the basis potassium, thorium and uranium, expressed as
for interpreting ancient sedimentary environ- percentages or API units.
ments (Selley 1978; Cant 1992). The spectral Spectral gamma ray data recorded from
gamma ray tool measures radiation produced by outcrop have been used for correlation and to
the radioactive decay of naturally occurring define sediment facies in Upper Carboniferous
radioactive elements. The most common natu- deltaic sediments (Myers & Bristow 1989;
rally occurring radioactive elements in sedimen- Davies & Elliot 1995). Spectral gamma ray data
tary rocks are potassium, thorium and uranium. have also been used to characterize marine
As each of these elements decay they give off bands in the Upper Carboniferous (Archard &
gamma radiation of a particular energy mea- Trice 1990; Leeder et al. 1990). In this study we
sured in MeV (millions of electron volts). The have attempted to apply the methodology of
principle energies for each element are 1.46 MeV Myers & Bristow (1989) to Carboniferous rocks
for potassium, 0.68MeV for thorium, and 1.12 in the Southern North Sea. We have examined
and 0.98 MeV for uranium (Desbrandes 1985). spectral gamma ray logs from thirteen wells in
The radiation from potassium (K 40) is a single the Southern North Sea (Wells 1-13). Sedimen-
energy while uranium and thorium have a series tary logs of core were available for seven of the
of isotopes producing radiation with a range of boreholes and stratigraphic information showed
energies which overlap (Rider 1986). In addi- that two sandstone units 'A' and 'B' could be
tion, Compton scattering leads to a reduction in correlated between three and six of the wells,
energy and the total gamma radiation is a respectively. Unfortunately due to confidential-
complex spectrum. The spectral gamma ray tool ity agreements we are unable to identify the wells
samples the spectrum around specific energy in question or the names of the correlated units.
levels, 1.46MeV for potassium, 1.76MeV for G a m m a ray logs are affected by hole condi-
uranium and 2.62 MeV for thorium (Rider 1986; tions, 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
2 C.S. BRISTOW & B. J. WILLIAMSON

a decrease in gamma ray response. To provide good separation of curves on log plots, all
some control on data quality, gamma ray potassium values were converted to deci. %.
measurements were plotted against caliper data. Having reduced the log data to cored depth
Another common borehole effect is the use of intervals, element data and element ratios were
KC1 in drilling mud. The potassium in the plotted for each well on cross plots and logs, and
drilling mud produces an increase in absolute between wells on cross plots and box plots.
values of potassium on the spectral gamma ray
log. This is supposed to be corrected in the Observations
processing and we have assumed that the
contractors have made the right corrections to Potassium-thorium cross plots discriminate
the data. However, there appears to have been
no correction for variations in mud chemistry lithofacies
down hole. Of the thirteen wells that we have Cross plots of potassium against thorium,
examined, ten were logged by one contractor thorium against uranium and potassium against
and the remaining three were logged by a second uranium were produced for each cored well. The
contractor (Table 1). Seven of the wells were cross plots provide an easy to read display of the
drilled using an oil based mud, five were drilled range of measurements from each well by facies
using a water based KCI mud and one was and for comparing each facies between wells.
drilled with a salt saturated polymer. The cross plot of potassium against thorium
from Well 1 (Fig. 1) is a typical example; It
Table 1. Summary of well characteristics shows limestones with very little gamma radia-
tion clustered in the lower left-hand corner of
Well Logging Drilling Correlated the plot with cross-stratified sandstones in a field
number contractor mud sandstone around 0.01 deci % potassium and 10ppm
thorium. Finer grained lithologies, silty sand-
1 1 KC1 A
2 1 Oil stones, sandy siltstones and claystones are less
3 1 KC1 well discriminated but all show relatively high
4 2 Oil A potassium and thorium. One intriguing feature
5 2 Oil of this plot (Fig. 1) is the negative values for
6 1 Oil potassium in the limestones. Negative values for
7 ! KC1 potassium are very small, less that 0.005 deci %,
8 1 Oil A and were only found in Well 1; however,
9 1 Oil B negative values for uranium were found in wells
10 1 KC1 B
2, 6 and 7. The negative values indicate a
11 2 polymer B
12 1 KC1 B problem with the algorithm used by the con-
13 1 Oil B tractor to calculate elemental concentrations.
Other wells such as Well 5 (Fig. 2) show clear
discrimination between lithofacies although the
Methods absolute values are different from those in Well
1. Cross-stratified sandstones generally contain
A simple seven category lithofacies scheme was slightly less potassium and less thorium, most
adopted for the cored wells with classification on claystones have relatively high values of potas-
the basis of lithology and sedimentary struc- sium but a few have almost no potassium.
tures: cross-stratified sandstones, silty sand- Changes within lithofacies for a particular well
stones, s a n d y siltstones, claystones, coal, could be due to differences in the detrital
limestones and rooted beds. After depth match- composition and diagenetic history of the
ing of core and log and corrections for core to claystones. However, low values may also be
log slip, depth intervals for each lithofacies were encountered where the gamma ray response is
defined. Lithological boundaries were picked at averaged across a bed boundary. The resolution
the shoulder of gamma ray curves to take of a gamma ray tool is typically about 3 0 4 0 cm
account of readings which 'smear' across bed depending on the speed that the tool was run
boundaries. The gamma ray log data were then (Rider 1986) and where the sampling interval
assigned a lithofacies classification on the basis coincides with a bed boundary the measurement
of core descriptions. Where core log depths were will not represent either lithology, but a mixture
metric and the wireline log data in feet, the data of two different lithologies. Another possible
were recalculated to metric units. For several of source of error is in the core to log calibration
the wells, potassium values were given in deci. % where reconstruction of a core may lead to small
rather than as percentages. As deci. % units give offsets in the core to log slip.
SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS 3

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 diagenesis are unlikely to produce such a clear


systematic difference. Other possible explana-
tions are that the sandstones were deposited in
On Fig. 3, which shows cross plots of cross- different deltaic environments: mouth bar, dis-
stratified sandstones from all seven cored wells, tributary channel or shoreface; or that the
the data for individual wells form distinct sandstones are stratigraphically different and
clusters. The differences within wells is less than have different detrital sources or different
the differences between wells, which suggests diagenetic histories. One way of testing these
some systematic changes between wells. Geolo- hypotheses is to examine the character of
gic factors such as a change in provenance or correlated sandstones.
4 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. variation between wells could be due to lateral


This has been correlated stratigraphically be- facies changes, but these are unlikely to have
tween three wells. The cross plot of potassium produced the observed shift in absolute values.
against t h o r i u m (Fig. 4) shows the same The similar shape of the trends combined with
sandstone in three different wells plotting in their differences in absolute values indicates a
slightly different areas. There is almost no systematic change between wells which we
overlap between the three data sets and although attribute to changes in the borehole environ-
the trends appear to be similar in each well, there ment. The factors most likely to affect the logs
is a clear difference in the absolute values. Some are caving, the use of different drilling fluids, and
SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS 5

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.
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 differ-
ence? 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
Fig. 6. Box plot of total gamma for sandstones and appear most likely that the correction factor for
claystones. Claystones usually have higher total potassium in the mud has left a residual of
gamma than sandstones although there is some over-
enhanced potassium values. One might wonder
lap in Wells 3 and 6. The lower than usual values in
these claystones may be due to deposition in an why the other Wells (10 and 12), with water
interdistributary bay rather than a prodelta environ- based mud and relatively high KC1 contents, lie
ment. 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
variations in the procedures of different logging appear therefore that the choice of logging
contractors. There is very little difference in contractor can have a significant effect on
caliper data between wells and no evidence for results.
significant caving, which leaves two possible
e x p l a n a t i o n s for the differences observed.
Box plots show differences between wells
Firstly, Well 1 was drilled with water based
mud, while Wells 4 and 8 were drilled with an oil Box plots have been used for a comparison of
based mud. Secondly, Wells 1 and 8 were logged total gamma ray values for cross-stratified
by a different contractor to Well 4. Reduced sandstones and claystones between wells, using
values for potassium in Well 1 are most likely to lithofacies defined from core. Each plot (includ-
be due to an over-correction for potassium in the ing boxes and whiskers) shows the spread of
KC1 drilling mud. observations about the median. The box repre-
6 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, Eliminating inter well differences using ratio
the whiskers extend to the minimum and
maximum data values. Median values for plots
cross-stratified sandstones are generally 50 API Element ratio vs element ratio plots were
units or less, although they do vary between generated to eliminate the systematic variations
wells (Fig. 6). Total gamma ray response for in gamma ray tool response between wells
sandstones is almost always less than the total (usually due to varying well conditions) which
gamma ray response for claystones, where the may have been inadequately compensated for in
median value is close to 100 API units, although logging company calibration procedures. The
there is some overlap in Wells 3 and 6 where the plot of K/Th ratio against K/U ratio for Unit A
claystones have lower total gamma ray response (Fig. 7) shows that measurements from Wells 4
than the other claystones. There is no obvious and 8 overlap while measurements from Well 1
reason for the lower total gamma ray response in are clearly lying on a different trend. Wells 4 and
these two wells. Well 3 was drilled with a water 8 were both drilled with an oil based mud while
based mud, but so were Wells 1 and 7, while Well 1 was drilled with a water based mud
Well 6 was drilled with an oil based mud as were containing KC1. The K/Th cross plot (Fig. 4)
Wells 2, 4 and 5. Wells 3 and 6 are from broadly shows low potassium values for Well 1, and the
similar stratigraphic units but Wells 5 and 7 are ratio plot (Fig. 7) shows an offset due to low
from the same Group. One possible explanation potassium values. In addition, Fig. 7 shows an
is that the claystones in Wells 3 and 6 were exceptionally good correlation between thorium
deposited in slightly different environments. The and uranium. We suspect that the correction
core logs indicate a prodelta environment for factor applied to compensate for KC1 mud in
claystones in Wells 1, 2, 4, and 7 and an Well 1, has over-compensated for potassium and
interdistributary bay environment for claystones also affected the measurements of thorium and
in Wells 3, 6 and 5. Re-examination of the core uranium.
logs indicates that the claystones in Well 5,
originally attributed to an interdistributary bay,
are significantly thicker than other interdistribu- Conclusions
tary bay deposits and could be re-interpreted as
prodelta deposits. If this is the case, then the Lithofacies for Carboniferous deltaic sequences
total gamma ray response is discriminating from the Southern N o r t h Sea have been
between sedimentary environments, not just identified from core descriptions and compared
between lithofacies. with spectral gamma ray logs. The results show
SPECTRAL GAMMA RAY LOG CORRELATION PROBLEMS 7

that lithofacies can be discriminated within References


single wells. However, comparison of correlated
sandstones shows that variations between wells ARCHARD, G. & TRICE, R. 1990. A preliminary
are greater than variations within wells. There is investigation into the spectral radiation of the
too much variation between wells to make an Upper Carboniferous marine bands and its
unequivocal assessment of lithofacies and de- stratigraphic application. Newsletters on Strati-
graphy, 21, 167-173.
positional environment on the basis of spectral
CANT, D. J. 1992. Subsurface facies analysis. In:
gamma ray logs alone. The differences between WALKER R. G. • JAMES, N. P. (eds)Facies
wells are attributed to changes in logging Models, Geological Association of Canada, pp.
environment, mainly mud characteristics, bore- 27-45.
hole quality and different logging companies DAVIES, S. J. 8~ ELLIOT, T. 1995. Spectral gamma ray
which have made detailed correlations impossi- characterisation of high resolution sequence
ble. For sandstones showing low total gamma stratigraphy: examples from upper Carboniferous
ray response, small errors associated with fluvio~leltaic systems, County Clare, Ireland. In:
calibrations and correction factors will make a HOWELL, J. A. 8z AITKEN, J. F. (eds) High
Resolution Sequence Stratigraphy: Innovations
comparatively large difference to results. In three
and Applications. Geological Society Special Pub-
wells, negative values for uranium were noted lications No. 104, pp. 25-35.
and in one well negative values for potassium DESBRANDES, R. 1985. Encyclopedia of well logging.
were found which suggests a problem with the Institut Francais du Petrole, Graham and Trot-
algorithm used to calculate elemental concentra- man Ltd, London.
tions. Cross plots of correlated sandstones DRESSER ATLAS. 1982. Well logging and interpretation
indicate that correction factors for KC1 in techniques (3rd edition). Dresser Industries Inc.,
drilling muds are not always successful, and USA.
there appears to be a difference between the LEEDER, M. R., RAISWELL,R., AL-BIATTY,H., MCMA-
HON, A. & HARDMAN, M. 1990. Carboniferous
results achieved by different contractors in this
stratigraphy, sedimentation and correlation of
respect. Corrections for KC1 appear to be based well 48/3-3 in the southern North Sea Basin:
on a single value for each well although mud integrated use of palynology, natural gamma/
chemistry will almost certainly change down sonic logs and carbon/sulphur geochemistry.
hole. More detailed tool calibration is required Journal of the Geological Society, London, 147,
before subsurface correlations and facies analy- pp. 287-300.
sis can be reliably made using spectral gamma MYERS, K. J & BRISTOW, C. S. 1989. Detailed
ray response alone. The influence of downhole sedimentology and gamma ray log characteristics
environment could be further tested by compar- of a Namurian deltaic succession II: Gamma ray
logging. In: WHATELEY,M. K. C. & PICKERING,K.
ing the geochemical composition of core with
T. (eds) Deltas." Sites and Traps for Fossil Fuels,
gamma ray response. In the meantime avoid Geological Society Special Publications No. 41,
trying to read too much from spectral gamma pp. 81-88.
ray response where KC1 mud is involved. RIDER, M. H. 1986. The Geological Interpretation of
Well Logs, Blackie Halsted Press, Glasgow.
The authors thank Mobil North Sea for funding this SELLEY, R. C. 1978. Concepts and methods of subsur-
work and for permission to publish the results. The face facies analysis. American Association of
manuscript has been improved by the comments of J. Petroleum Geologists, Continuing Education
S. Schweitzer and P. Corbett. 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--measure-
ment 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 including time. The latter, on the other hand,
is a problem of scale in reservoir engineering applies on some coarse grid as a replacement of
(e.g. Warren et al. 1961; Haldorsen 1986). The a fine grid domain, but it may change radically if
volume of a reservoir under production greatly the boundary conditions are changed. It will
exceeds the volume of rock recovered from cores emerge from our discussion that we are fre-
or investigated by wireline logs. There are many quently talking about pseudo properties when
efforts underway to improve the modelling of we refer to core-log data integration.
reservoirs, which particularly address the extra- The petrophysical community have appre-
polation from the sparse core-log data to the ciated for some time that there are also scale-
interwell volumes. Computer flow models of up problems in making comparisons between
reservoirs involve grid blocks that are by core and log data (e.g. Knutson et al. 1961).
necessity large, relative to the investigation However, historical practice relied on the
volumes of core or logs. Therefore engineers sampling of cores with plug-size measurements
have to integrate the core and log data for use in at one-foot spacing (Fig. l a). These were then
simulation models in a process loosely referred compared directly with the log measurements,
to as 'up-scaling'. Permeability is a particular recorded at half-foot intervals. Shifts between
property of interest and several techniques have core depths and log depths accounted for the
been developed for its up-scaling, e.g. power offset (if present) between the core and log.
averaging, renormalization, and pseudo-isation. Occasionally, a primitive up-scaling technique
The aim of up-scaling is to estimate the using a running average ( 1 : 2 : 1 weighting) was
'effective' or equivalent properties at the chosen used for the plug data prior to comparing with
volume scale, e.g. grid blocks. the log data. Although the scale discrepancies
The adjectives 'effective' and 'pseudo' are were often appreciated, there was not much else
often used interchangeably in the petroleum that could be done.
literature to denote an up-scaled property, but The development of high resolution petro-
there is a subtle difference. The former attempts physical measurements in the laboratory (probe
to be intrinsic to the rock/fluid system and aims permeameter) and downhole (image logs) has
to be independent of boundary conditions, 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. 2. A comparison between (a) Up-scaling and (b)


Cross-scaling. Numbers refer to approximate volumes
Fig. 1. A comparison between (a) the traditional core in cubic metres. Refer to the text for definition of these
plug and logging tool scales of measurement and terms.
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 being developed. This approach is tied directly
section. In (b) there is more scope for identifying small to the needs of reservoir modeUers and offers a
scale heterogeneities and less sensitivity to missing core way of integrating data and procedures from the
material. Depth matching is also improved. original geological conceptual model, through
the petrophysical data acquisition, the up-scal-
ing/cross-scaling, and the construction of the
issues between core and log measurements. numerical reservoir simulation model.
These high resolution measurements image the
geology far more effectively than the conven- Definitions
tional, low resolution devices. Indeed, image
logs were developed specifically to image the In this paper, we define the terms up-scaling and
geology in the subsurface, potentially replacing cross-scaling as follows (Fig. 2):
the need for core data. With data at high
sampling densities and small volumes of mea- Up-scaling: The determination of an effective
surement, the comparison between logs and (or pseudo) property at a scale larger than
cores becomes more tractable (Fig. lb) and that of the original measurement. An exam-
therefore gives us a feasible approach to core- ple would be using the arithmetic average of a
log scaling. Since the laboratory probe permea- set of layer permeabilities as an estimator of
meter measures a different physics (gas flow rate) the horizontal permeability of the composite
to a subsurface image log (acoustic reflection or layered media (Jensen et al. 1997, pp. 137-
electrical conductivity) with different boundary 139). Comparing probe to plug to well test
conditions, there are also cross-scaling relation- permeabilities is an up-scaling problem (Cor-
ships (see below) that must be considered in bett et al. 1996a).
addition to volume scale and sampling density
effects. The issue of measurement scale for the same
In this paper, we illustrate the cross-scaling petrophysical property is the process of up-
and up-scaling of permeability between core and scaling. Reservoir engineers are familiar with the
wireline logs for subsurface prediction of perme- up-scaling of permeability for reservoir simula-
ability. Larger scale dynamic data are used to tion. Cross-scaling is a much less familiar
justify the methods presented. Having reviewed concept and may be defined as follows:
the method, we discuss the implications for
other properties and outline a new approach to Cross-scaling: The determination of a rela-
petrophysics--genetic petrophysics--which is tionship between two different physical prop-
A REVIEW OF UP-SCALING AND CROSS-SCALING 11

erties. Using regression to summarize the


relationship between porosity and permeabil-
ity 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 measure-
ment 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 Fig. 3. Measurement of properties in the laboratory at
inappropriateness--of the surrogate property. similar volume scales with a resistivity probe (above)
The definition of these terms helps us distin- and permeability probe (below). Refer to Jackson et al.
guish the impact of geology (largely up-scaling) (1994) for more details.
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 up-
scaling of permeability can be achieved in
practice.

Case studies
We consider three examples of the cross-scaling
between permeability and resistivity which have Fig. 4. Correlation between resistivity (shown a
been carried out and which have been reported formation resistivy factor= measured resistivity/brine
in the literature. resistivity) against probe permeability for a slab of
Lochabriggs Sandstone.
Laboratory study
Jackson et al. (1994) measured permeability and (Fig. 5a) and provided the basis of a perme-
resistivity with probe devices for an aeolian ability predictor which was a considerable
sample that was saturated with brine in the improvement over methods based on the density
laboratory (Fig. 3). The resistivity probe was log and core plugs (Fig. 5b).
carefully designed to investigate a volume
similar to that of a steady state probe permea- Morecambe Bay study
meter and both volumes were comparable to the
sample's scale of sedimentary variation. A Thomas et al. (1996, 1997) undertook a detailed
strong relationship was observed (Fig. 4) and probe study over a fluvial interval for which
this can be related to the fundamental physical resistivity images had also been acquired. They
control. found a strong correlation between the probe
permeability and microscanner resistivity (Fig. 6).
As-Sarah study In all of these cases, an empirical relationship
existed and was reflected in the measurements at
Ball et al. (1997) carried out a probe permea- similar scales. Such relationships could reflect an
meter study on a fluvial sandstone. They found underlying physical relationship, explained by
that averaged probe data (at 10cm spacing) an existing analytical model (e.g. Biot's and the
correlated reasonably well with microresistivity 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 Fig. 6. Comparison between probe permeability,
(averaged over a 30 cm window) and micro-spherically formation image and FMI resistance for an interval
focussed log (MSFL) resistivity and (b) plug perme- of Sherwood Sandstone. Refer to Thomas et al. 0996,
ability and wireline density for an interval of PUC-B 1997) for more details.
Reservoir. Refer to Ball et al. (1997) for more details.

1994; Thomas et al. 1996) show very strong


insight into the need for new petrophysical resistivity-permeability relations, while lamina-
analysis. set measurements (Ball et al. 1997) exhibit a
weaker, though still useful, relationship. In all
Cross-scaling permeability and resistivity cases, the effects of geological variation were
mitigated by chosing similar measurement vo-
In the three studies just mentioned, the relation- lumes.
ship is driven by the effects of pore geometry and
porosity upon both the hydraulic and electrical Up-scaling permeability
conductivities. Several workers (e.g. Doyen
1988; Katz and Thompson 1987) have shown In two of the cases presented above, the
that both transport properties depend on a permeability was up-scaled for comparison with
characteristic pore size in the rock. The form some larger scale dynamic data.
of that dependency differs for hydraulic and In the As-Sarah study, the permeability
electrical conductivity, thus making the hydrau- predictor developed from the microresistivity
lic--electrical relationship strength dependent was used to predict permeability in the uncored
also upon the level of rock heterogeneity. In a sections of several wells. With a continuous
homogeneous sample, one characteristic length permeability log, the cumulative permeability-
and its mutual effects upon both permeability thickness product, the transmissivity, was com-
and conductivity will give rise to a strong pared with a production log spinner survey. A
electrical-hydraulic relationship. Heterogeneity, good comparison was found supporting the
however, will diminish the relationship strength appropriateness of the predictor (Fig. 7). This
because different portions of a sample will have predictor continues to form the basis for
differing characteristic sizes. This explains why permeability models in the field (von Winterfeld,
data at the lamination scale (e.g. Jackson et al. pers. comm.).
A REVIEW OF UP-SCALING AND CROSS-SCALING 13
-11975" : -11900

-12025" ~ P r o b e -11950" ~ Probe

~ -12125 ~~ Cum. Prob~


-12000- ~i

-12050 -
PLT

i I
~LT
-12175'

-12225 , , , ~ 9 , 9 , , , , , 9 , 9 -12150 , , , , ' , 9 , 9 , 9


0 50 100 150 200 250 300 350 400 0 250 500 750 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 properties (effective kv/kh ratio) compared well
t t with a larger-scale dynamic measurement (Fig. 8).
The importance of the geology in the up-
I scaling is well illustrated in two ways in this

'~ ~ I~T ~
~ enrlda~:; % o P ~ : 2 a ttimates
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
~. 0.001t I I
of similar beds in outcrop, that the layers or beds
o 5 10 15
observed at the wellbore extend well beyond the
Measurement interval (ft) volume of investigation of the dynamic measure-
ment. This is in contrast to the anisotropy shown
Fig. 8. Validation of probe/FMI resistance predictor by plug scale measurements which is notably
for kv/kh with pressure data from a Modular Dynamic poor in estimating effective kv/kh at larger scales.
Tool (MDT) for an interval of the Sherwood Averaging plug scale kv/kh ratios is also an
Sandstone. The probe estimator uses the harmonic
average (over a moving window--to represent vertical inappropriate up-scaling method for this para-
permeability over a measurement interval) divided by meter, which is very sensitive to scale changes
the arithmetic average (horizontal permeability) over (Corbett et al. 1996b, Cowan 7 Bradney 1997).
the same expanding window. Refer to Thomas et al. The comparison of up-scaled permeability
(1996, 1997) for more details. (probe, plug, or wireline) with the well test can
provide additional corroboration of permeabil-
ity predictors. For these larger scales, the effects
In the Morecambe Bay study, the up-scaling of the organization of the geology (i.e. sedimen-
of both horizontal (kh) and vertical (kv) perme- tary structure) can also be important. This level
ability were required for comparison with a of up-scaling is beyond the scope of this review
borehole pressure measurement. The horizontal (refer to Corbett et al., 1996a). Nonetheless, it is
permeability was up-scaled by taking the arith- important to note that up-scaling from core to
metic average of the probe data, the vertical log must be tied with a consistent geological
permeability by taking the harmonic average framework to the scales of well tests and full
(Thomas et al. 1996, 1997). The up-scaled field numerical grid blocks.
14 P.W. M. CORBETT E T AL.
4"
Plug permeability - density log cross-scaling
revisited
We can revisit the As-Sarah example to compare
~3-
;~ 2" 9 '.
9 . .ii

J
~-.
9 ] : 9

i,iii.
the probe-microresistivity method with the plug-
9 9

density method for permeability prediction 9 This 0


will reveal the nature of improvements provided ...
2-047
r -1
to the petrophysicist by the smaller scale
measurements 9 It seems ironic that solutions to
the up-scaling problem have been facilitated (i.e. 0 ' ;0 ~0 3'0
they are more accurate, not necessarily faster) by Porosity (%)
having more 'smaller' scale petrophysical mea- Fig. 9. Core plug porosity and permeability relation-
surements. This irony, however, overlooks the ship for the PUC-B Sandstone. This type of relation-
role of the geology in the scaling process: ship is typical in texturally heterogeneous fluvial
smaller-scale measurements are often more reservoirs. Clay content and cementation variations
easily interpreted in their geological context 9 at plug scale due to clay drapes and rhizocretions also
The geology provides information regarding the impact these data. These factors combine adversely to
volume and shape of each event, allowing make a complex relationship between permeability and
analysts to make inferences about the validity porosity, one which cannot be used with any
confidence for permeability prediction. A more textu-
and frequency of the value in the unsampled rally sensitive surrogate property is needed and was
regions 9 provided (in this study) by the MSFL resistivity 9Refer
If we examine the porosity-permeability to Brayshaw et al. (1996) for more discussion on the
relationship (Fig. 9) for the As-Sarah reservoir, textural controls on permeability and to Ball et al.
we see that it is very weak. The lack of (1997) for more details of the PUC-B study.
relationship is due to a number of factors--
variable grain size and sorting in the fluvial Genetic petrophysics
sediments, patchy rhizocretionary cements, plug
orientation with respect to heterogeneities, and The Morecambe Bay example shows the power
others. Weak porosity-permeability relation- (for prediction) of scale-compatible cross-scaling
ships in fluvial reservoirs are often observed and geologically-assisted up-scaling. Fig. 8
(Brayshaw et al. 1996). The cross-scaling rela- shows that the effective property (in this case,
tionship, in this case, is strongly obscured by the k v / k h ) varies at certain geological length scales.
geological heterogeneity--a smaller or larger There is a significant and abrupt change at the
volume scale is suggested or separation of the bed scale (4ft) and the bedset scale (12ft). Above
grain size classes (Hogg et al. 1996). Any the bedset scale, there appears to be less
porosity-permeability relationship from these variability in the estimates and close agreement
data will be associated with a high degree of with the Modular Dynamic Tool (MDT) re-
uncertainty if used to predict permeability 9 sponse. While the cost implications of MDT
On the (weak) assumption that porosity and versus image log have to be considered, image
permeability are related, the wireline density log based predictors, calibrated by MDT mea-
derived porosity might be used to predict surements at carefully selected intervals, hold
permeability. The density log has a volume of potential for improved anisotropy estimates in
investigation that is larger than the small scale the future. Anisotropy in sediments is strongly
(lamination) textural features that control per- affected by bedding, so it is only appropriate
meability. In this example, it also proved very that a predictor based on a log that 'sees' the
difficult to depth match the plug data with the bedding will be better than estimates from small
log data, the probe data were more useful in this volume, plug measurements.
respect. The 'true' variation in permeability The length scales (i.e. the geological architec-
shown by the probe did not correlate well with ture) provide important guidance for the petro-
the poor resolution of the density log. Any physicist--the length scales for combining or
permeability predictor based on the latter will comparing appropriate measurements and also
eliminate a scale of heterogeneity that may be the length scales to be avoided for sampling
important to the sweep efficiency of the reser- intervals. Sampling close to the frequency of the
voir. The up-scaling of permeability, if this data (volume or wavelength) is a notoriously
method had been followed, would result in a poor procedure in geophysical measurements.
more uniform reservoir permeability field, which Unfortunately, the 1-inch plug size and 1-foot
may have been inappropriate for modelling oil sampling interval are close to the Nyquist
recovery 9 frequency of lamina and beds!
A REVIEW OF UP-SCALING AND CROSS-SCALING 15

Smaller scale measurements mean more data References


and more work in reducing, summarizing, and
integrating the data. In that respect, their BALL, L. D., CORBETT,P. W. M., JENSEN,J. L. & LEWIS,
development is not welcome in the time- and J. M. 1997. The role of geology in the behavior
personnel-challenged climate demanded by in- and choice of permeability predictors. SPE
dustry. However, the very fact that there are Formation Evaluation, 12, 32-39.
BRAYSHAW,A. C., DAVIES,G. W. • CORBETT,P. W. M.
representative elements within reservoirs (e.g.
1996. Depositional controls on primary perme-
stratal elements, genetic units, architectural ability and porosity at the bedform scale in fluvial
elements) can also be exploited. The fundamen- reservoir sandstones. In CARLING, P. A. &
tal rock measurements can be targetted on these DAWSON, M. R. (eds) Advances influvial dynamics
representative elements. The effective properties and stratigraphy, John Wiley & Sons, 373-394.
(or even pseudos) can then be determined--by CORBETT, P. W. M., RINGROSE,P. S., JENSEN, J. L. &
simulation or measurement for these elements SORBIE,K. S. 1992. Laminated clastic reservoirs--
(Corbett et al. 1992; Ringrose et al. 1993; Pickup The interplay of capillary pressure and sedimen-
et al. 1995; Huang et al. 1995). The modelling of tary architecture. SPE 24699. Proceedings of the
67th SPE Annual Technical Conference and
these elements--geobodies in G O C A D - - c a n
Exhibition, October, Washington, 365-376.
then be accompanied by the appropriate petro- --, PINISETTI, M., TORO-RIVERA, M. & STEWART,
physics--genetic petrophysics. This method in- G. 1996a. The comparison of plug and well test
volves a more selective use of petrophysical permeabilities, Dialog, 4-8.
measurement which is intended to be more cost- --, GOOD, T., JENSEN, J. L., LEWIS, J. J. M.,
effective. Indeed, a genetic petrophysics ap- PICKUP, G., RINGROSE, P. S. & SORBIE, K. S.
proach which explicitly recognizes and solves 1996b. Reservoir description in the 1990s: A
the cross-scaling problem may be the only perspective from the flow simulation through
successful route to true data integration. layercake parasequence flow units. In: GLENNIE,
K. & HURST, A. (eds), AD 1995: N W Europe's
Hydrocarbon Industry, Geological Society, Lon-
Conclusions don, 169-178.
COWAN, G. & BRADNEY,J. 1997. Regional diagenetic
Cross-scaling between petrophysical properties controls on reservoir properties in the Millom
is best achieved when the scales and density of accumulation: implications for field development.
measurements are comparable. In: MEADOWS,N. S., TRUEBLOOD,S. P., HARDMAN,
Up-scaling of petrophysical properties bene- M. & COWAN,G. (eds) Petroleum Geology of the
fits when the geological architecture is accounted Irish Sea and Adjacent Areas. Geological Society,
for. London, Special Publications, 124, 373-386.
DOYEN, P. M. 1988. Permeability, conductivity, and
Probe data and image logs can be jointly used
pore geometry of sandstone. Journal of Geophy-
to predict permeability (horizontal and vertical sical Research, 93, 7729-7740.
in the subsurface), demonstrating that consis- HALDORSEN, H. H. 1986. Simulator parameter assign-
tent-volume cross-scaling and geologically-con- ment and the problem of scale in reservoir
strained up-scaling can be effective. engineering. In: LAKE, L. W. & CARROLL, H. B.
The tools, understanding, and techniques are (eds), Reservoir Characterisation, Academic Press,
now available for the development of a more Orlando.
geologically-based petrophysics m e t h o d - - w h i c h HOGG, A. J. C., MITCHELL,A. W. & YOUNG, S. 1996.
we refer to as genetic petrophysics--that is fit for Predicting well productivity from grain size
analysis and logging while drilling. Petroleum
the purposes of reservoir modelling. It is under- Geoscience, 2, 1-15.
stood that improved modelling prepares the way HUANG, Y. RINGROSE, P. S. & SORBIE, K. S. 1995.
for improved oil recovery--the ultimate motiva- Capillary trapping mechanisms in water-wet
tion behind this work. laminated rocks. SPE Reservoir Engineering, 10,
The authors acknowledge the support of Wintershall 287-292.
and British Gas in the studies discussed above. They JACKSON,M. A., BOWEN,D. G., JENSEN,J. L. & TODD,
also wish to acknowledge the support of EPSRC and A. C. 1994. Resistivity and permeability mapping
industrial co-sponsors (Amerada Hess, Amoco, BHP at the lamina scale. Proceedings of the Interna-
Petroleum, British Gas, Chevron, Fina, Saga, Schlum- tional Symposium of the Society of Core Ana-
berger, Shell, Statoil, Texaco) for continued work in lysts, Stavanger, 12-14 Sept., paper SCA-9415,
this area under the PEGASUS project. The authors 163-172.
also acknowledge the contributions of the various JENSEN, J. L., LAKE, L. W., CORBETT, P. W. M. &
authors of the case studies from which this overview GOGGIN, D. J. 1997. Statistics for Petroleum
has been drawn L. Ball, J. Lewis, S. Thomas, D. Engineers and Geoscientists, Prentice-Hall, New
Bowen and M. Jackson. Their insights while working Jersey.
with the data have helped to formulate and illustrate KATZ, A. J. & THOMPSON, A. H. 1987. Prediction of
these concepts. rock electrical conductivity from mercury injec-
16 P. W. M. CORBETT ET AL.

tion measurements. Journal of Geophysical Re- Field: a numerical approach using probe permea-
search, 92, 599-607. metry. In: OAKMAN, C. D., MARTIN, J. H.
KNUTSON, C. F., CONLEY, F. R., BOHOR, B. F. & CORBETT, P. W. M. (eds) Coresfrom the Northwest
TIMKO, D. J. 1961. Characterization of the San European Hydrocarbon Province." An illustration of
Miguel Sandstone by a coordinated logging and geological applications from exploration to devel-
coring program. Journal of Petroleum Technology, opment. Geological Society, London, 197-203.
13, 425-432. - - & JENSEN,J. L. 1996. Permeability and
PICKUP, G. E., RINGROSE, P. S., CORBETT, P. W. M., permeability anisotropy characterization in the
JENSEN, J. L. ~; SORBIE, K. S. 1995. Geology, near well-bore: a numerical model using probe
g e o m e t r y , and effective flow. Petroleum permeability and formation micro-resistivity data,
Geoscience, 1, 37-42. Transactions of The Society of Professional Well
RIN~ROSE, P. S., SORBIE, K. S., CORBETT, P. W. M. & Log Analysts Thirty-Seventh Annual Logging
JENSEN, J. L. 1993. Immiscible flow behaviour in Symposium, New Orleans, 16-19 June, paper JJJ.
laminated and cross-bedded sandstones. Journal WARREN,J. E., SK1BA,F. F. & PRICE, H. S. 1961. An
of Petroleum Science and Engineering, 9, 103-124. evaluation of the significance of permeability
THOMAS, S. D., CORBETT, P. W. M. & JENSEN, J. L. measurements. Journal of Petroleum Technology,
1997. Permeability anisotropy estimation within 13, 739-744.
the Sherwood Sandstone, Morecambe Bay Gas
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 different penetrabilities to X-rays. As a result
established role in non-destructive imaging of the linear scanner produces an image which
both slabbed and whole core and has been reflects the density variation in the section
routinely used in visual assessment and quality analysed (Tolansky 1961).
control of material being subjected to other A project carried out on 108ft of hetero-
physical measurements (for example Algeo et al. geneous sediments (Duncan et al. 1996) has
1994; Rigsby et al. 1994). Since core may be shown that a digital measure of the X-ray
observed in real time, whole core can be oriented penetrability values ('luminance') can be ex-
to maximum dip prior to plugging or slabbing, tracted in order to produce a surface density
especially useful where the core has been resin- variation log.
stabilized within fibreglass, pvc or aluminium These X-ray luminance values may yield data
liners. This ability to examine interactively, in at close and equally spaced points producing a
detail and non-destructively, the 3-D nature of log with significant advantages over the data
the internal structure of the core material is from conventional core analysis (where sample
particularly important. Linear scanning is there- spacing may be irregular, widely spaced and
fore useful in the observation of both hetero- lithologically chosen, or where Gamma Ray
geneous and apparently homogenous lithologies response may be poor). Such data can be
and the following features are commonly char- compared directly with the wireline logs and it
acterized: is found that the X-ray luminance values show a
linear relationship with the downhole Formation
(a) bedding features and sedimentary struc- Density Log (FDL). The X-ray luminance data
tures; may therefore provide an accurate tool for the
(b) bioturbation (ichnofacies analysis), espe- correlation of core density with log density.
cially in slabbed sections;
(c) identification of remnant structure (not
readily visible to the naked eye) which has Database
been obscured by bioturbation; The Scott partner group provided access to a
(d) natural and coring-induced fractures and range of core and associated materials:
shears (cemented/uncemented/open);
(e) cement distribution; (a) 108ft of lithologically/mineralogically
(f) small scale grain size variation; variable sediments (resinated archive
(g) assessment of resin competence in pre- slabs);
served and/or sleeved core. (b) wireline logs for the analysed interval
including the appropriate FDL traces;
These features are distinguished by their (c) the sedimentological composite log;

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

(d) petrographic data for the seven thin again assigned to Piper Formation Depositional
section samples which fall within the Unit 4b. They also consist of interbedded fine or
analysed interval; very fine grained sandstones and silty, argillac-
(e) core analysis data (porosity, permeability eous deposits. The modal grain size of the
and grain density) for the analysed interval. sandstones is seen to decrease towards the lower
part of the analysed section. Pervasive calcar-
Brief description of cores eous and dolomitic, nodular cements are locally
common. Thin section analysis indicates that the
Section A predominant cement is dolomite (42.5-48%)
with subordinate calcite (3.5-5%) and minor
The analysed interval commences within rela- quartz (0.5-2%). Authigenic kaolinite accounts
tively 'clean', blocky, medium grained sand- for only 0.5%.
stones. These sandstones are assigned to the
Piper Formation Depositional Unit 4c and are Methodology
interpreted to be shoreface sandstones, possibly
representing gully fill within an upper delta front Production of the X-ray scan images
system. At a depth of 6 ft below top of section
these deposits are underlain by variably argillac- A schematic representation of the scanning
eous sandstones with sandy, argillaceous silt- system is shown in Fig. 1.
stone interbeds. The sandstones are generally Although the imaging system has been devel-
fine grained and are frequently apparently oped to operate with core material of various
structureless or faintly laminated. Current rip- forms and dimensions, the present investigations
ples and burrows are locally observed. Bioturba- employed 3 ft resinated archive slabs for the
tion is more commonly observed in the finer imaging and quantitative density measurements.
grained units, some units are micaceous and This presents a thickness of rock material for
locally contain carbonaceous material. Nodular analysis which is relatively constant, both across
calcareous cement is observed at approximately the core diameter and along its length, and for
36 ft below top of section. These lower deposits which the 3-D inhomogeneities are reduced. In
are assigned to Piper Formation Depositional this way, differences in the core thickness and
Unit 4b and are interpreted to belong to an variability due to the curvature of the core are
offshore transition zone. They are believed to be reduced and interpretation can be simplified to
the deposits from turbidite flows in the lower essentially 2-D.
delta front to pro-delta. Thin section analysis The X-rays passing through the rock create an
indicates that the predominant cement is quartz inverted image of the material on an electronic
(8-11.5%) with relatively minor calcite (1- image intensifier. This visible image of the X-ray
3.5%). Authigenic clays are dominated by field is picked up by a CCD camera and
kaolinite (0-3 %). subsequently digitized. This image may be
viewed in real time (i.e. the 'live' image on
Section B screen moves as the rock is transported along the
gantry) and approximately 6-7in of core are
The sediments within Section B are also assigned observed at any one time within the camera
to Piper Formation Depositional Unit 4b and image frame. These frames may be enhanced by
similarly consist of relatively clean, fine grained a dedicated image processing computer and can
sandstones interbedded with siltier, argillaceous be combined to produce a composite image of
sediments which are moderately to highly the 3 ft section.
bioturbated. The finer material is frequently After positioning the core, each 'live' frame is
micaceous and carbonaceous debris is locally frozen and a digital filter, which enhances the
recorded. Some of the coarser units show edge and structure information, is used to
development of calcareous cement which is sharpen the image. Once the optimum image
locally nodular. Thin section analysis indicates has been captured it is electronically transferred
that the predominant cement is calcite (4-40%) to a PC computer terminal and stored as a TIFF
with subordinate quartz (1-12%). No authigenic format file. Overlaps of approximately lin
clays are recorded from the two samples between neighbouring frames are used in order
analysed. to ensure the optimum matching in the compo-
site. The individual images are manipulated on
Section C the PC, using conventional image processing
software, to produce the composite image,
The sediments analysed from Section C are 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 real time, at any given point across the image.
printer matching the resolution of the digitized Thus, by taking regularly spaced readings, a
images. profile of the variation in the grey scale can be
To further ensure accurate and straightfor- produced.
ward matching of each frame to form the 3 ft These data (referred to as luminance values)
composite image, a steel mesh with a grid of indicates the penetrability of the rock material to
V2in is placed alongside each resin slab as it is X-rays and are, therefore, related to the density
inserted into the scanner. This is especially useful of the rock (Tolansky 1961). Higher luminance
in sections of the core which appear structureless values represent greater penetration of the rock
and homogenous. The steel mesh is positioned by the X-rays and, therefore, lower density.
to avoid obscuration of the core and its image Conversely, lower luminance represents areas of
may optionally remain on the composite image rock with higher density and therefore greater
for scaling and quality control purposes. X-ray 'stopping ability'. The luminance values
Where no rock is present the image appears to (which vary between approximately 60 and 200
be very bright (white). This is due to saturation for typical core material) are, therefore, inversely
or 'burn out' within the image intensifier, caused related to the rock density.
by the higher intensity of X-rays where there is Despite the high quality of the imaging system
little core material present to block them. This used in the capture of the information, each of
'burn-out' of the image artificially increases the the images contains an astigmatic error. This
intensifier output in closely neighbouring areas, means that there is an apparent density variation
resulting in the surrounding rock appearing between the centre of the image compared with
'bleached'. In order to avoid the possible the edges. While this does not significantly effect
misinterpretation of the X-ray intensity in these the visual interpretation of the images; it is
areas, disks and/or strips of lead shielding undesirable in the point luminance data. To
approximating the density of the resinated slab eliminate this error, therefore, the luminance
material are placed into plug holes, and other measurements are recorded from a fixed point
significant gaps. within the X-ray field. The luminance profile is
obtained by moving the rock (using the scanner
transport mechanism) and recording the values
Production of the quantitative X-ray density at the known fixed point within the field of view.
data Measurements may be recorded at any required
spacing; with 1 92 in spacing being used in the
The digital images which are obtained from the current project.
scanner are composed of pixels of varying grey The luminance measurements are made using
scale (0 to 255). The grey scale can be read, in 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

190 Ox x i x Luminance
x o ox• i

o Corrected
170 Luminance
--Luminance
~'~ 150
_~ . ~ o ~, j . . . . . . i R2=0.6445
o e j x ~ ~ x|
" ~ ' x x o x t~ x '- ..... Corrected
~ ~ o :! o i F~ o~ ~ ~ ~- ~$ ' ~ ' ~ ~x I ~, ; Luminance
t30

e=
C 110
E
,-1
,-I
9O

50
2.35 2.4 2.45 2.5 2.55 2.6 2.65 2.7 2.75 2.8
Bulk 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 Discussion


filter (i.e. each measurement is made on an
image comprising the average of 20 scans of the The luminance values indicate the penetrability
stationary rock). This is done automatically, and of the rock material to X-rays and are, therefore,
in 'pseudo real time', using the image processing related to the density of the rock. Higher
computer. luminance values represent greater penetration
Measurements of the thickness of the slab at of the rock by the X-rays and, therefore, lower
each luminance recording point are noted. In density. Conversely, lower luminance represents
addition, luminance values for an aluminium areas of rock with higher density and therefore
block 'standard' placed at the top and base of greater 'stopping ability' of the X-radiation. The
each 3ft section are measured. Where necessary luminance values are, therefore, inversely related
the scanner controls can be adjusted prior to to the rock density.
scanning to ensure that the observed luminance A comparison of this luminance data, repre-
from these calibration standards remains con- senting density, with traditional wireline log
sistent. These data, along with the luminance density measurements is presented in Figs 3(a
values are entered into a spreadsheet and stored and b). Figure 3(a) shows the correlation of the
on the PC for subsequent analysis luminance data with the FDL trace. The tie lines
indicate the core to log depth shifts appropriate
Description of X-ray images for this core material. Clearly excellent correla-
tion between the luminance profile and wireline
Figure 2 presents a single frame showing the X- log is observed, with Fig. 3(b) showing the
ray image at the point 'F' marked on Fig. 3. The luminance data depth shifted and superimposed
bright, irregular lines represent core breaks on the FDL trace. The luminance values are
which are likely to be coring induced. The core smoothed (using a simple 5 point moving
breaks are locally bedding-parallel. This frame average filter) but are otherwise unprocessed.
displays very clearly a partially cemented frac- A cross plot of luminance against bulk density
ture running subvertically through the core. The (from the wireline log) is shown in Fig. 4.
contrast produced by variations in the core Luminance values are shown both uncorrected
material density allows detailed examination of and corrected for slab thickness variations. The
these and other features. A conventional core correction is performed assuming a simple
analysis plug hole, with included masking, is reciprocal relationship between thickness and
shown, as is the '/2 in alignment grid. luminance value: this is considered to adequately
22 A. R. DUNCAN E T AL.

190 . . . . . . . o
io 6 . i

170 ........ o ...... ~


I
o ~
~1~01 . . . . . . . . . . . . . . . . . . ~ o. 0 ~ i o

~o o oo o~ ~ _~ oo ~o_
130 . . . . ~ ....... ~ . . . . . . . . . ,, ~ -. - : - ~ = - - L o ......
0 0 ~ O 0 '
I o o o ~ o
, o ^ o~..~f o o o ! '

"~
o
o
I..IU~
!~ / o
. o
- o 2v ~ !
~
. . . . .

~ _ o~ ~9 i , o Corrected
_1 9o i - ~ '~ ~. . . . . . . . . . i .... ~ Luminance
oel o o i '
oo .~o o ! i
70 ._0 0 O0 o; ~ -- : t ~C~
: , i Luminance
i 1 R2=0.6135
i
0 2 4 6 8 10 12 14 16 18

Porosity (plug) (%)

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 ! i D BD1 ,, B D 2
i
i i__ 1 o BD3 x BD4
170 ---4 .......
] ! x BD5 , BD6
'
:
ii - - A l l data R2=0.88
~.~ 150 ~ ~Nxx -
m

.Q
=" 130
i "-%
0 x ;
C x ' __
(~ 110
e"

E
--I 90

1.80 1.90 2.00 2.10 2.20 2.30 2.40 2.50 2,60 2.70 2.80

Bulk Density (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 Detailed conventional core analysis and sedi-
bulk material over the relatively small observed mentological analysis has been carried out on
variation in both luminance and thickness. This the core sections analysed for this project; this
thickness c o r r e c t i o n is seen to h a v e little data has been c o m p a r e d with the luminance data
significance in the final correlation of the logs. measured from the slabbed section close to the
QUANTITATIVE X-RAY DENSIMETRY 23

plug locations. Figure 5 shows, for example, a the loading of the core material and, in
cross plot of porosity of the CCA plug samples particular, in the automated capture of the
against the luminance values. Again the linear image and luminance data could allow higher
relationship between luminance and this key throughput; enabling greater data sampling
physical property is well defined. densities and potentially more detailed data
As further confirmation of these relationships, processing and analysis.
X-ray luminance values were measured for an Perhaps of greatest interest is the elimination
independent and varied collection of conven- of the optical distortion error in the individual
tional core analysis plug samples whose physical image frames. While the necessary geometry of
and geological properties are well established the scanner is a major contributory factor to this
(Duncan 1993). Figure 6 shows the bulk density error, it is believed that image processing may
for these samples plotted against luminance; the yield a significant and reliable reduction. By
strong linear relationship is again confirmed. viewing the image of a homogenous standard
These data are further differentiated by lithology (such as an aluminium block of similar size to
and while it is interesting to speculate on the the core section), it is possible to store the
relationship between lithology and luminance, variations in the image due to this error in digital
this dataset is considered too sparse to prompt form. By 'deconvolving' the standard image
any reliable conclusion. obtained in this way from the images of the
Interestingly, and as a positive demonstration scanned rock it may be possible to provide a
of the utility of these X-ray densimetry measure- much flatter response from the imaging system.
ments, the original core to wireline shifts for the This intermediate processing would allow an
slabbed core sections were taken to be: Wireline accurate digital representation of the density of
Log Depth equals Core Depth + six feet. Com- the core across the full image to be produced.
parison of the quantitative FDL (log) and Instead of collecting data at specific points, it
luminance measurements, however, indicates would then be possible to map the density
that a core to wireline correction of eight feet variation of the rock slab in two dimensions.
(downhole) for section A is more appropriate. A This would provide, not only a more accurate
shift of six feet for Sections B and C is confirmed log for comparison with downhole logs, but also
by comparison of the density and luminance allow the density variation to be plotted as a
traces. These revised depth shifts have been three-dimensional map, potentially highlighting
applied to the luminance data presented in Fig. more subtle variations in the core structure.
3(b). Initial work carried out is encouraging.
While the sections analysed for this project
were chosen in part for their known variation in
sedimentological structure, the success of this Conclusion
correlation technique in its most basic form
without any significant data processing clearly Linear X-ray scanning has an established role in
demonstrates the potential of these measure- non-destructive imaging of core, with the varia-
ments. It is believed that refinement of the tion in image reflecting the density variation in
processing could yield considerable additional the core section. The techniques described here
data which, coupled to the non-destructive allow not only the qualitative X-ray image to be
nature of the methods, the ability to analyse produced, but also quantitative luminance va-
material within opaque liners and the speed of lues to be extracted. These correlate very well
data capture, makes the technique of very with physical core properties, for example bulk
considerable importance. density and porosity, derived from wireline or
conventional core analysis techniques. These
luminance values thus provide a valuable core
Development to log correlation tool which may be of
particular value where traditional Gamma Ray
Technically, the performance of the scanner is or Core Analysis techniques are unavailable or
excellent. Quantitative investigations of the relatively unreliable due to poor response or
physical performance of the scanner, for exam- sparse data. The possibility of improving the
ple of the effects of variations in X-ray power operating procedures, in particular the sampling
output or the influence of 'burn-out' around interval and processing methods, as well as
unshielded plug holes etc., could potentially lead ultimately providing full density maps of the
to direct calibration of the luminance values in core section promise to yield even greater
terms of physical properties of the core material. benefits, and confirm the importance of X-ray
Improvements in the operating procedures, in imaging as a core analysis tool.
24 A . R . DUNCAN ET AL.

We gratefully acknowledge the permission to publish DUNCAN, A. R., 1993. A sedimentological, petrographic
this material granted by the Scott partner group: and reservoir geological study of the Devonian age,
Amerada Hess Limited, Amoco (UK) Exploration, old red sandstone of Gamrie Bay, Grampian
Deminex (UK) Oil and Gas, Enterprise Oil, Kerr Region. M.Sc. Thesis, University of Aberdeen.
McGee Oil (UK), Superior Oil (UK) and Premier Pict DUNCAN, A. R., MATHESON,F. E., & COLLIE, D. A. L.
Petroleum. Our thanks goes to F. Matheson at 1996. Quantitative X-ray density imaging of
Robertson Research Int. Ltd who diligently and selected cores. Robertson Research International
expeditiously undertook the preparation and measure- Ltd Project Report No D213 for Amerada Hess
ments of the core material analysed during this project. Ltd.
RIGSBY, C. A., ZIERENBERG,R. A. & BAKER, P. A.
1994. Sedimentary and diagenetic structures and
References textures in turbiditic and hemiturbiditic strata as
ALGEO, T. J., PHILLIPS,M., JAM1NSKI,J. & FENWlCK,M. revealed by whole-core X-radiography; Middle
1994. High resolution X-radiography of lami- Valley, northern Juan de Fuca Ridge. Proceed-
nated sediment cores. Journal of Sedimentary ings, Scientific Results, ODP leg 139, 105-111.
Research A: Sedimentary Petrology and Processes, TOLANSKY, S. 1961. Introduction to Atomic Physics.
A64, 665-668. 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 geochem- measurements were made on aliquots of the
ical logging tools is to transform the raw oxide same crushed and thoroughly homogenized rock
curves into mineralogy logs. In a recent exercise powder for each sample. There is, therefore,
aimed at calibrating geochemical logs in a UK essentially no scaling problem involved to
borehole a number of core samples (103) were explain variations in the modal estimates, and
taken and analysed extensively in the laboratory a minimal problem of sub-sampling from the
for both chemistry and mineralogy, to provide a rock powder.
database to support the log calibration. For all
103 core plugs quantitative mineralogy was Background
determined by X-ray diffraction at the British
Petroleum laboratories in Sunbury and by Through the use of pulsed neutron devices,
infrared spectroscopy (MINERALOG) at Core direct activtion of the formation by appropriate
Laboratories. In addition a petrographic exam- isotopes, and the natural gamma spectra it is
ination was carried out, and a minimal point possible to obtain an almost complete, and
count made (200 points per thin section) on continuous log of the major element chemistry
approximately half the samples to provide of a formation. These techniques were pioneered
approximate modal data. All core plugs were by Schlumberger (Hertzog & Plasek 1979;
also chemiclly analysed by X-Ray Assay La- Hertzog et al. 1987a, 1987b, 1989; Galford et
boratories (XRAL) in Ottawa for all major and al. 1988; Rupp et al. 1989) with their Geochem-
all potentially significant trace elements (a total ical Logging Tool (GLT) offering measurements
of 69 elements per sample). From the chemical of Si, A1, Ti, Fe, Ca, K, S, the minor elements
data, estimates of the modal mineralogy were Gd, Th, and U, together with H and C1. Other
calculated using a selection of different algo- tools are now available (Wyatt & Jacobson et al.
rithms. Together these analyses result in a range 1993; Odom et al. 1994; Jacobson & Wyatt 1996,
of modal estimates and the purpose of this Herron & Herron 1998). Transformation of the
contribution is to compare these estimates in an major elements into the more conventional oxide
attempt to evaluate the accuracy of the different form gives virtually complete major element
methods. Apart from the petrographic work, all oxide analysis at each measured depth interval,

HARVEY,P. K., BREWER,T. S., LOVELL,M. A. & KERR, S. A. 1998. The estimation of modal 25
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
26 P. K. HARVEY E T AL.

typically every 15cm, down the borehole. One on a routine basis (Herron 1986; Harvey et al.
approach to the interpretation of the resulting 1990; Harvey & Lovell 1992; Harvey et al. 1992;
geochemical logs is their conversion into com- Lofts et al. 1994, 1995b).
puted mineral assemblages, the so called 'che- For the case study described here a particu-
mical modes' of Wright and Doherty (1970). The larly complete dataset is available consisting of
resulting mineralogy logs are valuable in their conventional (physical) modal mineral measure-
own right but may be used in addition with a ments, together with comprehensive chemical
suitable rock classification filter to produce analyses from which calculated modes could be
lithological logs (Herron 1988), or estimation obtained. It is the particularly complete nature
of other formation properties such as matrix of these data which justifies a comparison
(grain) density, porosity, Cation Exchange between physical and calculated methods of
Capacity (Chapman et al. 1987; Herron 1987b; modal analysis, and the opportunity to make
Herron & Grau 1987), thermal conductivity some comment on the accuracy of modes. Using
(Dove & Williams 1988), heat flow (Anderson & natural samples (borehole core plugs) in this
Dove 1988), photoelectric factor, Pe (Kerr et al. case, however, precludes any definite means by
1992), magnetic susceptibility (Harvey et al. which one method can be chosen as 'more
1997), fluid saturation (Hastings 1988), neutron accurate' than another, unless a particular mode
capture cross-section (Herron 1987b), and, is 'obviously' wrong. As a first order assump-
indirectly and probably only in formation- tion, however, it is likely that if two or more
specific situations, permeability (Herron 1987a). unrelated methods of estimation give essentially
The transformation of a rock's elemental the same result then they are probably close to
composition to mineral assemblages has been the true value. This level of uncertainty arises
the subject of many contributions. Igneous because all methods of modal estimation can
petrologists have long employed the C.I.P.W. give seriously erroneous results sometimes, and
norm (Cross et al. 1903), and similar ideas have for some minerals; it is not simply a question of
been extended to metamorphic rocks with calibration and precision (repeat measurement
Niggli's normative procedures (Burri 1964). In error). With the spectral techniques particular
calculating norms there is no requirement that problems arise from spectral overlap and poor
the mineral assemblage used is that which resolution at the lower concentrations. With
actually occurs in a rock, and as such the computed modes it is the choice of the correct
C.I.P.W. norm, for example, was developed mineral assemblage, the correct 'composition'
originally for purposes of classification and and possible problems of compositional co-
employed a strict and standard list of 'possible'. linearity which are important.
minerals. However, for the purposes to which Modes are usually obtained by micrometric
geochemically derived mineralogy logs might be analysis (point counting of a thin section) and as
put (lithological analysis, basin modelling, pet- such are usually expressed in volume percent of
rophysical estimation) it is necessary to estimate the optically identified minerals. In contrast, a
the percentage of the minerals that are actually set of modal proportions may be calculated, to
present in the rock. The latter is the 'mode' of give 'norms', by assuming an ideal or theoretical
the rock which is conventionally determined suite of minerals, and the compositions for those
directly by micrometric analysis (point counting minerals. From these data some sort of fit may
of a thin section) or spectral methods such as X- be found that partitions the mineral composi-
ray diffraction or infrared spectroscopy (Harville tions within the initial rock analysis. Norms are
& Freeman 1988; Adam et al. 1989; Matteson & usually expressed in weight percentages, and
Herron 1993). have found very wide application, particularly in
The alternative approach is to compute the igneous petrology, for characterization and
mode from a complete chemical analysis. Nu- classification. In these applications comparison
merous authors have offered specific solutions to between rocks is made with a common set of
this inversion problem including modified nor- (chosen) minerals, unlike the mode, which
mative schemes (Imbrie & Poldervaart 1959; reflects the actual minerals present. Normative
Nicholls 1962; Pearson 1978), graphical models mineralogy logs may be useful in the early stages
(Miesch 1962; Fuh 1973) and a variety of of an investigation but are no substitute for the
numerical models including least squares mini- estimation of the actual mineral percentages
mization, linear programming and genetic algo- present if attempts are to be made later to
rithms (Wright & Doherty 1970; Albarede & generate, for instance, a grain density log. In this
Provost 1976; Fang et al. 1996), while others report the attempt is made to calculate chemical
have considered the strategy and associated modes, or the mineral proportions of the actual
practical problems of performing the inversion minerals present in the sample.
THE ESTIMATION OF MODAL MINERALOGY 27

Table 1. Mineralogy of the core samples.

Technique XRD MINERALOG Petrography

Silica Quartz * * *
Feldspars Albite * * *
K-feldspar * * *
Carbonates Calcite * * *
Ankerite *
Dolomite * *
Siderite * * *
Clays/Micas Muscovite *
Illite * * *
Smectite *
Kaolinite * * *
Chlorite *
Minor phases Zircon *
Barite * *
Pyrite * * *
Apatite *

* 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 dolomite and siderite are clearly distinguishable
petrographically in stained thin sections and are
The mineralogy of the core samples derived quantified as those species in the infrared
from a combination of XRD, infrared (MINER- ( M I N E R A L O G ) results. In the X R D analyses,
ALOG) and petrographic information is sum- ankerite (a Ca, Mg, Fe carbonate) is quantified
marized in Table 1, and also shown as a in place of dolomite, so that the two spectral
downhole mineralogy log in Fig. 1. The 100 techniques are not estimating the same carbo-
metre sequence consists of clastics covered by nate species. The carbonates occurring in Units
some 8 m of limestone. For purposes of descrip- 2 and 3 are dominantly dolomite or ankerite,
tion the section can be divided into five units: with minor amounts of siderite, while the
limestone unit at the top of the logged section
(Unit 1) (360-368 re)virtually pure calcite lime- (Unit 1) is a virtually pure calcite limestone.
stone; For the clay and other phyllosilicate minerals,
(Unit 2) (368-382m) quartz rich (60-65%) sec- kaolinite and chlorite are measured together in
tion with sub-equal quantities of the X R D analyses. Chlorite was only seen in
feldspar (both albite and K-feldspar) thin section in occasional grains and was not
and clays. Both kaolinite and illite (the detected in the infrared figures. For this study
latter generally in excess) are present; chlorite is considered to be absent. Muscovite
(Unit 3) (382-397m) quartz-carbonate domi- and illite are also determined together by XRD.
nant lithology with quartz in excess, Occasional distinct flakes of white mica are
and sub-equal proportions of kaolinite present in a number of sections but in no case
and illite; would these make up more than a fraction of
(Unit 4) (397-430 m) Very inhomogeneous sec- one percent of the rock. While white mica
tion with 40 to 70% quartz, no (assumed to be muscovite) is known to be
significant carbonate, and clay con- present in a very small amount in some samples
centrations up to c.30%; kaolinite it was not detected by infrared spectroscopy, and
generally in excess of illite. Locally is virtually impossible to calculate with any
high concentrations of pyrite (included reliability due to a strong compositional co-
with 'minor' minerals in Figure 1). linearity with K-feldspar, illite and kaolinite
( 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 which are present in significant proportions.
quartz-rich (70-80%) section with Included amongst the minor phases which
minor feldspar, about 15% clay, some occur in at least some of the samples are zircon,
two-thirds of which is kaolinite, with a barite, apatite and pyrite, all of which have been
few percent of minor minerals. identified petrographically. Of these, only pyrite
occurs locally in sufficient quantity to be
Of the possible carbonate phases calcite, 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 From the mineral data above, all the observed
detected by infrared, but both apatite and zircon mineral assemblages can be established, and
were too low for the spectral methods. these, minor phases excluded, are summarized in
Table 2; in all a total of thirteen different
Numerical modelling of the core sample assemblages.
From the chemical viewpoint the following
mineralogy components are available for modelling: SiO2,
The estimation of a modal mineral assemblage A1203, TiO2, Fe203, MgO, CaO, Na20, K20,
from the chemical analysis of a sample requires MnO, P205, S, CO 2 and H 2 0 + , expressed in
the minerals in the assemblage to be chosen, and weight percent, together with Ba and Zr which
the compositions of those minerals to be defined. were reported in parts per million. No other
Given this information there are a variety of 'minor' elements are in sufficient concentration
solution methods and strategies that can be to be expected to form discrete mineral phases,
employed to solve for the mode (Harvey et al. or significantly alter the modal estimates of
1990, 1992; Lofts et al. 1994; Fang et al. 1996). other mineral phases in which they might occur
For the modelling of the samples described as trace lattice components.
here, the main minerals are quartz, feldspars Zr and Ba are considered to occur only in
(albite, K-feldspar), carbonates (calcite, dolo- zircon and barite, respectively. In addition,
mite, siderite), clays (kaolinite, illite) and the amongst the oxide components P205 almost
minor phases (zircon, barite, apatite, pyrite). certainly occurs at significant levels only in
THE ESTIMATION OF MODAL MINERALOGY 29

Table 2. Observed mineral assemblages in the core samples (excluding minor phases).

Assemblage l 2 3 4 5 6 7 8 9 10 11 12 13

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

* 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 petrographi- BaO, ZrO2, S and P205 were removed from the
cally. TiO2 poses a problem, and in the first data matrix leaving SiO2, A1203, FeO, MgO,
instance is best calculated as rutile, though there CaO, Na20, K20, CO2 and H 2 0 + to be
is no evidence that this mineral actually occurs distributed, as appropriate, between the impor-
in any of the samples. TiO2 may also be present tant remaining mineral phases: quartz, albite, K-
in one of the clay phases, and this problem is feldspar, calcite, dolomite, siderite, kaolinite,
discussed later where there is good evidence that illite and possibly smectite. The simultaneous
it actually occurs in different minerals in estimation of all these minerals together would
different parts of the section. constitute a fully determined system for methods
Sulphur is assumed to be present only as a of inversion involving the solution of systems of
component of pyrite. Other minerals, such as equations.
gypsum or anhydrite are possibilities, though
there is no evidence for any sulphates being
present, and pyrite is the only identified sul- Strategies for extraction of the main mineral
phide.
phases: Models A, B, C
Manganese, which is only present at a very
low level (maximum 0.75% MnO, 90% of To remove complications related to methods of
measurements less than 0.18% MnO), was solution a simple unconstrained and unweighted
a d d e d to iron (as FeO) for purposes of least squares method has been used throughout
computation. Manganese often substitutes for (Harvey, et al. 1990) for inverting the different
iron, and the significant correlation (at a = 0.05; models. One consequence of the lack of con-
r = 0.58) between the two elements in these data straint is that mineral proportions may be
is consistent with this occurring here. Removing negative, implying an insufficiency of a combi-
MnO leaves a total of 14 possible mineral phases nation of elements with respect to a 'perfect'
and 15 chemical components to consider. solution. Such negative estimates, while impos-
Of the several strategies employed in the sible, offer a guide to the fact that either the
modelling of the mineral assemblages in this modelled assemblage is wrong, or one or more
case history three simple methods are presented of the mineral compositions are in error. Clearly,
here. In each case the data were pre-processed to such a solution is unacceptable. One approach,
remove the minor phases rutile, apatite, barite, then, is to model a given composition with all
zircon and pyrite which were calculated out of likely minerals, and to reject those minerals
each core analysis assuming ideal stoichiometric which turn out negative. This is the basis of
compositions. Provided the chemical analyses of Model A, described below. Another approach is
the core samples are accurate this procedure to model each given composition to all the
gives excellent estimates for these minerals which mineral assemblages which are known to occur
cannot be matched by any direct measurement. in the section (Table 2), and chose the best fit as
Although treated here as a minor phase, pyrite the appropriate solution. This is the basis of
does reach significant concentrations in a few Model B. For Model C the mineral assemblage
samples; the variation in pyrite downhole is obtained from the X R D analysis was taken as
shown in Fig. 2, and is discussed later. correct, and the given composition fitted to that
With extraction of these minor minerals TiO2, assemblage. This latter approach, in principle,
30 P. K. HARVEY ET AL.
360 ..... 9 9 t ' ' ' l ' ' ' ~ ' ' " l ' ' ' 360
9 ' i 9 " ' i ' 9 ' i 9 , , I ' ' '

Unit 1

370 370

380 Pyrite .~so

390 390

_: _~:~ ,.
4O0 Quartz ....... ].... :-- "'~-~ .... 4O0

~ ~i~--~_~.: .:_ ~_

410 ........ 41o


3
420 420
.........
430 ~z-~, , Py-/a) 430
9 Py-(b)

440 ' - - Py-(m) 440


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

450 450

9 ' ] ' " ' I ' * * Z t r ' * r ' ' 460


20 40 60 8O 100
20 40 60 80 1O0

360 . .. 360

370 370

380 380

390 390

400 400

410 410

420 1 Unit4 420

430 430

440 440

450 I Uni|5 450

460 460
0 20 40 60 80 100 0 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 however, cannot be justified because of the
mineral assemblage. complex interaction of phases in a least squares
model. With little or no formal justification one
M o d e l A. iterative removal o f n e g a t i v e procedure we have found very effective is to
remove the most negative phase and re-solve the
minerals
system. If negative phases still occur the
Solutions with negative compositions imply that procedure is repeated until all phases are
there is an inconsistency in the postulated positive. The procedure is illustrated in Table 3
mineral compositions (as stated above) and a for sample K78. The least squares fit using all
simple means of overcoming this is by removing nine minerals is good but dolomite is slightly
the mineral from the analysis. Wholesale re- negative (a) at --0.21. Removing dolomite as
moval of all negative minerals in one pass, one of the phases sends calcite slightly negative
THE ESTIMATION OF MODAL MINERALOGY 31

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

a b c XRD MINERALOG

Quartz 79.81 79.97 80.05 88 80


Albite 0.21 0.12 0.00 0 0
K-feldspar 5.42 5.47 5.30 3 7
Calcite 0.01 -0.12 0 0
Dolomite -0.21 0 0
Siderite 0.22 0.21 0.13 0 0
Kaolinite 12.13 12.35 12.44 7 5
Illite 0.00 1.14 0.81 0 7
Smectite 1.29 1.22 0.96 0 0
Std. Err. 0.011 0.057 0.067

(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 Model C: fit to the individual mineral
the final estimate. The mineralogical pseudo-log assemblage for each core sample
shown in Fig. 1 was produced using this model.
For completion in Fig. 1, the trace minerals were For each core sample a list of the minerals
added and the Model A derived mineralogy present had already been identified by X R D ,
normalized to make the assemblages sum to petrography or infrared analysis. For this model
100%. The standard error given in Table 3 is a the X R D mineral assemblage was chosen for
measure of the fit of the core and mineral each sample and then fitted accordingly. Poor
chemistry and is computed between the original fits, often with negative mineral estimates,
(input) chemistry, and the composition back- identify a real incompatibility between the rock
calculated from the derived (output) mineralogy and mineral chemistry assuming that the phases
(Harvey et al. 1990). are correctly identified by the XRD.

Model B. choice o f known mineral Comparison of measured and calculated


assemblages estimates of mineralogy
Thirteen possible parageneses for the core Figs 2 through 4 summarize the variation in
samples have been identified, within limits of measured (XRD & M I N E R A L O G ) and com-
detection, from the X R D , petrographic and puted (Models A & B) values for quartz,
infrared data. These are summarized in Table feldspars and pyrite (Fig. 2), carbonates (Fig.
2. Most assemblages contain five or six phases; 3), and the clays (Fig. 4). A more detailed
only one (assemblage 3) contains over six (7), so comparison may be made by examination of
that the number of minerals is generally at least Table 4 which shows the correlations between all
two less than the number of chemical compo- models and the physically derived measure-
nents. The procedure was to fit each sample to ments. In Table 4, based on 103 core samples,
each of these possible assemblages. The opti- correlations > 0.19 are different from zero at a
mum assemblage was then chosen using the significance level of 0.05, and > 0.25, at a level of
following criteria: 0.01.
For quartz, the best correlation is between
(a) in virtually all cases this procedure yields a Model A and the infrared ( M I N E R A L O G )
selection of potentially acceptable (non- data; the relationship is linear (Fig. 5) and close
negative) assemblages; the one with the to the 1:1 line. The agreement between the two
smallest standard error is then chosen; is particularly good in the lower half of the
(b) but, if all possibilities gave at least one section (Units 4 & 5), though in the upper part,
negative mineral proportion the assem- below the limestone cap, the computed quartz
blage with the smallest absolute negative estimates are almost consistently lower. Regres-
sum was chosen. sion analysis of this relationship gives a slope of
32 P. K. HARVEY ET AL.
, 9 , ~ . 9 . ~ , 9 , 360

370

Unit2

380 380

390
Umt 3 390

400 400

4.10
Siderite Ankerite / Dolomite 41o
420 42O

430 .............. 430


i 9 Sid-(a) 71 9 Dol-(a)

I 9 Sid- ( b) i 9 Dol-(b)
44O 440
iI Sid-(m) [ - - Dol-(m) :

[ ..... Sid-(x) j - - - Ank-(x)


450
f! ............. 450

4(30 . . . i . , , I 9 - . i . , . i . . . _ . . I , . , i , , , i , , . t ' ' 460


0 20 40 60 80 0 20 40 60 80 1(30

360 36O

370 2%.~_ 370

380
~ "
39O

4OO

411)
Calcite Total carbonates 4~o
420 420

;.m

43O ................. 430


9 Cal-(a) 9 Carb-(a)

9 Cal.(b) 9 Carb-(b)
440
- - Cal.(m) - - Carb-(m)

..... Cal-(x)
450 i ..... Carb-~x) 450

9 i , , , i , , , i , . . i . , . 460

20 40 60 80 0 20 40 60 80 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 estimates averaging between 7% and 9% for
removal of four outlying points. In view of this Models A and B, respectively. The infrared
excellent relationship is of note that the correla- figures agree at 7.4%; the X R D average is lower
tion for quartz between the X R D and the at 2.6%. The highest linear correlation is again
infrared is slightly lower with the X R D mea- shown between Model A and the MINERA-
surements being generally higher than the LOG, and it is likely that these methods are
infrared, often by several percent. Overall for giving close to the true result. For the calculated
quartz, all models perform reasonably well and modes the albite concentration is constrained by
show similar patterns of variation. It is not at all the sodium concentration, and in the absence of
clear which set of data is correct! any other sodium bearing mineral, an accurate
Of the feldspars, both albite and potash- estimate should be expected. With albite occur-
feldspar are present in small quantities. Albite ring in concentrations below 10% there are
is virtually restricted to Units 2 and 3, with problems of sensitivity and detection limit with
THE ESTIMATION OF MODAL MINERALOGY 33

360 .,.,..,,...,..., . . .
On, I
9 . L .
370 :~':.; Kaolinite :y~.j Total clays
UnitZ
~80 :,(,...~:
u.*,3 !;-'<; " Umt3

4304204,040039~
i'~~;
_
umt4 thai,4

Unit $ Ill-(b) u,~l 5


- - Ill-(m)
450 ~ , ~ ) ..... Kaoi-(x) - ~
Y~. :":. -. .....
. . . Ill-(x) -

o 20 40 6o 8o Ioo o 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 B/Mlg C/MIg Mlg/XRD A/XRD B/XRD C/XRD

Quartz 0.987 0.980 0.954 0.962 0.966 0.953 0.948


Albite 0.865 0.837 0.712 0.847 0.862
K-feldspar 0.193 0.431 0.515 0.146 0.433
Kaolinite 0.928 0.916 0.662 0.785 0.809 0.784 0.586
Illite 0.025 0.512 0.192 0.587 -0.071 0.227 0.274
Calcite 0.976 0.975 0.973 0.994 0.983 0.979 0.978
Siderite 0.869 0.852 0.848 0.836 0.802
Dolomite 0.901 0.587 0.760 0.953 0.904 0.529 0.771
E carbonates 0.998 0.998 0.994 0.993 0.994
E clays 0.911 0.888 0.856 0.875 0.821
Pyrite 0.975 0.975 0.975 0.968 0.972 0.972 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 computed models (Harvey et al. 1992).
the agreement between all methods is remark- For the carbonates there is overall excellent
able. agreement between all the methods (Figs 3 & 6)
Potash feldspar occurs throughout the section in that the total amount of carbonate (calcite + -
below the limestone cap at concentration levels dolomite/ankerite+ siderite) determined by the
similar to albite (Fig. 2). There is poor agree- different methods is essentially the same. In
ment in detail between the models; Model B detail, however, there are some distinct differ-
shows the highest correlation of the computed ences between the measured and calculated
models, but is little lower than the weak mineral percentages. These effects are seen
correlation of 0.515 between the X R D and clearly in Fig. 3 where calcite is correctly and
infrared figures (Table 4). The estimation of K- accurately estimated by all methods in the
feldspar suffers both from problems of low virtually pure calcite limestone cap, but in Units
concentration in the physically derived estimates 2 and 3 below, where more than one carbonate
and potential compositional co-linearity in the mineral is present, dolomite is severely under-
34 P.K. HARVEY E T AL.

Fig. 5. Crossplot of measured and estimated quartz Fig. 7. Regression expressing kaolinite calculated using
contents for the three computed models and XRD (y- modal Model A as a function of the infrared
axis), shown relative to the infrared (MINERALOG) (MINERALOG) measurements. The relationship is
(x-axis) measurements. Qtz-(a): quartz computed from seen to be linear with a slope close to unity, but an
Model A; Qtz-(b): ditto, for Model B; Qtz-(c): ditto, intercept of some 5%.
for Model C; Qtz-(x): XRD measurement.

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 compro-
mised 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 deter-
mined 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
Fig. 6. As Figure 5 for total carbonate (calcite + relationship is shown in Fig. 7. Apart from
dolomite/ankerite + siderite)percentage. Carb-(a): three obvious outlying points the latter is
total carbonate computed from Model A; Carb-(b): essentially linear, with a slope close to 1.0, but
ditto, for Model B; Carb-(c): ditto, for Model C; Carb- an intercept which over-estimates the Model A
(x): XRD measurements. 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
estimated by the calculated models, and siderite section which the core photographs indicate to
is slightly over-estimated. Calcite is included in be very inhomogeneous. Sample preparation
results of both Models A and B but was not should have removed this problem for labora-
detected by either X R D or infrared methods. tory work; their gross deviation remains a
These differences are due essentially to the use of problem.
ideal carbonate compositions (Table 5) in the For illite there is no real agreement between
mineral calculations and the compositions of the any of the models. The closest relationship is
actual carbonates present in these samples not shown between the X R D and infrared figures,
being available. The use of an ankeritic compo- 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.

Qtz Ab K-f Cal Dol Sid Kaol Ill Smec

SiO2 100.00 68.74 64.76 0.00 0.00 0.00 45.48 57.28 51.14
A1203 0.00 19.44 18.32 0.00 0.00 0.00 39.29 18.55 19.76
FeO 0.00 0.00 0.00 0.00 0.00 62.02 0.65 5.11 0.83
MgO 0.00 0.00 0.00 0.00 30.41 0.00 0.14 2.07 3.22
CaO 0.00 0.00 0.00 56.03 21.86 0.00 0.41 1.59 1.62
Na20 0.00 11.82 0.00 0.00 0.00 0.00 0.00 0.43 0.11
K20 0.00 0.00 16.92 0.00 0.00 0.00 0.00 5.11 0.04
H20 0.00 0.00 0.00 0.00 0.00 0.00 14.04 8.86 22.80
CO2 0.00 0.00 0.00 43.97 47.73 37.98 0.00 0.00 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 comparison between a range of methods for the
the highest correlation of the computed models estimation of the proportions of different
(r=0.512), with the infrared estimates, though minerals in a set of samples.
its correlation with the X R D data is non- The question is, which is the 'best' (most
significant (at a level of 0.05). Depending upon accurate?) method for calibrating mineralogy
which set of data is believed illite is present logs, or whether the calculations on their own
throughout most of the section (Fig. 4) with the are actually superior. All methods obviously
highest values (10-20%) in Units 2 and 3. It is have their strengths and weaknesses. The major
difficult quantify using the physical methods problems with the spectral methods concern
because of poor sensitivity and spectral varia- sample preparation (and presentation), sensitiv-
tion, and to compute because of uncertainty ity and spectral resolution at low levels (lowest
about the mineral chemistry and chemical few percent) and spectral interferences. The
variation in these rocks. Despite these problems main problems with calculating the mode con-
the total clay curve shown in Fig. 4 shows cern the accuracy of the sample's chemical
surprisingly good agreement over the range of analysis, the need to solve for the correct phase
methods. assemblage and to have a good estimate of the
phase compositions, and compositional co-
Discussion linearity. Modern methods of quantitative geo-
chemical analysis are much more precise than
The results described here came from a study of the direct methods of modal analysis compared
mineral inversion methods to determine the way here, and with appropriate calibration may be
in which the most accurate mineralogy log could expected to produce estimates of the chemical
be obtained from a suite of geochemical logging components which are close to the actual values.
data. The chosen method would have to involve Some effects of concentration errors on element-
calculation from the geochemistry, and valida- to-mineral inversion are discussed by Herron &
tion of these results would need to be by Chiaramonte (1993).
comparison with some independent method. We have shown elsewhere that if the chemical
For the samples used in this validation a analysis is correct, the phase assemblage known,
particularly comprehensive dataset was available and the minerals in that assemblage analysed,
from X R D and infrared ( M I N E R A L O G ) ana- then the calculated mode is in excellent agree-
lyses made on the same samples. The latter, and ment with the 'true' mineral proportions (Lofts
the geochemical analyses upon which the modal et al. 1995b). While this may seem self evident
calculations were based, were all measured on the 'quality' of a calculated solution very rapidly
aliquots of the same homogenized powder for deteriorates as analytical errors increase. Where
each sample. All measurements were, hence, an element is virtually restricted to one mineral
made on essentially the same material, and on phase, and particularly if that mineral is only at
very similar volumes of the same material in levels of a few percent or less, such as sulphur in
each case. In the comparisons made here there pyrite, sodium in albite or barium in barite, then
is, therefore, no real problem of heterogeneity or the calculated mode for these minerals is likely
scaling (up or down) involved; just a simple to give the best estimate. In the more general
36 P.K. HARVEY E T AL.

case where elements are partitioned between For minor minerals, such as zircon, apatite
different minerals, a good fit between the input and barite, which are definitely known to be
sample chemistry and the chemistry recon- present (confirmation generally by petrographic
structed from the mineral proportions (Harvey examination) the only reliable method of esti-
et al. 1990) may provide very good evidence that mation is calculation. In most samples these
the solution is close to the 'true' mineral minerals were not detected by the spectral
proportions. methods, but as they contain elements which
Similar arguments may be produced for the do not normally occur at significant levels in the
quality of XRD and M I N E R A L O G solutions other minerals present (Zr in zircon, P in apatite,
under specified conditions. For several of the and Ba in barite), and the chemical analyses can
minerals examined here, close agreement can be regarded as both more accurate and precise
occur between different methods and it may be than the spectral mineral techniques, then
postulated that if a number of unrelated accurate estimates may be expected. The same
methods give essentially the same result for a is true for pyrite (constrained by the S content)
sample it is likely that that result is close to the in the absence of other sulphur bearing minerals
'true' value, though this cannot be proven. such as gypsum or anhydrite.
This would suggest that for the results If the chemical compositions of all the mineral
described here, quartz, pyrite and (total) carbo- phases were known, then their use, together with
nate are reasonably accurate, despite a minor 'good' chemical analyses should always produce
bias in the quartz crossplot, and variations the most accurate modal estimates by calcula-
between the individual carbonates could be tion using the actual mineral assemblage for a
easily reconciled by the use of the correct given sample. The mineral compositions, how-
carbonate mineral compositions. It is significant ever, are rarely known in sufficient detail and
that these minerals which do give reasonable quite erroneous estimates can be made if the
agreement between widely differing methods are wrong compositions are used.
characterized by a small number of cations and In the case history described here, Model A
restricted compositions which result in well was finally chosen for modelling the geochemical
defined XRD and infrared spectra. log data following further experiments using a
The variation between the different estimates reduced number of components to correspond
for the feldspars (K-feldspar and albite) and the to those measured by the geochemical logging
clays (illite and kaolinite) is more complex. Both tool.
spectral methods (XRD and infrared) suffer
from problems of overlapping lines for these Conclusions
minerals and poorer sensitivities (i.e. detection
limits) at low concentrations. Likewise, calcula- (1) There was generally good agreement be-
tions of the modes from the chemical analysis tween methods in the estimation of quartz,
suffer from two other problems. The first is the total carbonate and pyrite. It is reasonable
use of the 'true' chemical composition of each to assume, but cannot be proven, that these
mineral phase used in a model. It has been estimates are close to the 'true' values. It is
demonstrated previously (Lofts et al. 1995b) significant that those minerals which do
how sensitive a solution can be to changes in show good agreement between widely differ-
compositions of the minerals used in calcula- ing methods have fairly simple and limited
tions; ideal compositions are rarely appropriate. compositions (i.e. are stoichiometric).
The actual compositions of the kaolinites, illites (2) Agreement for the clay minerals and the
and the different carbonate minerals were not feldspars is much more variable due to
known in this study. The second problem is that problems of sensitivity and spectral inter-
of compositional co-linearity; that is, where four ference for the two physical methods of
or more of the minerals to be modelled lie on (or analysis, and problems of uncertain mineral
very close to) the same compositional plane composition and compositional co-linearity
(Harvey & Lovell 1992). This leads to essentially for the computed models. Good agreement
an infinity of solutions, or, if forced, a very is seen between methods for albite, even at a
unstable solution. The variable range of esti- low level, and kaolinite. The results for illite
mates for K-feldspar amongst the computed and K-feldspar were comparatively poor
values almost certainly results from this pro- and it is considerably more difficult here to
blem, given that the actual compositions of the judge which figures, if any, are close to the
clays (which have a number of chemical correct values.
components in common with the feldspars) are (3) Despite the problems with illite, excellent
not known. agreement is seen between the methods for
THE ESTIMATION OF MODAL MINERALOGY 37

total clay content, which is particularly rithm for least-square fitting and general error
useful for the calulation of shale components analysis. IPGP NS 252, 309-326.
in lithofacies modelling from the geochem- ANDERSON,R. & DOVE, R. 1988. The determination of
ical logs. heat flow in a wellbore in the South Eugene Island
area of offshore Louisiana: implications for fluid
(4) Poor agreement between methods results
migration and hydrocarbon location in the sub-
from low sensitivity (especially at low con- surface. Transactions Spectroscopy and Geo-
centrations) and spectral interferences in the chemistry Symposium, Schlumberger-Doll
X-ray and infrared techniques, problems of Research, Ridgefield, CT. Paper K.
compositional co-linearity and uncertainty BURRI, C. 1964. Petrochemical calculations based on
about the actual compositions of some of the equivalents (Methods of Paul Niggli). Israel
minerals in the calculated modes. Program for Scientific Translations, Sivan Press,
(5) If the chemical compositions of all the Jerusalem.
mineral phases were known, then their use, CHAPMAN, S., COLSON,J. L., FLAUM,C., HERTZOG, R.
C., PIRIE, G., SCOTT,H., EVERETT,B., HERRON,M.
together with 'good' chemical analyses,
M., SCHWE1TZER,J. S., LA VIGNE,J., QUIREIN,J. &
should always produce the most accurate WENDLANDT, R. 1987. The emergence of Geo-
modal estimates by calculation using the chemical Well Logging. The Technical Review, 35,
actual mineral assemblage for a given 27-35.
sample. The mineral compositions, however, CROSS, W., IDDINGS,J. P., PIRSSON, L. V. & WASHING-
are rarely known in sufficient detail and TON, H. S. 1903. Quantitative classification of
quite erroneous estimates can be made if the igneous rocks. University of Chicago Press,
wrong compositions are used. Chicago.
(6) Whether 'absolute' accuracy is actually DovE, R. E. & WILLIAMS, C. F. 1988. Thermal
conductivity estimated from elemental concentra-
important obviously depends on why he
tion logs. Transactions Spectroscopy and Geo-
mineralogy log is required; probably not if chemistry Symposium, Schlumberger-Doll
simply to illustrate diagrammatically the Research, Ridgefield, CT. Paper J.
relative variation in a sequence but impor- FANG, J. H., KARR C. L. & STANLEY, D. A. 1966.
tant if the data are to be used, for instance, Transformation of geochemical log data into
for quantitative basin modelling or physical Mineralogy using genetic algorithms. The Log
property log estimation. Analyst, 37, 26-31.
(7) Underlying these comparisons is the fact FUH, T. M. 1973. The principal of constituent analysis,
that the accuracy of any measured para- with special reference to the calculation of weight
percentages of minerals in metamorphic rocks.
meter is virtually impossible to specify,
Canadian Journal of Earth Science, 10, 657-669.
except in (usually quite unrealistic) limiting GALFORD, J. E., HERTZOG, R. C., FLAUM, C. &
cases. Even with this particularly compre- GALINDO. G. 1988. Improving pulsed neutron
hensive dataset no definitive conclusions can gamma ray spectroscopy elemental weight percent
be demonstrated concerning the accuracy of estimates through automatic dimensioning of the
different estimates; the analysis of these data spectral fitting process. Society of Petroleum
does, however, demonstrate the difficulty of Engineers, SPE 18151, 423-430.
obtaining accurate modal estimates. And in HARVEY, P. K. & LOVELL, M. A. 1992 Downhole
this case history there are no real problems mineralogy logs: mineral inversion methods and
the problem of compositional colinearity. In:
of homogeneity or scaling in the estimation
Hurst, A., Griffiths, C. M. & Worthington, P. F.
process! (eds) Geological Applications of Wireline Logs IL
Geological Society, Special Publications No. 65,
The authors would like to thank BP Exploration 361-368.
Operating Company for agreeing to the publication of --, BRISTOW,J. F. & LOVELL,M. A. 1990. Mineral
this work. We would also like to thank Core transforms and downhole geochemical measure-
Laboratories for the provision of the MINERALOG ments. Scientific Drilling, 1, 163-176.
measurements which were made in December 1988. --, LOFTS,J. C. & LOVELL,M. A. 1992 Mineralogy
logs: element to mineral transforms and composi-
References tional colinearity in sediments. 33rd Annual
Symposium of the Society of Professional Well
ADAM,H. G., HARVILLE,D., MEER, D. & FREEMAN,D. Log Analysts, Oklahoma city, Oklahoma.
1989. Rapid mineral analysis by Fourier trans- --, Lovell, M. A., Lofts, J. C., Pezard, P. A. &
form infrared spectroscopy, Society of Core Bristow, J. F. 1997 Petrophysical estimation from
Analysts, Annual Technical Conference Reprints, downhole mineralogy logs, In." LOVELL,M. A. &
v. 1 (198%1989), part II, Society of Professional HARVEY,P. K. (eds) Developments in Petrophysies,
Well Log Analysts, paper SCA-8809, I. Geological Society, Special Publications No. 122,
ALBAREDE, F. & PROVOST,A. 1976. Petrological and 141-157.
geochemical mass-balance equations: an algo- HARVILLE,D. G. & FREEMAN,D. L. 1988. The benefits
38 P. K. HARVEY ET AL.

and application of rapid mineral analysis pro- JACOBSON, L. A. & WYATT, D. F. 1996. Elemental
vided by Fourier transform infrared spectroscopy. yields and complex lithology analysis from the
Society of Petroleum Engineers, SPE 18120, 141- pulsed spectral gamma log: The Log Analyst, 37,
150. 50-64.
HASTINGS, A. F. 1988. Using the derived elemental KERR, S. A., GRAU, J. A. & SCHWEITZER,J. S. 1992. A
concentrations to improve the accuracy of fluid comparison between elemental logs and core data.
saturations determined from well logs. Transac- Nuclear Geophysics, 6, 303-323.
tions Spectroscopy and Geochemistry Sympo- LOFTS, J. C., HARVEY, P. K. & LOVELL,M. A. 1994. A
sium, Schlumberger-Doll Research, Ridgefield, stochastic approach to mineral modelling of log
CT. Paper T. derived elemental data. Nuclear Geophysics, 8,
HERRON, M. M. 1986. Mineralogy from geochemical 135 148.
well logging. Clays and Clay Minerals, 34, 204- & - 1995a. Reservoir character-
213. ization from downhole mineralogy. Marine and
- - 1987a Estimating the intrinsic permeability of Petroleum Geology, 12, 233-246.
clastic sediments from geochemical data. SPWLA &- - 1995b. The characterisation
28th Annual Logging Symposium. paper HH. of' reservoir rocks using nuclear logging tools:
- - 1987b. Future applications of elemental con- evaluation of mineral transform techniques in the
centrations from geophysical logging. Nuclear laboratory and log environments. The Log Ana-
Geophysics 1, 197-211. lyst, 36, 16-28.
- - 1988. Geochemical classification of terrigenous MATTESON, A. & HERRON, M. M. 1993. Quantitative
sands and shales from core or log data. Journal of mineral analysis by Fourier transform infrared
Sedimentary Petrology, 58, 820-829. spectroscopy, Society of Core Analysts, Annual
& GRAU, J. A. 1987. Clay and framework Technical Conference Proceedings, Society of
mineralogy, cation exchange capacity, matrix Professional Well Log Analysts, Paper SCA-9308.
density and porosity from well logging in Kern MIESCH, A. T. 1962. Computing mineral compositions
County, California. American Association of of sedimentary rocks from chemical analyses.
Petroleum Geologists, 71, 567 575. Journal of Sedimentary Petrology, 32, 217-225.
HERFtON, M. M. & HERRON, S. L. 1998. Quantitative NICHOLS, G. D. 1962. A scheme for recalculating the
lithology: open and cased hole application derived chemical analyses of argillaceous rocks for com-
from integrated core chemistry and mineralogy parative purposes. American Mineralogist, 47, 34-
data base. In." HARVEY, P. K. & LOVELL, M. A. 46.
(eds) Core-Log Integration. Geological Society, ODOM, R. C., STREETER, R. W., HOGAN, G. P. &
Special Publication 136 (this volume). T~TTLE, C. W. 1994. A new 1.625 inch diameter
HERRON, S. L. & CHIARAMONTE,J. M. 1993. Impact of pulsed neutron capture and inelastic/capture
element-to-mineral matrix concentration errors spectral combination system provides answers in
on geochemical log interpretation. Nuclear Geo- complex reservoirs. 35th Annual Logging Sympo-
physics, 7, 375-381. sium Transactions, Society of Professional Well
HERTZOC, R. C. & PLASEK,R. E. 1979, Neutron excited Log Analysts. Paper O.
gamma-ray spectrometry for well logging. IEEE PEARSON, M. J. 1978. Quantitative clay mineralogical
Transactions on Nuclear Science, NS-26, 1158- analyses from the bulk chemistry of sedimentary
1563. rocks. Clays and Clay Minerals, 26, 423-433.
, COLSON,L. SEEMAN,B. O'BRIEN, M. SCOTT,H. RuPP, J., HERTZOG, R. & SCHWEITZER,J. 1989. Using
MCKEON, D. WRAIGHT, P. GRAU, J. A. ELLIS, D. log-derived elemental concentrations in strati-
SCHWEITZER, J. ~ HERRON, M. 1987a, Geochem- graphic correlation of argillaceous sediments:
ical logging with spectrometry tools. SocieO'of Third International Symposium on Borehole
Petroleum Engineers, SPE 16792, 447-460. Geophysics for Minerals, Geochemical and
, SORAN, P. D. ~ SCHWEITZER, J. S. 1987b. Groundwater Applications.
Detection of Na, Mg, A1 and Si in wells with WRIGHT, T. L. & DOHERTY, P. C. 1970. A linear
reactions generated by 14 MeV neutrons. Nuclear programming and least-squares computer method
Geophysics, 1,243-248. for solving petrologic mixing problems. Geologi-
- - , COLSON,L. SEEMAN, B. O'BRIEN, M. SCOTT,H. cal Socieo' of America Bulletin, 81, 1995-2008.
McKEON, D. WRAmHT, P. GRAU, J. A. ELLIS, D. WYATT, D. F., JACOBSON, L. A. & HASHMY, K. 1993.
SCHWEWZER,J. & HERRON, M. 1989. Geochemical Elemental yields and complex lithology analysis
logging with spectrometry tools: SPE Formation from the Pulsed Spectral Gamma log. 34th
Evaluation, 4, 153-162. Annual Logging Symposium Transactions, So-
IMBRIE, J. & POLDERVAART,A. 1959. Mineral composi- ciety of Professional Well Log Analysts. Paper
tions calculated from chemical analyses of sedi- UU.
mentary rocks. Journal of Sedimentary Petrology,
588-595.
29,
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 , P. D. J A C K S O N , 2 T. S. B R E W E R , 1 G.
WILLIAMSON 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 tion of two datasets, which may comprise
one of the many attempts to utilize geological observations ranging from qualitative through
data obtained by measurements at different to quantitative, with the aim of providing the
scales. This use of data from different sources best data on which to base our interpretation
involves the reconciliation of different observa- and hence 'explain the meaning of' our observa-
tions which may be inter-related through their tions. Integration in turn may be defined as 'to
inherent property or physical basis (e.g. labora- find the total value of', and without necessarily
tory and in situ velocities or porosities), or implying total amalgamation of all available
through their similar volumes of interrogation data in a non-selective manner. Yet there
(e.g. porosity and permeability measurements on appears to be much concern and considerable
core plugs). Alternatively the data to be inte- effort directed towards deciding which of the
grated may not be related in either of these ways two different datasets represents the truth--log
(e.g. core descriptions and FMS images). or core?
Integration involves the reconciliation of such In this paper we review the basic principles
data in a way which is defined by the overall involved in data acquisition and interpretation
aims and objectives of the study. It may involve through consideration of the measurement
the calibration of one dataset through some process, measurement calibration, and measure-
equalization procedure, whereby one dataset is ment integration. Figure 1 summarizes the
assumed to be correct. Another scenario is problems involved. The measurement itself
where the two or more datasets are integrated may be characterized in terms of its scale,
through the selective addition of components to resolution and quality. These are functions of
enhance the overall picture of the formation the parameter being measured and the measure-
represented both downhole and in the recovered ment environment as well as the selected target.
core. These datasets may be multiple measure- These measurements can then be calibrated,
ments of the same physical parameter by either relatively (locally) or absolutely (globally);
different techniques or measurements of com- if the calibration is simply between the indivi-
pletely different parameters, In this latter ap- dual core and log measurements then the result
proach each dataset is respected for both its may be equalization of values with correspond-
inherent fundamental nature and scale, both ing loss of total information. I n t e g r a t i o n
datasets are assumed to be correct, neither through reconciliation of the different measure-
dataset is defined as superior in preference to ments towards an optimized solution should
the other, and the interpreter attempts to extract yield the best interpretation; but this depends
the maximum information from the total data strongly on the assumptions involved and these
available. should be directly related to the overall aim of
Today we are faced with core and log data in the study. Indeed the best approach to the
increasing quantity and sophistication. Integra- problem of core-log integration is through
tion of core and log data concerns the combina- judicious choice of interpretation target, careful

LOVELL,M. A., HARVEY,P. K., JACKSON,P. D. BREWER,T. S., WILLIAMSON,G. & WILLIAMS,C. G. 39
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
40 M.A. LOVELL E T AL.
' ' ' i , , , i , , , i ' ' ' i , , , i , , ,

~ AIMS/OBJECTIVES ~ ~- Data collected over the years


between 1965 and 1980.

If MEASUREMENT 1
scale - resolution ~ size / shape / orientation
quality ~ precision / a c c u r a c y / bias

800 I000 1200 1400 16IX) 18(l(I 20(~)

Pairs of breeding storks

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


relating measurements: a good correlation does not
(CALIBRATION1 "(INTEGRATION~/ imply a causal relationship (after Sies, 1988).
absolute - relative -"--~ 1 selective addition
equalisation k reconciliatiotl

gamma ray attenuation measurements) or sim-


ply an empirical relationship (for example
Fig. 1. Measurement, calibration, and integration. The porosity and saturation from electrical resistivity
choice and specification of the measurement should be measurements). The model used in relating the
dictated by the aims and objectives of the study, and measurement to the derived parameter may fall
should dictate both whether there is a need for within a wide range of variable sophistication.
calibration and the route to integration. This may impact on both the accuracy and
precision of the derived parameter since even
given accurate and precise measurements a
and appropriate selection of measurement tech- model that is too simple may not result in a
niques including necessary calibration, and true representation of the actual derived physical
optimization using well-defined integration. properties.
The aim of this paper is to document funda- We do not propose to consider the role of
mental concepts in the context of core-log mathematical or statistical procedures here (see
integration in order to form the foundations Moss 1997 for a review of the partitioning of
for data integration and interpretation. In petrophysical data) but do sound a warning note
examining these aspects of integration further on attempts to define the relationship between
we need first to consider the nature of measure- different parameters or measurements. Correla-
ment itself. tion coefficients are often used to support an
interpretation of a causal relationship between
two parameters, but a high coefficient does not
Measurement necessarily imply such a causal relationship.
Indeed, all too frequently, unrelated parameters
Measurements are made to determine the value are correlated in an attempt to derive some
of some parameter. Unfortunately the para- solution from our data. Figure 2 demonstrates
meter we are often interested in (for example such a classic mythical relationship between two
porosity) cannot be measured directly but is unrelated parameters. Even in this example the
obtained indirectly through the subsequent high correlation is provided through only two of
processing of raw measurements of a related the points and any rigorous testing of the
parameter(s). The relationship between the relationship is likely to be invalid.
measured parameter(s) and the derived para- Returning to the subject of measurement we
meter may have a well-constrained physical examine three primary aspects below: resolution,
basis (for example density estimates from 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 is the minimum bed thickness for


which the sensor measures, possibly on a limited
Resolution concerns the minimum separation portion of the bed, a parameter related to the
between two features such that the two features real value of the formation.
can be identified individually rather than as one This latter definition still falls short of ideal
combined feature (see for example Sheriff & since, as Theys (1991) points out, it does not
Geldart 1982). In terms of log measurements this necessarily measure the true value of the
relates to the physical separation of two features parameter concerned for the thin bed. If a
along the length of the well (usually in a vertical logging tool is to measure a parameter and yield
sense assuming a vertical drillhole). With respect a true value for even a limited portion of the
to core measurements this definition equally bed, then the bed thickness must be at least as
applies, although it may be complicated by the large as the vertical resolution. This vertical
consideration of lateral variations or heteroge- resolution will depend not only on the tool
neities visible in the core sample. While the design, but also on the formation and borehole
concept of resolution is easily described, the characteristics. Thus in terms of log measure-
strict numerical definition of it varies. Theys ments the vertical resolution must be as much a
(1991) provides a theoretical definition of log local value if it is anything, but this value will
vertical resolution: 'The full width at half constrain the conditions under which the mea-
maximum of the response of the measurement sured value relates to the true value of the
to an infinitesimally short event'. He then individual formation. In terms of interpretation,
includes other non-attributable definitions from the true value may not be important except
elsewhere in the literature: qualitative: vertical where the aim is to quantify that particular
resolution is the minimum distance, x such that parameter (e.g. porosity or saturation). The
the logging tool is able to resolve distinct events matrix (or grain) density log from Thistle Hole
separated by this distance; quantitative: vertical 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 (e.g. frequency) and the actual physical dimen-
'calcite dogger' which is sharply bound on either sions over which the measurement is made (e.g.
side by sandstone. The line is derived from the size and shape). Typically, scale may be defined
continuous log measurement, whereas the points quantitatively with precise descriptions of the
relate to specific individual point determinations size and shape of the measurement (see for
on core samples. This is the classic 'shoulder- example Clark 1979 or H o h n 1988). It is linked
bed' effect and the discrepancies are due to the to the measurement technique and hence the
averaging caused by interrogation of a larger design of the tool. Except in isotropic, homo-
volume by the log measurement. In effect, genous media the different aspects of scale will
neither estimate of matrix density is wrong: they be important and will contribute to the measure-
simply relate to different volumes of rock ment data value. Scale may also be linked to the
constrained by the measurement design and thus resolution of the measurement. A simple exam-
require slightly different interpretations. Bed ple of the effect of scale concerns the measure-
resolution is nearly always a problem even with ment of porosity on core plugs. Doveton (1994)
the most finely resolved tools and bed boundary shows how for two porosity datasets extracted
effects are always present. by Baker (1957), respectively from whole core
and plugs, the mean values may be the same but
Scale the variability of the whole core is less than that
of the plugs. There is an apparent rotation of the
Differences of scale are evident in terms of the relationship between the two porosity determi-
relative dimensions of the measurement itself nations in which the extremes in the smaller
INTEGRATION OR CALIBRATION? 43

samples are averaged out. Doveton (1994) corresponding to that measurement.


emphasizes that this core-based example is
equally applicable to the different volumes Quality
sampled by core and log measurements.
As an example of the importance of scale, Quality is defined by the precision, accuracy and
consider a series of different measurements in the bias of the measurement (Murphy 1969). A good
laboratory on a sample of Penrith Sandstone quality measurement will be characterized by
(Fig. 4). In addition, the role of orientation of high precision and accuracy, and a zero bias.
the measurement is also considered. Orientation Precision refers to the closeness of agreement
becomes important where the rock is not both between the results obtained by applying the
homogeneous and isotropic (i.e. for most experimental procedure to a sample several
measurement scenarios in nature). In this times under prescribed conditions. Accuracy
specific experiment the sandstone is an aeolian refers to the closeness of the measurement to
deposit, comparable to the Rotleigendes of the the true value. Whilst precision may be quantifi-
North Sea, of Permian age. It is characterized as able for both the laboratory and downhole
a clean reservoir-type sandstone with rounded measurements, the accuracy is more difficult to
quartz grains, quartz overgrowths, and an assess as the actual (true) values are unknown,
absence of clay phases. Generally the quartz and standards (samples for whom specific
overgrowths reduce the porosity and increase measurements are assumed to be known as
resistance to both electrical and fluid flow 'correct' within definable limits) are virtually
(Harvey et al. 1995). The three images in Fig. 4 non-existent for the measurements under discus-
were obtained through the application of mea- sion here. Consequently the true value is an
surements at the same spacing to the upper idealized concept and the accuracy is a qualita-
surface of the slab (see Jackson & Lovell 1991; tive concept (Theys & Woodhouse 1994). In this
Lovell & Jackson 1991; Harvey et al. 1995). way we are not aware of the correct answer in an
Porosity was measured using image analysis of absolute sense, and hence any bias present
the essentially 2-D surface visual texture, whilst remains unknown. Figure 5 (after Kimminau
permeability was determined using minipermea- 1994) illustrates the concepts of accuracy and
metry measurements which involve transient precision, two terms which are frequently mis-
pressure impulses at point locations, again on understood or confused.
the upper surface. Conductivity refers to the ODP Hole 926B on the Ceara Rise penetrates
electrical conductivity (inverse of resistivity) and sediments which are predominantly ooze and
was measured by an array of surface mounted chalk, with varying concentrations of nannofos-
potential electrodes with remote electrodes sils, foraminifera and clay, together with minor
passing a uniform current through the full components of iron oxide and sulphides. The
volume of rock. The three comparable images variation in CaCO3 in ODP Hole 926B is plotted
relate to very different volumes of rock: the in Fig. 6. The continuous line is derived from
porosity data is restricted to the surface, whilst shore-based measurements in which some 70g
the permeability investigates a hemispherical samples were finely crushed, sub-divided and the
volume of rock (in homogenous isotropic major elements determined by X-ray fluores-
material); in contrast the conductivity is an cence (XRF) spectrometry. The carbonate per-
average value integrated over a vertically orien- centage was determined directly assuming that
tated rectangular prism. Given these significant all the calcium measured occurred as calcium
differences in both scale and orientation the carbonate. The individual points are from the
images may not always show good correlation shipboard measurements which were obtained
although as the figure demonstrates, for the from measurement of acid-liberated CO2 and
Penrith Sandstone, with its relatively simple the assumption that this was all bound up in
structure, there is a reasonable relationship calcite. The precision of this method at one
between the different properties. This is in part standard deviation is reported as less than 1%
because the sands that we have studied are (Explanatory Notes, Curry et al. 1995) as is that
relatively uniform, but perhaps more impor- for the CaO XRF determinations. Thus while
tantly the primary fabric of the samples is the measurement method and the analytical
perpendicular to the longest axis. In less homo- volumes are different for the two datasets, the
geneous materials the different sample volumes precision is similar, and a bias may have been
investigated would lead to greater disparity introduced by the respective (unknown) sam-
between the images. In this way different pling strategies. It is impossible to say which of
measurements may perceive different degrees of the datasets is correct even though there are
homogeneity as a function of the sample volume significant differences between them.
44 M.A. LOVELL E T AL.

accurate and accurate but


precise imprecise

J
parameter value parameter value

A inaccurate but
inaccurate and
imprecise
precise

) L
parameter value parameter value

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

Thompson & Theys (1994) note that quality Calibration


requires the definition of specified requirements
rather than expense or luxury. Thus the defini- Calibration is the process by which measure-
tion of the target or aim of the measurement ments are compared with known standards for
needs to be carefully detailed before we can the purpose of enabling the quantitative com-
assess its quality. Straley et al. (1995) demon- parison of measurements. Thus, calibration
strate the importance of defining the aims in requires samples for which supposedly 'true'
considering the use of N M R in partially values are known in order that accuracy may be
saturated rocks. In attempting to compare defined (Ruth & Pohjoisrinne 1993). This
downhole and laboratory measurements they calibration procedure may involve recourse to
note that mercury porosimetry actually char- local standards in which case the calibrated
acterizes the pore throat size compared to the measurements may be termed relative. Locally
N M R T1 which primarily responds to pore body calibrated data can be easily compared and used
size. Thus the two separate measures of pore without knowledge of their relationship to
dimensions respond to different aspects with world-wide measurements of the same para-
consequently different answers. meter. These local standards may, in turn, be
INTEGRATION OR CALIBRATION? 45

i ' I i
9 J

i
i
i

9 CaCO 3 (shipboard laboratory)


16 d
!
! 9 CaCO 3 XRF shore laboratory

17
9 !

Leg 154

Hole 926B
19 Ceara Rise

i
!

20

9 iI
i
J
i

i
t
22
5 6 7 8 9

% 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 of the raw measurement to usable values is


standards providing so-called absolute values designated separately as the tool response.
of measurement which can be related through Consequently, calibration is based on a log
the same measuring stick. Whilst there is much measurement in large volume artificial or
to be said for global standards enabling com- natural formations. Theys (1991) also considers
parison of all data for a single parameter, that the checking of a logging instrument in well-
calibration does not necessarily imply truth: defined conditions both before and after a
inherent bias in the measurement of a systematic logging run should really be labelled verification,
nature, in which the measurement effectively whilst the matching of surface electronics to
measures something other than what was downhole signals could be better described as
intended (Eisenhart 1962), can simply yield surface system alignment. This is because neither
consistency. The importance of calibration in considers the actual calibration of the data,
the integration process depends more on the simply the overall working of the tool within
nature of the measurements and how they are to predefined calibration constraints.
be used. Core measurement calibration, meanwhile, is
In terms of downhole logging calibration is well-documented and generally involves the use
often confused with other measurement checks. of standards of a similar scale to the samples
The process of calibration of logging tools under test. These small volume local standards
concerns the production of a specific signal in can be readily controlled and related to national
response to known measurement values within a or international standards. Skopec (1992), how-
formation (Theys 1991); thus the transformation ever, 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 measure- techniques, strategies, or acquisition procedures.
ments are compared (e.g. nuclear spectroscopy R a t h e r it involves the process of relative
logs), each method must be examined carefully calibration between small and large volume
to determine experimental limitations, accuracy measurements, typically between core and log.
and precision in testing, as well as potential Where datasets do not agree it is important to
mineral alteration processes that can occur when ask whether the issue is one of data quality or
a rock is sampled. the quality of interpretation (Owens 1994;
Given a satisfactory understanding of our Harvey et al. 1998).
measurement base we can proceed to analyse the As an example of this process, consider Fig. 7.
data. Often we are concerned with combining Here two datasets are shown at different scales,
two datasets of the same parameter with the aim but both relate to measurements on core. The
of producing one, more complete dataset. In this continuous electrical Formation Factor log is
way core data may supplement sections of log derived from the electrical resistivity image as
data, or indeed duplicate it. The normal presented in Fig. 4. These data are then
procedure here is to assume that one of the compared with the solid bar which is the
datasets is correct and to adjust the other to electrical Formation Factor determined in a
create a best fit. This equalization can create traditional manner on a minicore or plug. This
better coverage of the total borehole section but was originally done during attempts to calibrate
will inevitably involve manipulation of at least the novel imaging system against industry
one dataset and the loss of inherent absolute standards (Jackson et al. 1994). The match
values. This is a standard approach to so-called between the two datasets is improved visually
core-log integration. Unfortunately it does not by smoothing the image data further (dots). The
consider discrepancies between the two measure- smoothing was carried out with a simple moving
ment sets created by different measurement average with a window width corresponding to
INTEGRATION OR CALIBRATION? 47

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 resolu-
tion.
A related problem occurs when we are trying
to predict petrophysical properties from unre-
lated 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. Effec-
tively we calibrate our model or algorithm to
give answers which are compatible with labora-
tory 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 Fig. 8. Comparison of core and log data simulated for
and core data the principles remain true for ODP Sites 792 and 793 (Ocean Drilling Program Leg
integrating these data with smaller scale (thin 126) demonstrating the integration of measurements at
section, SEM) and larger scale (VSP, seismic different resolutions.
reflection) data.

Integration based on ODP Sites 792 and 793 in the region


of the Izu Bonin Arc (Lovell et al. 1992), was
Integration involves the reconciliation of data- simulated in three steps:
sets with or without the equalization involved in
calibration procedures. Often this will include (1) generation of a lithological sequence with
the selective addition of data. Different datasets abundances and thicknesses of simulated
may relate to the same measurement, the same units corresponding to the appropriate
scale, or either or both of these attributes may be distribution obtained from core logging;
different. The overall aim of integration is to (2) generation of the rock chemistry at 1 cm
maximize the i n f o r m a t i o n available in an intervals throughout the section, preserving
optimal manner. Towards this aim, the objective the average and variance/covariance rela-
is not simply to compare data but to constrain tionships of the chemistry for each lithol-
and characterize some geological process or ogy;
effect. (3) sampling of the simulated chemical se-
The effects of sample size and tool resolution quence.
in core-log integration is easily demonstrated by
a simple Monte Carlo experiment which could In Figure 8 this simulated dataset is sampled for
go some way to explaining the variation seen in alumina (a) as it would respond to the geochem-
the CaCO3 estimates shown in Fig. 6. In this ical logging tool (GLT) by averaging over a 60
particular experiment (Fig. 8) a 30 m section of cm window and reporting measurements every
oceanic sediment, with basaltic lava flows, was 15 cm to provide the continuous log curve, and
simulated to evaluate the suitability of core (b) as a set of core plug results obtained by
measurements as guides to the accuracy of randomly sampling a small number of the total
geochemical log measurements. The section, 3000 simulated compositions to provide the
48 M.A. LOVELL E T AL.

individual point measurements. Both datasets


present different perspectives of the same che-
mical sequence; neither is wrong and conse-
quently 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 informa-
tion 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 sedimen-
tary cover in Hole 896A and the position of Fig. 9. Stratigraphy derived from core recovery based
sediment/basement interface was based upon on core barrel descriptions, compared with stratigra-
rubble being felt by the drill bit at 179 mbsf phy and based on downhole electrical FMS images and
(metres below sea floor) and the hole was cored core observations.
from 195.1 mbsf to 469 mbsf (Alt et al. 1993).
Within this drilled section, core recovery aver-
aged 26.9%. Pillow lavas (57%) and massive based on an overall poor core recovery (26.9%,
flows (38%) dominate the cored material, with Alt et al. 1993), which is also very variable
breccias (5%) and two small dikes accounting within individual sections of the borehole (Fig.
for the remainder of the material. With the 9). Shipboard scientists produced the lithology
exception of pillow rims, the majority of the shown in Fig. 9 based solely on visual observa-
rocks are slightly altered (< 10%) and variably tions of recovered core.
veined (Alt et al. 1993). Pervasive background In contrast, shore-based scientists (Brewer et
reducing alteration coupled with saponite and al. 1995) have produced a comparable stratigra-
minor pyrite replacement of olivine has led to phy based on the variations in texture of the
the grey colour of the core. Oxidative alteration downhole Formation Micro Scanner (FMS)
is manifested by dark grey to yellow and red Images together with sonic, resistivity, and
alteration halos which commonly occur around gamma ray logs. The FMS tool produces images
smectite veins (Alt et al. 1993). In the pillow of the borehole wall dependent on variations in
lavas and massive flows, veins are usually the measured electrical resistance and these
< 1 mm in thickness and commonly infilled by images can be analysed texturally to develop a
dark and light green saponite and aragonite. log-based stratigraphy with reference to the
Other vein minerals include analcite, fibrous core. As Brewer et al. (1995) demonstrate, there
zeolite and pyrite. All of the previous data were are substantial differences between the stratigra-
INTEGRATION OR CALIBRATION? 49
resistivity smoothed pixel mean pixel permeability
(ohm-m) value value (roD) porosity (%)

i j

e
\
0
-4
i o , .t-_.
.... 0"2 4 ........ 9
O. . . . . . . - JO
Ik

e O" --~-e ....


e-..t
"lP. --|ii.
11

e.. e r e ....
i~ 9i:e
:,
e
o o

ie
.!
k i 2 k ~ i

3 4 5 165 175 185 160 170 180 0 2000 4000 0 20 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 ment of electrical textures on recovered core for
integrating core and log information. comparison with downhole images, thus con-
Here the problem may initially be seen as one straining in a quantitative manner the inter-
of constraining the downhole data through pretation of the downhole data.
selected core observations, knowing that the A different example of the integration of data
recovered core is present in the drilled section, from different sources is shown using the
and utilizing the downhole data to extend the electrical conductivity image from Fig. 4 con-
interpretation to the full depth of the hole. verted to a resistivity image (by taking the
However, the core data are inherently biased, inverse of each plotted value). In Fig. 10 this
due possibly to preferential sampling of some resistivity image is converted into a microresis-
lithologies, incomplete recovery, and the defined tivity log by averaging across each row of values.
criteria and procedures used to identify and Similarly, the porosity and permeability images
extrapolate recovered material over the total are displayed as averaged micrologs. These logs
depth drilled. Thus whilst the core does indeed again show the variable nature of the formation
represent the truth, its allocation to a particular but could easily be incorporated into a petro-
lithology, distribution with depth and continuity physical analysis routine for producing im-
may be questioned. In contrast the FMS images proved estimates of both fluid volume and flow
are relatively new and lack precise calibration in parameters. Here the photographic image has
terms of textural detail and lithological re- been converted to a pixel log, again by row
sponses. They are usually continuous and are averaging; this micro-log is then smoothed to a
based on electrical, not visual, properties which similar resolution to the electrical micro-log.
may or may not be equable, and whose equality There is an inverse correspondence between the
may vary within the hole. These images prob- two which suggests the use of photographic
a~ contain bias in addition to that within the images for comparison with electrical images in
core. Thus, rather than accept one dataset in its clean sandstones. Where clay minerals may
entirety as the truth and reject the other, it contribute to conduction processes, the rock is
would be better to use the ground-truth of the contaminated with mud, there are clay-filled
recovered and described core as calibration fractures, or the pore fluid is resistive (e.g. fresh
points for the interpretation of the downhole water), the correspondence between photo-
images. This would ideally include the measure- graphic and electrical images may not be as
50 M. A. LOVELL ET AL.

reliable, nor as predictable. Thus discrepancies Hole 896A from FMS images. Scientific Drilling,
between optical and resistivity images or logs 5, 87-92.
may also yield information about the nature of CLARK, I. 1979. Practical Geostatistics. Elsevier,
the pore space. London.
CURRY, W. B., SHACKLETON,N. J., RICHTER, C. & the
Shipboard Scientific Party. 1995. Proceedings.
Summary ODP Initial Reports., 154: College Station, TX
(Ocean Drilling Program).
(1) Interpretation of core and log data should DOVETON, J. H. 1994. Geological Log Analysis Using
involve consideration of the measurement pro- Computer Methods. American Association of
cess, calibration, and integration. Calibration Petroleum Geologists, Computer Applications in
and integration may not necessarily be included Geology, No.2.
in the first interpretation. EJSENHART, C. 1963. Realistic evaluation of the
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 Instru-
functions 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
attributes are considered in any data integration problem of accuracy in core-log calibration. This
exercise. volume.
(3) Measurements can be calibrated, either - - LOVELL, M. A., JACKSON, P. D., ASHU, P. A.,
relatively (locally) or absolutely (globally). Cau- WILLIAMSON, G., SMITH, A. S., BALL, J. K. &
FLINT, R. F. 1995. Electrical resistivity core
tion is essential since often this falls to inter-
imaging III: characterisation of an aeolian sand-
dataset 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.
achieve. Comparison of fracture apertures computed from
(4) Differences of scale in measurement sets electrical borehole scans and reflected Stoneley
may highlight geological features through varia- waves: an integrated interpretation. The Log
tions in the degree and nature of formation Analyst, 33, 50-66.
JACKSON, P. D. & LOVELL, M. A. 1991. The corre-
heterogeneity.
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, appro- London, UK. Paper J.
priate selection of measurement process, includ- 9HARVEY, P. K., BALL, J., WILLIAMS,
ing necessary calibration, and optimization C.I FLINT, R. F., ASHU, P. A. & MELDRUM,P. I.
using carefully-defined integration. This should 1994. Advances in resistivity core imaging.
realize an interpretation which maximizes the Transactions 35th Symposium, Society of Profes-
sional Well Log Analysts. Paper GG.
use of available data whilst remaining compa-
KIMMINAU, 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 geophy-
sical 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 Trans-
References actions: Society of Professional Well Log Ana-
lysts, Midland, Texas.
ALT, J. C., KINOSHITA, H., STOKKING, L. B. & the , PEZARD, P. A. & HARVEY, P. K. 1992.
Shipboard Scientific Party. 1993. Proceedings. Chemical stratigraphy of boreholes in the Izu-
ODP Initial Reports, 148. College Station TX Bonin Arc from insitu nuclear measurements.
(Ocean Drilling Program). Proceedings of the Ocean Drilling Program,
BAKER, P. E. 1957. Density logging with gamma rays. Scientific Results, 126, 593-601.
Petroleum Transactions of the American Institute Moss, B. 1997. the partitioning of petrophysical data:
of Metallurgical Engineers, 210, 289 294. a review. In: LOVELL,M. A. & HARVEY,P. K. (eds)
BREWER, T. S., LOVELL,M., HARVEY,P. & WILLIAMSON, Developments in Petrophysics, Geological Society
G. 1995. Stratigraphy of the ocean crust in ODP Special Publications, No. 122, pp 18l 252.
INTEGRATION OR CALIBRATION? 51

MURPHY, R. B. 1969. On the meaning of precision and Scientific Correspondence, Nature, 332, p.495.
accuracy. In. Ku, H. H. (ed.) Precision Measure- SKOPEC, R. A. 1992. Recent advances in rock
ment and Calibration. Statistical Concepts and characterisation. The Log Analyst, 33, 270-285.
Procedures. United States Department of Com- STRALEY, C., MORRISS, C. E., KENYON, W. E.
merce National Bureau of Standards Special HOWARD, J. J. 1995. N M R in partially saturated
Publication 300, 1, 357-360. rocks: laboratory insights on free fluid index and
OWENS~J. 1994. Fit-for-purpose data during field life. comparison with borehole logs. The Log Analyst,
The Log Analyst, 35, 58-60. 36, 40-56.
RUTH, D. & POnJOISRINNE, T. 1993. The precision of THEYS, P. P. 1991. Log data acquisition and quality
grain volume porosimeters. The Log Analyst, 34, control. Editions Technip, Paris.
29-36. THEYS, P. & WOODHOUSE, R. 1994. Society of profes-
SHERIFE, R. E. & GELDARX, L. P. 1982. Exploration sional well log analysts topical conference on
Seismology volume 1. history, theory and data quality, appendix: metrological definitions. The
acquisition. Cambridge University Press, Cam- Log Analyst, 35, p. 71.
bridge. THOMPSON, B. & THEYS, P. 1994. The importance of
SIES, H. 1988. A new parameter for sex education. 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 ICP-
AES 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 measure-
ment 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 Methods have been developed for the quantifi-
by an 'in situ' geochemical logging tool (GLT) cation of the errors arising from the sampling of
can have p r o f o u n d effects on the realistic one site, by either a single or multiple samplers.
interpretation of the geochemical variation For a single sampler, the methods have been
across a stratigraphic sequence. Although the applied both to the use of a single sampling
importance of the uncertainty is becoming protocol (Ramsey 1993) or comparisons be-
apparent, methods for the estimation of such tween several protocols (Ramsey et al. 1995b).
uncertainty are lacking. Broadly similar studies For the case of multiple samplers, different
on the estimation of bias and precision of GLT m e t h o d s have been devised d e p e n d e n t on
measurements have been reported (Wendlandt & whether all samplers were applying the same
Bhuyan 1990; Grau et al. 1990), but these did protocol (Ramsey et al. 1995a) or different
not attempt any rigorous mathematical estima- protocols (Argyraki et al. 1995). Applications
tion of the total uncertainty of measurement. of these methods were made for the estimation
For analytical measurements in isolation, the of heavy metals on contaminated land, but the
realistic estimation of uncertainty has already methodologies are equally applicable, in princi-
become an important issue (ISO 1993a; Euro- ple, to the measurement of elemental variation
cheni 1995; A M C 1995). In contaminated land within a borehole using a GLT.
investigations it has recently been shown that it This paper considers how estimates of mea-
is field sampling, rather than the chemical surement uncertainty can be derived, particu-
analysis, that can contribute the largest source larly for the case of a single sampler using a
of measurement error and will therefore limit the single protocol, utilizing a previously published
measurement uncertainty (Ramsey et al. 1995a). 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 53
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
54 M. H. RAMSEY ET AL.

the work are therefore: (Eurochem 1995). The alternative 'top down'
approach uses inter-laboratory trials to estimate
(l) to make the best estimate of uncertainty the total uncertainty of a measurement. In this
on the measurements from the case study, method, many selected laboratories ( n > 8 )
as an example of the general approach; would all analyse the same sample by the same
(2) to identify additional methodologies that analytical method (AMC 1995). The scatter of
could be applied to give improved esti- all reported measurements is then used as an
mates of uncertainty; overall estimate of uncertainty.
(3) to consider briefly how to identify accep- The 'bottom up' approach has the limitation
table levels of uncertainty for particular that it requires all of the sources of uncertainty
objectives of interpretation. 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,
Definition and terminology of measurement volumetric additions). However, the most im-
uncertainty portant source of uncertainty may not be explicit
in the method (e.g. lab. temperature), and is
The formal definition of uncertainty has been therefore easily overlooked, especially by in-
given as: 'A parameter associated with the result experienced practitioners. Furthermore, it can
of a measurement, that characterizes the disper- be a long and expensive procedure to quantify
sion of the values that could reasonably be all the component errors, if the method is to be
attributed to the measurand (ISO 1993a). A less applied rigorously.
formal but more understandable description of The benefits of the 'top down' approach can
measurement uncertainty is that it is 'an interval be appreciated from the differences that are
a r o u n d the result of a measurement that often evident between laboratories in inter-
contains the true value with high probability' organizational trials. These differences are often
(Thompson 1995). The standard uncertainty 'u' larger than can be accounted for by the
can be considered equivalent to one standard individual estimates of uncertainty within each
deviation which is often used to describe a laboratory (AMC 1995). This is because the
normally distributed random error. The ex- ' b o t t o m up' a p p r o a c h used by individual
panded uncertainty 'U" is equal to the product laboratories tends to give over-optimistic esti-
of the standard uncertainty and a coverage mates of the uncertainty. The limitation of the
factor 'k', which typically has a value of 2 or 3. 'top down' approach is that it depends on the
This is analogous to the use of multiples of selection of the laboratories that contribute. If
standard deviation for the quantification of the laboratories all use a similar source of
precision. In the formal definition, systematic calibration, they may all be equally biased and
errors are not included in estimates of uncer- therefore give an under-estimate of the uncer-
tainty, but only any residual random errors left tainty. Alternatively, one laboratory may have
after correction of the systematic errors. How- gross errors, atypical of the application of the
ever, in this application, the distinction between method as a whole, and this will cause an over-
systematic and random errors becomes blurred. estimate of the uncertainty.
The systematic error of one sampler becomes a
random error when assessed as part of a multi-
sampler comparison and systematic errors there- Estimating uncertainty for in situ borehole
by become incorporated into estimates of
uncertainty. measurements
There are two primary limitations in estimating
Estimating uncertainty in chemical analysis uncertainty of measurements made in situ in
boreholes, using the method described for purely
The methods recently developed for estimating laboratory based analytical systems. One pro-
uncertainty in chemical analysis in the labora- blem is that these methods ignore the uncer-
tory are of two types. They both need to be tainty arising from field sampling. It is often
evaluated as options for the application to in situ quoted that an analysis can never be of better
borehole sampling and analysis. In the 'bottom quality than the sample upon which it is made.
up' approach the random error from each What has been lacking, however, is the means of
individual component of a method is quantified estimating the size of the uncertainty which is
separately as a standard deviation (s). The introduced by field sampling. A second limita-
overall uncertainty is then estimated by sum- tion is in the quantification of systematic errors,
ming the individual errors by their variances (s 2) either from sampling or from chemical analysis.
ESTIMATION OF MEASUREMENT UNCERTAINTY 55

It is proposed to adapt the methods devised farm in Northumberland have been given else-
originally for estimating analytical uncertainty where (Sams et al. 1995). In brief, GLT
to the estimation of uncertainty from in situ measurements were made on a 40m length of
measurements by addressing these two limita- borehole drilled through a single Namurian
tions. cyclothem, with varied lithologies (Fig. 1).
With in situ measurements it is useful to The GLT has been described in detail by
consider field sampling and chemical analysis as Hertzog et al. (1987) and uses three different
just two parts of the same 'measurement' measurement techniques:
process, and to quantify their combined con-
tribution to the uncertainty. Such 'total mea- (i) The natural radioactivity of K, Th and U
surement' uncertainty has therefore four is used for their determination with a
potential components. These are the sampling N a t u r a l G a m m a Ray Spectrometer
and analytical contributions to random error (NGS; Mark of Schlumberger). The count
(i.e. sampling and analytical precision) and any rates obtained are directly proportional to
uncorrected systematic errors (i.e. sampling and the mass per cent of the element, provid-
analytical bias). ing a borehole correction factor is applied.
Taking the 'bottom up' approach to estimat- (ii) A neutron source of 252Cf, emitting
ing the total measurement uncertainty we can neutrons at about 2.3MeV is used to
review the methods available for the estimation activate A1, which is determined using an
of these four components. Analytical precision AACT; Mark of Schlumberger. Results
can be measured by the use of analytical obtained are proportional to the weight
duplicates (Thompson & Howarth 1976) or in per cent of A1 after an environmental
combination with sampling precision using a correction is applied.
balanced design of sampling and analytical (iii) After neutron capture from a burst of 14
duplicates (Garrett 1969; Ramsey et al. 1992). MeV neutrons, Si, Fe, Ca, Ti, S and Gd
This is possible for the GLT although, ideally, are determined using a tau-gated thermal
duplicate sampling would require the use of a neutron capture spectrometer with a
second borehole close to the first and assuming Gamma Spectrometer (GST; Mark of
lateral homogeneity. Analytical bias is usually Schlumberger). This procedure only pro-
estimated by the analysis of certified reference vides relative concentrations of these
materials (Ramsey et al. 1987), but this ap- elements, and they have to be converted
proach would be problematic for the GLT. In to absolute values by imposing a closure
addition, there are no methods in general use for relationship on the results obtained.
the estimation of sampling bias. For absolute
bias this may require the introduction of An oxide closure relationship is imposed on
reference sampling targets, analogous to refer- the results, in order to derive element concentra-
ence materials for the estimation of analytical tions from the raw GLT data It is assumed that
bias (Thompson & Ramsey 1995). Sampling bias each element occurs as a single oxide or
has already been estimated however, between carbonate in the formation and that the sum of
the concentration estimates made by the appli- the oxide and carbonate fractions is unity. This
cation of different sampling protocols at one site assumption is known to be in error, but it is
(Ramsey et al. 1995b). This approach does not, considered that the errors involved will be small
however, give bias against an 'accepted reference (< 5%) for most lithologies. The equation to be
value', as defined by ISO (ISO, 1993b). solved is (Hertzog et al. 1987):
Taking the 'top down' approach, it should be
possible to use measurements from inter-orga- F(~,Xi Yi/Si) + XK WK -[- XA1WA1 = 1.0
nizational sampling trials, such as sampling where:
proficiency tests (Argyraki et al. 1995) and
collaborative trials (Ramsey et al. 1995a) to F is a calibration factor to be determined at each
estimate uncertainty. The potential advantages depth point;
and practical feasibility of such an approach to Yi is the fraction of the measured prompt
in situ borehole determinations will be consid- gamma rays attributed to element i;
ered below. Si is the tool sensitivity for element i;
Xi is the ratio of the mass of the associated oxide
Experimental details of the GLT case study or carbonate to the mass of element i.

The details of the case study using the GLT at The mass fractions of potassium and aluminium
the Imperial College borehole at Whitchester (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 and mineralogy over a single sedimentary cycle
normally determined from the density log by using the same depth interval as the GLT (140-
assuming a quartz matrix. 180 m). Exact correspondence of the volume of
Duplicate field measurements were taken at rock analysed by both GLT and ICP-AES is
the site in Northumberland by lowering the clearly impossible (Fig. 2). The rationale behind
GLT twice through a section 110-230m in the sub-sampling of the core used for analysis is
depth. On one occasion all raw measurements that it should characterize the geochemistry of
were processed through the oxide closure to give the core at small intervals (e.g. 25cm), so that
elemental concentration, but on the second subsequent mathematical smoothing of results
occasion the raw results were initially unpro- can be employed to approximate the sampling
cessed. In order to achieve comparability, the volume analysed by the GLT (110,000 cm3). The
original dataset was divided by the correspond- inevitable sampling bias introduced by litholo-
ing raw counts for each element at a particular gies of widths less than 25cm (e.g. iron rich
depth. This procedure gave a value for the factor sideritic nodules) can then be estimated and
FXi/Si, and both F and Si should be constant at recognized as a systematic difference between the
a constant depth. If the raw counts for an two techniques. Discrete samples of about 30 g
element taken from the duplicate dataset are were selected with an average vertical spacing of
now multiplied by the corresponding value of 25 + 5 cm between samples, the rock chips being
FXi/Si, an estimate of the absolute value for that taken to represent the full lateral heterogeneity
element at a particular depth can be made. This of the core sample. A total of 147 analytical
procedure is not quite the same as comparing specimens were collected over the entire depth
data obtained from two runs processed inde- interval and powdered in an agate Tema Mill to
pendently. less than 75 #m.
The details of the analytical procedure have
Sampling and analysis of core samples been given elsewhere (Sams et al. 1995). In brief,
0.25g of dry rock powder was totally decom-
Core sampling was aimed at investigating the posed by fusing with LiBO2 and dissolving the
vertical variations in bulk rock geochemistry fused bead in dilute HNO3, with six elements
ESTIMATION OF MEASUREMENT UNCERTAINTY 57
Sampling design
Estimation of random component

The analytical precision of GLT measurements


was characterized as a function of concentration
using the method using duplicated analyses
25:1:5 cm
devised by Thompson & Howarth (1976). The
mathematical model of precision was:

sc = So + Pc (1)
ca. 60cm / /

/ /
where so=standard deviation at a particular
concentration c;
//
/ /
So= standard deviation at zero concen-
/ z
tration;
/ / c = concentration;
/ / 0 = slope factor, related to the high level
/ / precision and so, So and c all have the
same units of concentration.
20cm diameter borehole Corresponding core, with cut sections
for GLT analysis used for ICP-AES analysis The random component of the measurement
Fig. 2. Sampling designs for the measurements by GLT uncertainty, using a coverage factor of two, for
and by ICP-AES. zero concentration is given by:

Uo = 2s0
being determined by inductively coupled plas-
ma-atomic emission spectrometry (ICP-AES). and for a particular concentration c by:
This instrument was calibrated using 10 inter-
national reference materials, (BR, GA, GH, Ur = 2sc
NIM-N, SY-2, UB-N, NIM-G, DTS-1, AN-G
and MICA-MG). The certified values for the Thus, multiplying equation (1) by the coverage
reference materials are given in Govindaraju factor of two we get:
(1994) and thus, these act as the traceability of
the determinations to 'accepted reference va- Ur = Uo -+-20c (2)
lues', as required by ISO (ISO 1993b) for the
estimation of bias. The bias between measured The upper limit of the uncertainty on an
and certified values of the reference materials for estimation of concentration c, from equation
the 6 elements studied was generally less than 1 (2) is given by:
%. Values for the precision of analyses of these
six elements in silicate rocks by ICP-AES are c+ Ur=c+ 20c+ Uo
reported as < 0.5% in Ramsey et al. (1995c). 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
Ur% = 200Sc/C
GLT measurements
The two components of measurement uncer- At high levels of concentration where Ur>>Uo,
tainty considered initially are the analytical this can be expressed as:
precision (for random error) and the analytical
bias (for systematic error). The role of the Ur% = 2008 (4)
random component is explicit in the ISO guide
(ISO 1993a), but the systematic component for a This value of Ur% is approached asymptotically
particular technique such as GLT analysis has to at high concentration.
be estimated independently. Whether this latter Substituting for 0 from equation (4) in
component is used to correct the concentration equation (3) we get:
estimates, as suggested by ISO, or added into the
uncertainty estimate will be discussed below. 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 Random Uo Translational Rotational


(mass-%) Error, Ur % (mass-%) Bias, (mass-%) Bias, %

Si 044 2.7 1.80 - 1.53" 3.5


AI 0-12.5 7.6 0.62 1.12" - 10.1"
Fe 0-13 12.1 0.18 0.34* - 54.6*
Ca 0-38 5.0 1.28 0.38 4.3*
K 0.5 2.8 n.d. 0.48 0.20* - 13.6
Ti 0.1-2.7 n.d. n.d. 0.003 11.1
S 0.6-5 71 0.56 n.d. n.d.

n.d. = not determined.


* Bias value significantly different from zero (p = 0.05).

Table 2. Estimation of detection limit and precision of general relationship for all elements studied. The
GLT determinations (Sams et al. 1995). same results are transformed into the form of
relative uncertainty (Ur%) in Fig. 3B. For Si,
Element Detection High-level there is a significant increase in uncertainty with
limit (mass-%) Precision,% (Is) concentration, but when converted into relative
uncertainty at the high level, this was the best of
Si 2.70 1.35
any element at 2.7%. The uncertainty at zero
A1 0.92 3.83
Fe 0.26 6.05 concentration for Si was 1.8 mass %, which is
Ca 1.92 2.51 well below most of the samples from this
K 0.71 n.d. borehole. Hertzog et al. (1987) do not give a
S 0.56 35.45 standard deviation at zero mass% but give
precisions of 25.6% and 4.9% at 25 mass% Si
n.d. = not determined and 47 mass% Si, respectively.
For Fe, considering these random errors in
isolation gives a relative uncertainty of 12.1% at
These two components of random uncertainty
high concentration levels, and an uncertainty of
(Table 1) were estimated for 6 elements (Si, A],
0.16 mass% at zero concentration. The bias
Fe, Ca, K and S) by using a dataset consisting of
detected for this element, discussed below, will
600 duplicate borehole measurements taken over
however cause this latter estimate to be multi-
a section between l l 0 m and 230m in depth.
plied by a factor of 2.2, giving 0.35 mass%.
This method also provides standard errors on
Hertzog et al. (1987) quote a precision of 30% at
these uncertainty estimates, which can be used to
1 mass% Fe and one of 16% at 5 mass% Fe,
check whether the estimates are statistically
which appear broadly comparable.
greater than zero. When the So value is not
For S, the high level relative uncertainty was
significantly greater than zero, it is still possible
found to be over 70% at all concentrations, the
to calculate a maximum value for Uo, that is
highest for any element determined. This may be
equivalent to a maximum method detection limit
due to the proximity of sulphur concentrations
( M M D L ) (Ramsey et al. 1995c). The precision
in the samples to the detection limit of the GLT.
and detection limits of the G L T determinations
It is interesting to examine what limitations such
are listed in Table 2. Data used for calculations
a degree of uncertainty places on the geochem-
made in this paper is represented graphically in
ical interpretation of the concentration estimates
Fig. 3, and a summary of the duplicate data used
(see below).
to determine the precision of G L T measure-
ments is given in Fig. 4 in Sams et al. (1995). The
individual data sets obtained from ICP-AES and Estimation of systematic component
G L T measurements are available, but too
numerous to include in full here. They are A qualitative estimate of the analytical bias of
available from the Geological Society if re- the G L T measurements was made using X-ray
quired. fluorescence analysis on a limited selection of
The variations of uncertainty due to analytical core material by Hertzog et al. (1987). The
precision for Si and Fe are plotted against quantitative estimation of bias needed for
concentration (c) in Fig. 3A, to exemplify the uncertainty estimation has been reported by
ESTIMATION OF MEASUREMENT UNCERTAINTY 59

(a) (b)
1.4-

1.2"

o.
1.0-
.o

3.0
0.8-

0.6-

2.0
0.4-

.o'o 0.2-
o.

0.0
' I , I ' I , I ' I , I ' I , i l
10 20 30 40 1 2 3 4 5 6 7 8 9

(A) GLT Si (mass %) GLT Fe (mass %)

(a) (b)
100

80

60

60 . "

,m~ zo "

"0 , " ' " 0 i , i , I , i , i , i , i , i i


10 20 30 40 I 2 3 4 5 6 7 8 9

(U) 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 Before comparing results obtained using the
obtained from core samples using ICP-AES two techniques the question of sample size must
(Sams et al. 1995). The method employed used be addressed (Fig. 2). Samples collected for
simple linear regression to determine the rota- laboratory analysis represent a depth of approxi-
tional and translational components of bias. mately 1 cm. Overall, the GLT responds to an
Translational bias occurs when the intercept of approximately spherical volume of rock with a
the regression, line is statistically significantly radius of about 30cm, although each measure-
different from zero. Rotational bias occurs when ment technique used, i.e. NGS, AACT and GST,
the slope of the regression line is statistically will interact with different volumes of rock.
different from unity. The test for significance Assuming there is lateral homogeneity, the GLT
used in this study was the student t-test with a response is some weighted depth average of
probability of 0.05. Potential limitations in this vertical variations. In addition, it is usual to
general technique (Thompson 1982) have been apply a vertical smoothing process to the raw
shown not to be significant in this case (Sams et data before the oxide closure is performed. A
al. 1995). running average of 5 data points, or just over
60 M.H. RAMSEY ET AL.

Silicon Iron
25

4O 20.

8
~r 15'

E~ ~ 10"
I'--

(5
5.

9 .~':.':d'.'-..' ":'
, ' , ' i , , , , ~ , , , , L,,, ,i, ,,, 0
10 20 30 40 50 0 5 10 15 20 25

ICP mass percent 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. Estimates of the components of the bias in the
Therefore, raw measurements obtained from the element concentrations determined by the GLT
GLT and ICP-AES represent significantly dif- compared with ICP-AES are given in Table 1.
ferent depth intervals of rock. In order to make Significant rotational bias was found for all of
a comparison which is not biased by this the elements except Si, and significant transla-
difference, the ICP-AES data must be depth tional bias for all except Ca and Ti. Silicon
averaged in a comparable manner to the GLT shows a statistically significant translational bias
data, although this assumes that each laboratory of - 1 . 5 3 mass%, but no detectable rotational
measurement is representative of a 25 cm depth bias. This linear model for the bias is perhaps
interval, i.e. the sampling interval. Smoothing oversimplified as the majority of data points
was applied to the ICP-AES data using windows with less than 25 mass% Si fall below the
of varying lengths. The window chosen was one regression line (Fig. 4). In terms of the un-
that gave maximum correlation between the certainty there would seem to be a case for
GLT and ICP-AES data, and a square window adding 1.53 mass% to all the GLT estimates of
1.4m wide (equivalent to 6 analyses) was found Si concentration. This is rather simplistic,
to be the best. Thus the smoothed ICP-AES partially because of the over-simplified model
results, at any particular depth, are based on 6 used for the bias but also because this bias may
separate determinations around that depth. The well be different for different lithologies in
unsmoothed and smoothed ICP-AES measure- different boreholes. It is perhaps more prudent
ments for Si and Fe are plotted for comparison therefore to combine an estimate of the possible
alongside the GLT measurements against depth bias into the overall estimate of uncertainty. For
in Fig. 1. Fe, measurement by the GLT shows significant
In order to compensate for the difference in bias in both the rotational ( - 5 4 . 6 % ) and
sample interval the GLT determinations were translational (0.34 mass%) components, which
interpolated using a cubic spline, and depths again needs combining with the random com-
equivalent to the ICP-AES sampling were taken. ponent of the uncertainty for evaluation.
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 Combined estimates o f measurement
depth. Measurements taken by the GLT in the uncertainty
coal layers (150 and 154.3m in depth) were
removed from the comparison due to excep- The total expanded uncertainty (U) can be
tional errors caused by the effect of the estimated from the combination of the random
anomalous density of the coal propagated component Ur% described in equation (5), and
through the closure calculation (Sams et al. the systematic components resolved into rota-
1995). tional bias (BR) and translational bias (BT). The
ESTIMATION OF MEASUREMENT UNCERTAINTY 61

upper limit of the uncertainty for an estimated that the sensitivity factor for iron, Sve, is too
concentration c is then given by: high. Secondly, it is possible that occasionally
the sampling procedure used for ICP-AES
c + U = c (1 + Ur%/100)(1 + BR/100) + Uo + BT (6) determinations dramatically over-estimates the
Fe content of some samples. As mentioned
and the lower limit is given by: earlier, the ICP-AES data is assumed to be
representative of a 25 cm depth interval. How-
c - U = c (1 - Ur~ + BR/100 ) - - Uo + BT (7) ever, in the case of Fe this may not be true. Iron
occurs partially as sideritic nodules in the
For the case of Si, ( U r % = 2 . 7 % , Uo = 1.8 mudstones. If present in a core sample these
mass%, B R = 0 and BT = --1.53 mass%) equa- nodules could produce an analysis of up to 30
tion (6) gives: mass% Fe, which is probably a gross over-
estimate of the average value for that 25cm
c+U =1.027c+1.8+5.53=1.027c+3.33 depth interval. If a nodule is not sampled then
mass% the average value for the Fe concentration of the
25cm interval will be somewhat under-esti-
and equation (7) gives: mated. Overall therefore the heterogeneity of
the iron would not be expected to produce the
c - U = 0.973c- 1.8 + 1.53 = 0.973c-0.27 mass% large bias found in this case.

For an estimated Si concentration of 5 mass% Discussion


the uncertainty interval would therefore be 1.535
to 5.405 mass% Si, and for a high concentration The calculations used to estimate the overall
of 40 mass% it would be 35.59 to 41.35 mass% uncertainty of Si and Fe are equally applicable
Si. for the other elements measured by the GLT,
In the case of Fe, again assuming that the bias but there are a number of limitations which may
is not realistically correctable, the upper limit of mean the values obtained are under-estimates.
the combined uncertainty from equation (6), There are other causes of 'random' error in
where (Ur%=12.1%, Uo=0.18 mass%, the measurement system that have not been
BR = --54.6 and BT = 0.34 mass%) is given by: investigated. In contaminated land measure-
ments it has been shown that multiple applica-
c + U = c (5 + 12.1%/100) tions of the same measurement protocol, on
(1 - 54.6/100) + 0.16 + 0.34 different occasions, by different operators,
c + U = cx 1.121 x0.454 + 0.16 + 0.34 causes appreciable increases in the uncertainties
(Ramsey et al. 1995b). When different measure-
c + U = 0.509c + 0.50
ment protocols are selected to measure the same
quantity and applied by different operators then
and the lower limit of the uncertainty from
the uncertainty increases even further (Argyraki
equation (7) is given by:
et al. 1995). This suggests that a more rigorous
estimate of the uncertainty for the GLT mea-
c - U = c (1 - 12.1%/100)
surements would require the use of similar inter-
(1 - 54.6/100)- 0.16 + 0.34 organization trials with both multiple users of
c+ U=cx0.879x0.454 -0.16+0.34 one technique, the comparison of a number of
c + U=0.399c+0.18 probes in the same borehole, and the use of
closely spaced duplicate boreholes to investigate
For an Fe concentration of 1 mass% the lateral sampling errors, and small scale geo-
uncertainty would range from 0.579 to 1.009 chemical variability.
mass% Fe, and for 5 mass% Fe from 2.175 to A further limitation of the method reported
3.045 mass% Fe. The range of these uncertain- here is that it assumes that the ICP-AES analysis
ties are large, but could be allowed for in provides an 'accepted reference value' as re-
geochemical interpretation of the concentration quired by ISO for the detection of bias (ISO
values. 1993b). Although the ICP-AES was calibrated
Although not relevant to the calculation of using ten certified reference materials this does
uncertainty, it is interesting to speculate on the not ensure zero bias. This partially because the
cause of this systematic error. Two possible recommended values for these reference materi-
causes have been suggested for the large negative als also have specified uncertainties, but more
rotational bias for Fe determinations using the especially due to the problems of matching the
G L T (Sams et al. 5995). Firstly, it is possible sample volume. A better solution to the second
62 M. H. RAMSEY ET AL.

aspect would be to establish one borehole as a where a crude estimate will be sufficient and
'reference sampling target', against which new others where more reliable but more expensive
analytical probes and new operators could methods will be justified. Further clarification of
measure their systematic error (Thompson & these ideas awaits more case studies of the
Ramsey, 1995). The 'accepted reference value' of application of techniques like the G L T that
the elemental concentrations at specified depths include estimates of uncertainty, and an evalua-
in this borehole would have to be established by tion of its effects on the geochemical interpreta-
inter-organization sampling trials similar to tion of the information.
those described above, with a wide variety of
analytical techniques. Conclusions
One extra complication of the measurement
system used in the G L T is that the bias in any 1. Methods are available to estimate the
one element will be propagated through to the uncertainty of measurements made with
other elements by the oxide closure. Such the GLT.
multivariate effects in measurement uncertainty 2. The random component of the uncertainty
have not been investigated and may cause subtle can be estimated at a basic level, from a
and unforeseen effects on geochemical interpre- replicate set of measurements from the
tation. same borehole.
3. The random component of the uncertainty
Acceptable limits for uncertainty can be estimated from the chemical analy-
sis of core material by a technique such as
Once realistic values become available for ICP-AES, which can be linked directly to
measurement uncertainty then there will be a 'accepted reference values' of concentration
need to derive acceptable limits for the un- as required by ISO. The uncertainties in the
certainty. This is a separate aspect that relates analyses by the method used for compar-
the G L T measurements to the concept of'fitness ison (in this case ICP-AES) need to be
for purpose'. To take an extreme case, for the assessed, but for the ICP-AES method they
vertical correlation of statigrapaphy between were much smaller than differences be-
two boreholes it could be argued that systematic tween the measurement techniques. There
errors in the measurements are irrelevant. The are also problems with this approach in
depth of the coal horizons in the Northumbrian allowing for the effects of different volumes
boreholes, for example, could be correlated of rock sampled.
between boreholes even if the iron concentration 4. Progressively more realistic estimates of
is biased by --54.6%. It is only the random uncertainty would require the use of
component of the uncertainty that could mask different operators, on different occasions,
the position of such a feature, and as such even with different probes in inter-organi-
therefore could be specified in the fitness-for- zation trials.
purpose specification. The coal bands (shown at 5. There is a financial need to derive accep-
150 and 154.3m in Fig. 1) can be detected from table levels of uncertainty for particular
the sulphur concentrations measured by the applications, but further case studies re-
GLT, even though the random component of porting uncertainties must be examined
the uncertainty was estimated as 71%. This before this will be feasible.
shows that the limits for uncertainty need to be
related to the geochemical variance (Ramsey et We would like to thank D. Filmer who performed the
al., 1992). The requirements for uncertainty are ICP-AES analyses. We would also like to thank Agip,
very different if the elemental concentration Amoco, BP, Elf, Mobil, NERC, Schlumberger and
Statoil who funded the Imperial College borehole test
estimates were to be used to infer the mineralogy
site. The third author also acknowledges assistance
of a sample. In this case a bias o f - - 5 0 % on one given by Petronas Research and Scientific Services to
element could clearly have a major impact on enable this work to be completed.
the minerals inferred to be present and their
calculated proportion in the rock.
Because the estimation of uncertainty is more References
expensive than the simple estimation of elemen- ANALYTICAL METHODS COMMITTEE 1995. Uncertainty
tal concentration, cost-benefit analysis will need of measurements: implications of its use in
to be applied to identify the benefits that the analytical science. Analyst, 120, 2303-2308.
uncertainty information can bring. Further- ARGYRAK1,A., RAMSEY,M. H. & THOMPSON,M. 1995.
more, if there is a choice of several methods to Proficiency testing in Sampling: Pilot study on
estimate uncertainty, there will be some cases Contaminated Land. Analyst, 120, 2799-2804.
ESTIMATION OF MEASUREMENT UNCERTAINTY 63

EUROCHEM 1995. Quantifying uncertainty in analytical & HALE, M. 1992. Objective evaluation
measurement. Eurochem Secretariat, Teddington, of precision requirement for geochemical analysis
UK. using robust analysis of variance. Journal of
GARRETT R. G. 1969. The determination of sampling Geochemical Exploration, 44, 23 36.
and analytical errors in exploration geochemistry. , ARGYRAKI,A. & THOMPSON,M. 1995a. On the
Economic Geology, 64, 568 569. collaborative trial in sampling. Analyst, 120,
GOVINDARAJU, K. 1994. Geostandards Newsletter, 2309-2317.
Special Issue, July 1994. & 1995b. Estimation of
GRAU, J. A., SCHWEITZER,J. S. & HERTZOG,R. C. 1990. sampling bias between different sampling proto-
Statistical uncertainties of elemental concentra- cols on contaminated land. Analyst, 120, 1353-
tions extracted from neutron-induced gamma-ray 1356.
measurements. IEEE Transactions on Nuclear , POTTS, P. J., WEBB, P. C., WATKINS, P.,
Science, 7, 2175-2178. WATSON, J. S. & COLES, B. J. 1995c. An objective
HERTZOG, R., COLSON, L., SEEMAN, B., O'BRIEN, M., assessment of analytical method precision: com-
SCOTT, H., MCKEON, D., WRA1GHT, P., GRAU, J., parison of ICP-AES and XRF for the analysis of
ELLIS, D., SCHWEITZER, J. & HERRON, M. 1987. silicate rocks. Chemical Geology, 124, 1 19.
Geochemical logging with spectrometry tools. SAMS, M. S., WATKINS,P. J. & RAMSEY, M. H. (1995).
Society of Petroleum Engineers, Annual Techni- Validation of a geochemical logging tool for in
cal Conference 1987, Dallas. SPE 16792. situ major element analysis in boreholes using
Iso 1993a. Guide to the expression of uncertainty in inductively coupled plasma atomic emission
measurement. ISO, Geneva. spectrometry. Analyst, 120, 1407-1413.
Iso 1993b. 3534-1:1993 (EIF) Statistics, Vocabulary THOMPSON, M. 1982. Regression methods in the
and Symbols - Part 1. Probability and General comparison of accuracy. Analyst, 107, 1169-1180.
Statistical Terms, ISO, Geneva. 1995. Uncertainty in an uncertain world.
RAMSEY, M. H. 1993. Sampling and analytical quality Analyst, 120, 117N-118N.
control (SAX) for improved error estimation in - - & HOWARTH, R. J. 1976. Duplicate analysis in
the measurement of heavy metals in the environ- geochemical practice. Analyst, 101, 690-698.
ment, using robust analysis of variance. Applied - - & RAMSEV,M. H. (1995). Quality concepts and
Geochemistry, 2, 149-153. practices applied to s a m p l i n g - a n exploratory
, THOMPSON, M. & BANERJEE, E. K. 1987. A study. Analyst, 120, 261-270.
realistic assessment of analytical data quality from WENDLANDr, R. F. & BHUYAN,K. 1990. Estimation of
inductively coupled plasma atomic emission mineralogy and lithology from geochemical log
spectrometry. Analytical Proceedings, 24, 260- measurements. American Association of Petroleum
265. 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 of the methodology has therefore been tested on
between the small-scale, high-resolution sedi- most sedimentary rock types, giving a full
mentological and stratigraphical features visible spectrum of typical data.
at outcrop and the large-scale data available The overall aim of this paper is to reproduce
from seismic sections. Simple techniques have at decimetre resolution gamma ray and density
been developed for producing natural gamma wireline log trends from measurements on out-
ray and density logs from measurements on crops and rock samples, thus improving strati-
outcrop exposures and rock samples to correlate graphic correlation between outcrops and
with wireline logs from boreholes. Emphasis has boreholes. The data and interpretation pre-
been placed on producing field logs which are at sented in this paper are part of a wider study
the same resolution and of similar character to on the sequence stratigraphical interpretation of
typical downhole wireline logs, rather than wireline log signatures from over 100 boreholes
reproducing the absolute values which might from the Upper Jurassic of the Wessex Basin.
be expected downhole. This approach thus Where available, biostratigraphical data have
concentrates on reproducing patterns of cycli- been used to provide a framework for wireline
city, together with general decreasing and correlations between boreholes.
increasing trends, which in turn can be inter-
preted in terms of cyclostratigraphy; for exam- Geological setting of exposures and
ple, transgression and regression, sequence
boreholes
stratigraphical and Milankovitch cycles.
The techniques are illustrated for the Upper The Wessex Basin is a Mesozoic extensional
Jurassic (Oxfordian, Kimmeridgian and Por- basin which is divided into a series of half-
tlandian stages) of the Wessex Basin, Southern graben, or graben-like sub-basins (Fig. 1). The
England. This interval is represented by a wide Upper Jurassic exposures and boreholes used in
range of sedimentary rock types ranging from this paper are from four of these sub-basins. The
deep-marine siliciclastics to shallow and non- exposures where field measurements and rock
marine siliciclastics and carbonates. The validity samples were taken are on the Dorset Coast

AHMADI,Z. M. & CoE, A. L. 1998. Methods for simulating natural gamma ray and density wireline 65
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
66 Z, M. AHMADI & A, L. COL

~ VALEOF PEWSEY 9 9 BletchingleyI ~ ~0


SUB-BASIN Collendean 9 "~
~'~'-"""-.,~ Farm 1 9 Detention 1 ~ /..._-f
~ Ashdown 1 7"---.~ (

"~-~"--,,.,.~ WEALD SUB-BASIN ~ -


ST DORSET ~ _ -

"l~ ~ ~ CENTRAL CHANNEL '~.,~ c '7 '~


IV SUB-BASIN i ~'-" L,-~

-s,,) WESSEX BASIN ~ ~ J

t 7o s*o 9L0 SY0'0sz lb

90 ' 10 km ' DORSET ~-7~- 9>


Hamcliff Blackhead ~/:-~~ ~,~
[ ~ Redcliff/ .
ran Point ( ' 3-11 / 89
Y ~ I S" 2raRingstead Bay ~v

!-8o ,' ~"~-'~-~ "" 9 ' [ 98/11 4 80-


/ .'~ W E
. .Y. M
. .f ). I .I T
. .I -. - I ~"~- , encomoe
9 l .~~- -
/\ ( / - - Q B~YJ
t . . -. > . ". . ~ ' ~ SWANAGE
& Kimmeridge Bay St.Alban's
to Head
-70 Chapmans Pool 70-
70 80 90 Sy00LSZ 10
I I I 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 Corallian Group, a complex succession of


the edge of the Central Channel Sub-basin. shallow-marine siliciclastics and carbonates
Boreholes Encombe 1 (SY 9446 7785), 98/11-1 which show marked lateral and vertical varia-
(SZ 1187 8386) and 98/11-3 (SZ 1329 8459) are tion. The Kimmeridgian Stage (sensu anglico) of
on the up-thrown northern edge of the Central Dorset is represented by interbedded organic-
Channel Sub-basin, and borehole 98/11-4 (SZ rich and organic-poor mudstones, with a few
1187 8084) is in the Central Channel Sub-basin. thin beds of fine-grained sandstone near the base
Winterbourne Kingston 1 (SY 8470 9796) bore- and the top. These mudstones and sandstones
hole is in the Dorset Sub-basin and Marchwood comprise the Kimmeridge Clay Formation, and
1 (SU 3991 1118) is in the Mere Sub-basin. All of are generally of wide lateral extent. They can be
the other boreholes mentioned are from the correlated across the Wessex Basin and into the
Weald Sub-basin, a moderately deep graben in Wash area and Humberside using outcrops,
the eastern part of the Wessex Basin (Fig. 1). wireline logs and borehole cores (Gallois & Cox
The Oxfordian Stage is represented in the 1974; Cox & Gallois 1981; Penn et al. 1986;
lower part by mudstones of the Upper Oxford Melnyk et al. 1994, 1995). The Portlandian
Clay Formation and in the upper part by the Stage is represented by marine silty and clay-rich
CORRELATION OF WIRELINE LOGS WITH OUTCROP 67

dolomites deposited in a moderate water-depth Dorset using a Scintrex Scintillation Counter


(Portland Sand Formation) overlain by a (SCC) spectrometer. There are two problems
shallow and non-marine carbonate ramp system with the work of Talwar et al. (1992). Firstly,
which comprises the Portland Stone Formation they appear to have used an exceedingly short
and Lulworth Beds (Coe 1996). 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 &
Field and laboratory methods for reprodu- Mose 1987; Parkinson 1996). Secondly, their
cing wireline log trends correlation with two boreholes from the North
Dorset and Wiltshire area show little similarity
g a m m a ray logging because the lower two-thirds of the Oxfordian
strata examined in the boreholes is older than
Two types of wireline gamma ray sondes exist, any of the rocks which they illustrate from Bran
the conventional one which records the total Point, and thirdly they did not take into account
natural radiation, and the spectral gamma ray any of the unconformities in the Oxfordian
sonde which separately records gamma rays succession (Coe 1992, 1995).
emitted from 4~ 232Th or 238U and their decay
products (Serra 1984). The main uses of gamma Gamma ray logging field procedure. Two porta-
ray logs are: ble gamma ray spectrometers have been used
and compared in the work reported here: a
(i) as an indicator of lithology; geoMetrics GR310 (manufactured 1980) and an
(ii) to correlate the wireline signatures between Exploranium GR320 (manufactured 1996). Both
boreholes; tools use thallium-doped sodium iodide detector
(iii) to correlate separate wireline runs within crystals. The Exploranium GR320 was cali-
one borehole. brated in Toronto by Exploranium Ltd (Cana-
da) and the geoMetrics GR310 was calibrated
The fact that the gamma ray tool is run in all on the calibration pads at the British Geological
boreholes makes it the key wireline tool for any Survey, Keyworth. A value for background
attempt to make correlations between outcrop radiation was measured 2 km offshore from
and the subsurface. Swanage, Dorset for each tool at the same time
Field gamma ray logs can be constructed (Fig. 1). Detailed explanation of the calibration
using hand-held portable gamma ray spectro- of portable gamma ray spectrometers is pro-
meters, which were originally developed and vided by Lovborg (1984) and Lovborg & Mose
used for uranium ore exploration (Adams & (1987).
Gasperini 1970). Following the lead of Etten- The geoMetrics GR310 provides separate
sohn et al. (1979), total gamma ray logs have measurements of either total gamma ray count,
subsequently been used for surface to subsurface or diagnostic gamma radiation for either K, or
correlation of sedimentary strata (Chamberlain U, or Th, and only allows count times of 1, 10,
1984; Cowan & Myers 1988; Slatt et al. 1992; 100 and 1000s to be chosen. Source, detector
Van Buchem et al. 1992). More recently, and recorder are all housed in one unit
portable gamma ray spectrometers have also 9 cm x 18 cm x 28 cm, weighing 3.4 kg.
been used to study the distribution of K, U and There are several advantages of the Explor-
Th in sedimentary rocks, and as a tool for anium GR320 for this type of stratigraphical
stratigraphical correlation between rock expo- study. Total counts and counts in the K, U and
sures (Dypvik & Eriksen 1983; Myers & Bristow Th fields are all recorded during one counting
1989; Davies & Elliott 1996; Hesselbo 1996; period the length of which can be set by the user
Parkinson 1996; Bessa & Hesselbo 1997). anywhere in the range 1 to 9999 s. The instru-
Previous spectral gamma ray studies on the ment carries out automatic gain stabilization,
Upper Kimmeridge Clay Formation in Dorset unlike the geoMetrics GR310 which has to be
have been completed by Myers (1987) and calibrated by the user. Automatic gain stabiliza-
Myers & Wignall (1987), who took spectral tion is important because portable spectrometers
gamma ray measurements using an Explora- are prone to tool drift due to changes in
nium GR256 on the wave-cut platforms. They temperature and humidity. The fact that the
utilized these data for a sedimentological and Exploranium GR320 stabilizes itself at regular
stratigraphical interpretation of organic-rich intervals saves time and reduces the risk of
mudstones. Talwar et al. (1992) completed a errors due to incorrect manual stabilization. The
study of the gamma ray spectrometry of the inbuilt computer chip allows the spectra to be
Corallian Beds (Oxfordian) at Bran Point, displayed and the amount of K, U and Th to be
68 Z.M. AHMADI & A. L. COE

the sedimentary rocks in this study. Rocks with


a) ~ Mass of effective
lower density and the same amount of natural
sample = 49 kg
assuming a density radiation would result in a slightly larger
of 2.8 g/cm3 effective sampling region. The most precise
absolute values for a particular bed of greater
than about 14cm in thickness are obtained by
)ept~ =
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
Diameter = 84 c)C
m. ~" underlying bed or beds. The aim of this study,
b) however, was to compare the general trends of
field gamma ray logs with wireline data. There-
fore 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.
'
I
r Readings were only taken where the tool could
Cliff face Borehole be used at least 1 m above the base of the cliff,
Fig, 2. Sampled volume for a portable gamma ray thus avoiding errors due to gamma ray con-
spectrometer compared to a wireline gamma ray tribution from rocks on the beach.
sonde. (a) Dimensions of the sampled volume for a Count times of 100s for the geoMetrics
portable spectrometer (modified from Lovborg et al. GR310 and 200 s for the Exploranium GR320
1971). (b) Typical orientation and position of the were used in this study. This resulted in
sampled volume for the portable gamma ray spectro- theoretical tool precision errors of < 2.5% and
meter as used in this study. (c) Spherical sampled < 1.5%, respectively, for the total count reading.
volume for a wireline gamma ray sonde in a borehole.
Parkinson (1996) showed that, in practice,
This depends on the speed at which the tool is drawn
up the hole, as well as the density of the rocks, but departures of measurement geometry from a
typically has a radius of 30 cm (Rider 1991). The true plane far outweigh instrument precision as a
sampled volume tends to a more ellipsoidal shape source of experimental error. In this study, it
when the tool is drawn up the borehole faster. 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
calculated directly. The only disadvantage to probably due to slight lithological variations as
this instrument compared with the geoMetrics well as differences in the volume of the effective
GR310 is that it is bulkier and heavier. This sample size due to small undulations in the cliff
spectrometer comprises two parts, a detector face. A longer count time was used for the
(11.4x39.4cm) and a recording/processing unit Exploranium GR320 because spectral data were
(24 x 10 x 25 cm) which have a combined weight also recorded. Radioactive decay of natural
of 8.4 kg. elements is a r a n d o m process, so shorter
The effective sampling region of portable sampling periods give a greater statistical error.
gamma ray spectrometers is shown in Fig. 2a. Specifically, the percentage statistical error
The dimensions in the figure are only approx- varies with the number of counts collected: the
imate because the density value used by Lovbor~ higher the count, the more accurate the mea-
et al. (1971) to calculate them was 2.8gcm-- surement. For typical needs, 1000 counts (3%
which is about 0.3-0.5 gcm 3 higher than most of 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 71

Operating Manual; Lovborg 1984). A longer representative of the majority of the beds in the
count time had to be used for the Exploranium Upper Oxford Clay F o r m a t i o n , Corallian
GR320 because it takes longer to record Group and Upper Kimmeridge Clay Formation
sufficient gamma ray counts in the K, Th and were collected for density analysis. These
U windows than it does for the total gamma ray amounted to 116 samples over 90m of the
measurement. The spectral data recorded with Upper Oxford Clay Formation and Corallian
the Exploranium GR320 are not discussed Group (Fig. 5) and 260 samples over 280m of
further in this paper. the Kimmeridge Clay Formation (Fig. 6). All
The measurement procedure used for both the samples were dried in an oven at a
spectrometers was to take a reading once in temperature of less than 35~ prior to the
every bed of less than 50 cm thick and every 30- measurements being taken. A Ruska Universal
50cm in beds greater than 50cm thick. The Porometer (model 1051-801) was then used to
geoMetrics GR310 was used to record total measure the volume and grain density of the
gamma ray readings throughout the best Upper samples. The density of each sample was then
Jurassic exposures in Dorset, resulting in 1124 calculated using a single typical fluid density
total gamma ray readings with an average value of 1.06gcm -3 for pore fluids present
sample interval of 45cm over 503m (Fig. 3). within Upper Jurassic rocks of the Wessex
Part of the Kimmeridge Clay Formation was Basin; this actual value was recorded at Palmers
selected to compare the results from the two Wood 3 borehole (TQ 3655 5255) in the Weald
spectrometers. Full spectral gamma ray data Basin (pers. comm. P. Rowe).
were thus recorded with the Exploranium The raw density curves on Figs 5 and 6 show
GR320 at 824 sample points over 251 m of the the density values calculated from the actual
Kimmeridge Clay Formation (average sample samples measured. To obtain the box curve, two
interval 30cm; Fig. 4). The average sampling further procedures were applied. Firstly, beds
distance of 30-45 cm is within the limits of the from which no samples had been obtained were
effective sampled volume for each spectrometer assigned an average density typical for that
(84cm; Fig. 2; Lovborg et al. 1971) and each particular lithology, calculated from the mea-
consecutive reading overlaps the previous read- sured samples collected nearby. Secondly, the
ing resulting in a moving average, thus making it same density value was assigned to the whole
comparable with the wireline gamma ray tool as thickness of the bed. It was noted that the
it is pulled slowly up the borehole. sandstones of the Nothe Grit Formation and the
Bencliff Grit Member (top of the Redcliff
Density logging Formation) had lower density values than those
seen on the wireline density logs. This was
Wireline density logs record the bulk density of interpreted to be due to higher porosities of
rocks, by emitting gamma rays into the forma- these rocks at outcrop than in the subsurface,
tion and recording the number of back-scattered resulting from dissolution of calcite cement. The
gamma rays at a fixed distance from the source. density values of these beds were therefore
The bulk density is a function of the density of corrected as follows: their average porosity in
the matrix and the density of the fluids in the the subsurface was estimated by plotting typical
pore space. Therefore any attempt to construct a density and sonic values on porosity evaluation
field density log with the same character and log interpretation charts (Atlas Wireline Services
resolution as the wireline density log has to take 1985; Schlumberger 1994). The additional,
into account the density of the matrix and the secondary dissolution porosity that was calcu-
density of the pore fluid. The vertical resolution lated to be present in the rock samples was
for older single-detector tools is 40 cm and for multiplied by the difference between the density
more modern two-detector tools is 25 cm (Serra of calcite and the pore fluid and added to the
1984). calculated total density for the samples. The box
curve was then filtered using the Atlas Wireline
Density logging laboratory procedure. Fresh rock Services field acquisition filter (Atlas Wireline
samples of smaller than 3.1 cmx3.1 cmx3.7cm 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*-~

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

~ 0,.0 ~ ~

~9 ~,. ~ ~.~

.,..~ ~ r~

~- . N u

~ ~ ~ 0 0
~ o ~ "~

~..~ ~ " 0 ~ ~ 0

~ ~-'~ ~ ~

o p?,~,.0 N [-" ~
~. .~ =~ ~,~

.~~=~
~ ~ o
~ ~ ~~
CORRELATION OF WIRELINE LOGS WITH OUTCROP 73

~'~ ~ c~ 0.~ r'"

0~, +-~ . ~ ~.., r -~

..~

o "~ o

"~ ~,~

U~g~~
-~ ~-~-
~ .~~
~'~ ~ ~ o

~ ~ ~

..~ ~t" 9

~ ~ ' ~

o= '~

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

o== r ~ "~ ~ ~
74 Z. M. AHMADI & A. L. COE

sian-weighted, moving-average filter. The total Kimmeridge Clay Formation is 0.7 (Fig. 4) .
filter length used was 1.1 m. This results in a This was calculated using Corpac, a signal
filtered field density curve with similar character correlation computer program (Globex Consult-
and resolution to a wireline density log (Figs 5 ing Services, Ltd 1992) which is based on a
and 6). simple mathematical inverse method to correlate
two time series (in this case depth series)
described by Martinson et al. (1982). The reason
Surface to subsurface correlation the correlation coefficient is not higher is because
the Exploranium GR320 log has higher resolu-
Upper Jurassic composite field gamma ray tion, and because of the gaps in the data. Higher
log correlation coefficients are obtained if the two
logs are correlated over shorter intervals which
The field gamma ray data from nine different contain no gaps in the data.
locations along the Dorset coast were combined
to produce a composite field gamma ray log for Correlation of field and wireline gamma ray
the Upper Jurassic strata of Dorset (Fig. 3).
Comparison with the wireline gamma ray data logs
from borehole 98/11-4, which are plotted at the Upper Oxford Clay and Corallian Beds. The field
same scale, show that the same general trends gamma ray log shows, from the base, an overall
and wireline log patterns are present in both sets upwards decreasing and then increasing trend in
of data throughout the Upper Jurassic interval. the gamma ray values, as do the logs in
Clearly distinguishable in both log signatures are boreholes 98/11-4 and 98/11-3, reflecting the
the overall trends of decreasing and increasing change in lithology from mudstones to sand-
response which are interpreted as representing stones and limestones, and then back to
long-term facies changes controlled by relative mudstones and iron-rich sandstones (Fig. 7).
changes in sea-level (Coe 1992). For instance, The Nothe Grit Formation is a better defined
the overall upwards decrease in gamma ray gamma ray low in boreholes 98/11-4 and 98/11-3
values for the pallasioides Zone of the Kimmer- than in the field gamma ray log, probably
idgian to the anguiformis Zone in the Portlan- because the sands are cleaner in the boreholes
dian reflects the change from marine mudstones and the clays of the overlying Redcliff Forma-
to carbonates interpreted as a long-term low- tion contain a high percentage of carbonate in
ering of relative sea-level (Coe 1992, 1996). The the outcrop section. Three prominent gamma
one notable difference is that the Lower Kim- ray peaks in the Osmington Oolite Formation
meridge Clay (baylei to autissiodorensis zones) is can be seen on both the field gamma ray log and
thicker in borehole 98/11-4 than in the outcrop in 98/11-4 (Fig. 7). Over a wider geographical
section. This is due to the fact that the area the Corallian Beds are lithologically very
measurements for the Lower Kimmeridge Clay variable, being comprised of shallow-marine
were made on exposures situated on the footwall sandstones and limestones. Sequence stratigra-
of the Central Channel Sub-basin, where the phical interpretation of the wireline logs using
succession is apparently complete but thinner. the number and character of the cycles does
permit a correlation to be made across the
Comparison of the geoMetrics GR310 and Wessex Basin; however, the lateral lithological
the Exploranium GR320 variability makes correlation based purely on
the wireline log character difficult.
Figure 4 shows a comparison of the total gamma
ray measurements taken with the two spectro- Kimmeridge Clay Formation. The similarity
meters over part of the Kimmeridge Clay between the field gamma ray logs produced by
Formation. The decreasing and increasing the two different spectrometers and the wireline
trends, amplitude of variation, and the shape gamma ray log from Encombe 1 borehole
of the peaks and troughs correlate very well. The (approximately 1 km inland from the outcrops)
main difference between the two signatures is the is shown in Fig. 8. The data acquired using the
higher resolution of the Exploranium GR320 smaller sample interval with the Exploranium
log, which results from the 30 cm average sample GR320 spectrometer produces a higher resolu-
interval compared to a 45 cm average sample tion curve, despite the fact that the sampling
interval for the geoMetrics GR310. interval is about one third of the effective
The correlation coefficient between the two sampling diameter of the tool (Fig. 2). Using
field gamma ray logs over 245 m of the 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 I

Oo,0

gr

~t'q

qgoo
oo

,9, , . ~ o o

~r/2

O,.~

~~

. ,.,,~
76 Z.M. AHMADI & A. L. COE

9 O

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

"0 0

0,--
. ,,...~

0 ~

~ .

g~

,~ o

~ 0
o " N
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

~.=_ 2

O ~ .,..-,

O ~
78 Z.M. AHMADI & A. L. COE

coefficients described above, the correlation and to a certain extent with borehole 98/11-1.
coefficient for the Exploranium GR320 field The Winterbourne Kingston 1 borehole density
gamma ray log and the wireline log from the log is more difficult to correlate in the lower part
Encombe 1 borehole is 0.92, but it is only 0.82 due to the lack of variation on the large scale.
for the geoMetrics GR310 field gamma ray log Fig. 5 also shows the marked lateral and vertical
and the wireline log. variation of the Oxfordian strata between the
Figure 9 shows the similarity between the wireline logs of 98/11-1, Marchwood 1 and
geoMetrics GR310 field gamma ray log and Winterbourne Kingston 1. One notable example
wireline gamma ray logs from boreholes up to of this is the differences seen between the three
170 km away (Fig. 1). There are several wells for the density of the Nothe Grit Forma-
particularly prominent features, including the tion (or its equivalent) and the Osmington Oolite
two gamma ray lows with a low amplitude of Formation.
variation seen in the Collendean Farm (TQ 2480
4429) and Ashdown 1 (TQ 5005 3035) boreholes Kimmeridge Clay Formation. Figure 6 shows the
in the hudlestoni and wheatleyensis zones, which comparison between the processed field density
are often referred to as the 'Kimmeridge lime- log for the Dorset coast (filtered density log of
stones' (Hancock & Mithen 1987). At outcrop, Fig. 6) against the nearby Encombe 1 borehole,
these two gamma ray lows with a low amplitude and the Bletchingley 1 (TQ 3622 4772) and
of variation are prominent thick homogeneous Detention 1 (TQ 7478 4020) boreholes in the
calcareous mudstone units (middle and upper Weald Basin. The general trends and the
part of bed 40 and the lower part of bed 44; Fig. character of all of these logs is remarkably
4). The two gamma ray lows in Collendean similar. The four high peaks which straddle the
Farm 1 and Ashdown 1 (Figs 1 and 9) are pectinatus to hudlestoni zonal boundary in both
probably more enhanced than those in 98/11-4 the outcrop density log and Encombe 1 log
and the field gamma ray log because the represent more carbonate-rich cementstone
sediments have an even higher calcium carbo- beds. The distinctive increase in density seen in
nate content. A higher quartz sand content is all the logs at the top of the lower third of the
discounted because the lithology over the same hudlestoni Zone represents at outcrop a change
intervals in the nearby Warlingham borehole from interbedded organic-rich and organic-poor
(TQ 3476 5719) comprises argillaceous lime- mudstones to a thick calcareous mudstone (Coe
stones and calcareous mudstones (Worssam et 1992). Similar lithological changes are inter-
al. 1971). Prominent gamma ray lows on the preted to occur in the borehole sections.
field gamma ray log like those in the middle of
autissiodorensis Zone, at the base of elegans Conclusions
Zone and near the top of scitilus Zone, are
carbonate-rich cemented horizons. Similar sharp (1) The geoMetrics GR310 and Exploranium
gamma ray lows in 98/11-4 and Collendean GR320 gamma ray spectrometers can both
Farm 1 probably also relate to calcareous be used to produce field gamma ray logs
cemented horizons. which are comparable with borehole gamma
ray wireline logs. Whilst the newer Explor-
anium GR320 is more accurate and can be
Correlation of field and wireline density logs used to gather spectral data more quickly,
the older geoMetrics GR310 does produce
Upper Oxford Clay and Corallian Beds. Produc- excellent data with repeatable and compar-
tion of an outcrop density log over this interval able gamma ray trends. The most compar-
of mixed siliciclastics and carbonates is more able signal between hand-held spectrometers
problematic than that for the Kimmeridge Clay and wireline log tools is produced by using
Formation. Processing of the data in a similar the hand-held spectrometer perpendicular to
manner to that of the Kimmeridge Clay the bedding with a sample interval of 30 cm
Formation resulted in a filtered density curve or less.
with very little variation. However, further Field gamma ray logs produced for the
processing of the data to take into account Kimmeridge Clay Formation can be used to
dissolved carbonate cement at outcrop, as positively correlate, often down to the bed
described above, resulted in a curve which is (typically < l m ) but at least down to the
more similar to the borehole density logs. bed group scale (typically 10 in), with wire-
The general trends seen on the filtered density line gamma ray data from nearby boreholes.
log (Fig. 5) show a positive correlation with the Larger gamma ray features can also be
wireline density from Marchwood 1 borehole correlated with boreholes as far as 170km
CORRELATION OF WIRELINE LOGS WITH OUTCROP 79

away in the Weald Sub-basin. Field gamma References


ray and wireline gamma ray data for the
Oxfordian show similar trends but complex ADAMS, J. A. S. & GASPERINI, P. 1970. Gamma ray
local lithological heterogeneity may be mis- spectrometry of rocks, Elsevier, Holland.
leading. The concepts of sequence stratigra- ATLAS WIREL1NE SERVICES. 1985. Log Interpretation
phy (i.e. recognition of the metre to tens of Charts. Western Atlas International, Inc.
ATLAS WIRELINE SERVICES. 1992. WDS advanced log
metre scale cycles) considerably aid in
evaluation - documentation. Western Atlas Inter-
making the correlation. This is because the national, Inc.
interpretation relies on the recognition of BESSA, J. L. & HESSELBO, S. P. 1997. Gamma ray
wireline log trends rather than correlating character and correlation of the Lower Lias, SW
similar lithologies, and requires identifica- Britain. Proceedings of the Geologists' Association,
tion of stratigraphic gaps and condensed 108, 113-129.
intervals. CHAMBERLAIN,A. K. 1984. Surface gamma ray logs: a
(2) Small rock samples from outcrop can be correlation tool for frontier areas. American
used to produce a field density log. Some Association of Petroleum Geologists Bulletin, 68,
1040-1043.
processing of the data is required to produce
COE, A. L. 1992. Unconformities within the Upper
a signal which is directly comparable with Jurassic of the Wessex Basin, Southern England,
the wireline tool. The method could easily DPhil Thesis, University of Oxford.
be applied to small rock samples from core 1995. A comparison of the Oxfordian succes-
or washed cuttings. sions of Dorset, Oxfordshire, and Yorkshire. In:
Excellent results comparable with the TAYLOR, P. D. (ed.) Field Geology of the British
wireline signature were obtained for a thick Jurassic. Geological Society, London, 151-172.
succession of interbedded organic-rich and 1996. Unconformities within the Portlandian
organic-poor mudstones and cementstones Stage of the Wessex Basin and their sequence-
stratigraphical significance. In: HESSELBO,S. P. &
(Kimmeridge Clay Formation). Where the
PARKINSON, D. N. (eds) Sequence Stratigraphy in
lithology varies more widely and shallow- British Geology, Geological Society Special Pub-
marine sandstones and limestones (e.g. lications No. 103, 109 143.
Corallian Beds) are present, it is necessary COWAN, D. R. & MYERS, K. T. 1988. Surface gamma
to take into account the differences in ray logs: A correlation tool for frontier areas:
porosity between the borehole and outcrop Discussion. American Association of Petroleum
section and apply a further correction factor Geologists Bulletin, 72, 634-636.
to the outcrop density data. Cox, B. M. & GALLOIS,R. W. 1981. The stratigraphy
(3) The measurement and processing of the of the Kimmeridge Clay of the Dorset type area
and its correlation with some other Kimmeridgian
physical characteristics of rock exposures
sequences. Report of the Institute of Geological
to produce a wireline log signature is Sciences, 80/4.
invaluable in the understanding of boreholes DAVIES, S. J. & ELLIOTT,T. 1996. Spectral gamma ray
where core is not available. The data can be characterisation of high resolution sequence
readily used to supplement and enhance stratigraphy: examples from Upper Carbonifer-
conventional litho- and bio- stratigraphical ous fluvio~leltaic systems, County Clare, Ireland.
correlations between boreholes, and bore- In: HOWELL, J. A. & AITKEN, J. F. (eds) High
holes and outcrop. Resolution Sequence Stratigraphy." innovations and
applications, Geological Society Special Publica-
tions No. 104, 25-35.
Z. Ahmadi was supported by a Durham University DYPVIK, H. & ERIKSEN, D. O. 1983. Natural radio-
Research Studentship and an AAPG-PESGB Grants- activity of clastic sediments and the contributions
in-Aid grant for field and laboratory studies. We thank of U, Th and K. Journal of Petroleum Geology, 5,
Charlotte Martin and Toby Harrold for their assis- 4094 16.
tance in the field, and Brian Turner for the loan of his ETTENSOHN, F. R., FULTON, L. P. & KEPFERLE, R. C.
geoMetrics GR310 gamma ray spectrometer. The 1979. Use of scintillometer and gamma ray logs
Exploranium GR320 was purchased from a grant for correlation and stratigraphy in homogeneous
awarded to A. L. Coe from the Open University black shales. Geological Society of America
Research Development Fund. M. Oates of British Gas Bulletin, part II, 90, 828-849.
provided the wireline and biostratigraphical data from GALLOIS,R. W. 8z Cox, B. M. 1974. Stratigraphy of the
boreholes 98/11-1, 98/11-3 and 98/11-4, and H. Bailey Upper Kimmeridge Clay of the Wash area.
of the British Geological Survey provided the wireline Bulletin of Geological Survey of Great Britain,
data for the onshore boreholes in the Wessex Basin. 47, 1-16.
We would particularly like to thank N. Goulty for his HANCOCK,F. R. P. & MITHEN,D. P. 1987. The geology
constructive comments during the preparation of this of the Humbly Grove Oilfield, Hampshire, UK.
paper and two anonymous referees are thanked for In: BROOKS, J. & GLENNIE, K. (eds) Petroleum
reviewing this paper. Geology of North West Europe, Graham & Trot-
80 Z . M . AHMADI & A. L. COL

man, 161-170. using gamma ray spectrometry and palaeoecol-


HESSELBO, S. P. 1996. Spectral gamma ray logs in ogy: examples from the Kimmeridge Clay of
relation to clay mineralogy and sequence strati- Dorset and the Jet Rock of Yorkshire. In:
graphy, Cenozoic of the Atlantic Margin, offshore LE~GETT, J. K. & ZUFFA, G. G. (eds) Marine
New Jersey. In: MOUNTAIN, G. S, MILLER, K. G, Clastic Sedimentology - concepts and case studies,
BLUM, P., POAG, C. W. & TWlCHELL, D. C. (eds) Graham & Trotman, London, 172-189.
Proceedings of the Ocean Drilling Program Scien- PARKINSON,D. N. 1996. Gamma ray spectrometry as a
tific Results, 150. tool for stratigraphical interpretation: examples
LOVBORG, L. 1984. The calibration of portable and from the western European Lower Jurassic. In:
airborne gamma ray spectrometers - theoo', HESSELBO, S. P. & PARKINSON, O. N. (eds)
problems and facilities, Report Riso-M-2456, Riso Sequence Stratigraphy in British Geology, Geolo-
National Laboratory, Denmark. gical Society Special Publications, 103, 231-255.
& MOSE, E. 1987. Counting statistics in PENN, I. E., Cox, B. M. & GALLOIS, R. W. 1986.
radioelement assaying with a portable spectro- Towards precision in stratigraphy: geophysical
meter. Geophysics, 52, 555-563. log correlation of Upper Jurassic (including
, WOLLENBERG,H., SORENSEN,P. & HANSEN, J. Callovian) strata of the Eastern England Shelf.
1971. Field determination of uranium and thor- Journal of the Geological Society, London, 143,
ium by gamma ray spectrometry, exemplified by 381-410.
measurements in the llimaussaq alkaline intru- RIDER, M. H. 1991. The geological interpretation of
sion, South Greenland. Economic Geology, 66, well logs. Whittles Publishing, Caithness.
368-384. SCHLUMBEROER 1994. Log Interpretation Charts.
MARTINSON,D. G., MENKE,W. & STOFFA,P. 1982. An Schlumberger Wireline & Testing, Houston,
inverse approach to signal correlation. Journal of Texas.
Geophysical Research, 87, 4807~4818. SERRA, O. 1984. Fundamentals qf well-log interpretation
MELNYK, D. H., SMITH, D. G. & AMIRI-GARROUSSl,K. 1. The acquisition of logging data. Developments
1994. Filtering and frequency mapping as tools in in Petroleum Science 15A. Elsevier, Holland.
subsurface cyclostratigraphy, with examples from SLATT, R. M., JORDAN. D. W., D'AGOSTINO, A. E. &
the Wessex Basin, UK. In: DE BOER, P. L. & GILLESPIE, R. H. 1992. Outcrop gamma ray
SMITH, D. G. (eds) Orbital Jorcing and cyclic logging to improve understanding of subsurface
sedimentary sequences, International Association well log correlation. In." HURST, A., GR1FFITHS, C.
of Sedimentologists, Special Publications 19, 35 M. & WORTHINGTON, P. F. (eds) Geological
46. Applications of Wireline Logs H, Geological
, ATHERSUCH,J., AINSWORTH,N. & BRITTON, P. Society Special Publications, 65, 3-19.
D. 1995. Measuring the dispersion of ostracod TALWAR, A. D., HENDERSON, A. S. & HART, M. B.
and foraminifera extinction events in the subsur- 1992. Simple gamma ray response of the Upper
face Kimmeridge Clay and Portland beds, Upper Jurassic from the Dorset coast - a preliminary
Jurassic, United Kingdom. In: MANN, K. O., investigation using the scintillometer profile tech-
LANE, H. R. & SCHOLLE,P. A., Graphic correla- nique. Proceedings of the Ussher Society, 8, 70-72.
tion, Society of Economic Paleontologists and VAN BUCHEM,F. S. P., MELNYK,D. H. & McCAvE, [.
Mineralogists, Special Publications, 53, 185-203. N. 1992. Chemical cyclicity and correlation of
MYERS, K. J. 1987. Onshore-outcrop gamma ray Lower Lias mudstones using gamma ray logs,
spectrometry as a tool in sedimentological studies. Yorkshire, UK. Journal of the Geological Society,
PhD thesis, University of London. London, 149, 991-1002.
- - & BRISTOW,C. S. 1989. Detailed sedimentology WmTTAKER, A. (ed.) 1985. Atlas of Onshore Sedimen-
and gamma ray log characteristics of a Namurian tary Basins in England and Wales: Post-Carboni-
deltaic succession II: gamma ray logging. In: ferous Tectonics and Stratigraphy. Blackie,
WHATELEY, M. K. G. & PICKERING,K. T. (eds), Glasgow.
Deltas." Sites and traps for jbssil fuels, Geological WORSSAM, B. C., IVIMEY-COOK, H. C. 1971. The
Society Special Publications, 41, 81-88. stratigraphy of the Geological Survey Borehole
& W1ONALL, P. B. 1987. Understanding at Warlingham, Surrey. Bulletin of the Geological
Jurassic organic-rich mudrocks new concepts 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 lithol- tion. The clay, carbonate, and quartz-feldspar-
ogy from common geophysical logs is hindered mica portions of this interpretation have been
by a lack of sensitivity coupled with nonunique presented previously (Herron & Herron 1996).
responses to the minerals that reside in sedimen- This paper provides a brief introduction to the
tary rocks. The interpretation of lithology for the new geochemical logging capabilities in both
purpose of wireline petrophysical evaluation or open and cased holes and a detailed examination
geological characterization primarily consists of of the new core-based interpretation.
estimating fractions of shale, sand, and carbo-
nate. Nuclear logs, either gamma ray, photo- Elemental concentration logs
electric factor, and/or a combination of neutron
and density are the most commonly used logs for The recently developed technique to estimate
lithology interpretation, A desire for improved elemental concentrations from a single, induced-
accuracy in Ethological description led to the neutron gamma ray spectrometer (Herron 1995)
introduction of several generations of nuclear is an adaptation of a geochemical oxides closure
spectroscopy logs. Recent developments in open model already employed in the computation of
and cased hole logging have made it possible to elemental concentrations from multiple nuclear
obtain accurate concentration logs for the sondes (Hertzog et al. 1987; Schweitzer et al.
elements silicon, calcium, iron, sulfur, titanium, 1988; Grau & Schweitzer 1989; Grau et al. 1989).
and gadolinium at relatively low cost and high The most significant modifications are:
logging speeds (Herron 1995).
A new lithological interpretation has been (1) the elimination of aluminium and potas-
developed to capitalize on these new logging sium as necessary inputs to the geochemical
capabilities. It is founded on an extensive closure model, thus considerably reducing
database of core chemistry and mineralogy. the number of wireline sondes necessary to
The new interpretation provides quantitative produce elemental concentrations of potas-
estimates of: total clay, which is the sum of all sium;
clay minerals; carbonate, which is the sum of (2) a change in the elemental associations of
calcite and dolomite; anhydrite, which is the sum iron.
of anhydrite plus gypsum; and quartz-feldspar-
mica (Q F-M), which is the remainder of the Figure 1 presents examples of elemental concen-
formation essentially constituting the sand frac- tration logs from the new processing using data

HERRON, M. M. & HERRON,S. L. 1998. Quantitative lithology: open and cased hole application 81
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
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
Silicon wt%
50 0 20
Calcium wt%
40 0 10
Iron + .t4AI ~%
20 0 10
Sulfur wt%
20 0 2
Titanium wt%
4 0
[} 20 40
Gadolinium ppm

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

x 104
1.01

r-.

.,....

7-
1.03

.v
O..
g~
p,.

1.05

][ ,
~,
w
|
1.0"/
0 50 0 20 40 0 10 20 0 10 20 0 2 4 0 20 40
Silicon wt% Calcium wt% Iron + 14AI wt% Sulfur wt% Titanium wt% Gadolinium ppm

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

from an open hole Elemental Capture Spectro- core samples are shown for comparison. Two
scopy (ECS; Mark of Schlumberger) sonde. This points should be made when examining the data.
is a nuclear spectroscopy device which uses a The first is that since the uncorrected prompt
standard AmBe source and a BGO detector. It is capture yield for iron contains gamma rays from
combinable and can log at up to 540 m hr -l (1800 both Fe and A1, the log Fe should be approxi-
fthr-~). Chemical concentrations measured on mately equal to Fe+0.14A1. Accordingly, the
QUANTITATIVE LITHOLOGY

'~176wo,:-. /1 . . . . . . . . . . . . . . . . . .

- / I
~ .......... r
'O01we"/ Well6 ~ o ' o ~ I Well 7 "o / Well i

~ I/:"
oL~ ." .

~176
w~176/ Well 10

0 100 200 0 100 200 0 100 200 0 100 200


Gamma Ray Gamma Ray Gamma Ray 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 tion coupled plasma mass spectrometry, for
Fe+0.14A1. The second point is that the log whole rock elemental concentrations of silicon
concentrations agree well with core data. (Si), aluminium (A1), iron (Fe), calcium (Ca),
A second example is provided from a cased magnesium (Mg), sodium (Na), potassium (K),
hole Reservoir Saturation Tool (RST; Mark of phosphorus (P), titanium (Ti), manganese (Mn)
Schlumberger) log acquired from a well in and chromium (Cr), expressed as oxides, plus
Venezuela (Fig. 2). This example is processed Loss on Ignition (LOI) representing total vola-
using new elemental standards to derive the far tiles, H20 +, H 2 0 , sulfur (S), organic carbon,
detector capture yields (Roscoe et al. 1995), and thorium (Th), uranium (U), gadolinium (Gd)
corrections are made for casing and a 3.8cm and boron (B). A synthetic core gamma ray
cement annulus. The results show good agree- (GR) computed from core chemistry using the
ment between log concentrations and the sparse gamma ray response is given by equation (1)
core data.
G R = 4 T h + 8 U + 16 K (1)
Core database
where Th and U concentrations are in ppm and
The development of the new quantitative lithol- K concentrations are in wt% (Ellis 1987).
ogy interpretation begins with a core database The mineralogy fraction was analysed using a
that contains chemistry and mineralogy mea- new Fourier Transform-Infrared (FT-IR) proce-
surements on over 400 core plug samples from dure which simultaneously analyses the mid-IR
numerous wells on four continents. The wells are and far-IR frequencies. The mid-IR procedure
diverse in age and geographic location, but all was described in Matteson & Herron (1993).
are predominantly sands and shaly sands. Since that time the number of mineral standards
To analyse the samples, rocks were crushed has been increased to 26 with approximately the
and split with a microsplitter into chemistry and same level of accuracy (better than +2 wt %).
mineralogy fractions. The chemistry fraction was The mineral standard set includes quartz, albite,
analysed at X-Ray Assay Laboratories using X- anorthite, K-feldspar, muscovite, biotite, kaoli-
Ray fluorescence, neutron activation and induc- nite, illite, smectite, chlorite, glauconite, calcite,
84 M. M. HERRON & S. L. HERRON

o 1~176

9 dP

%
e 9 1 4 90 ~ ~ 9 9 9 9 e"
10 20 0 5 0 2.5 5
Thorium ppm Uranium ppm Potassium wt%
loo

+t
_~ so
o

00

loo[
+/
10
t 9

Aluminum wt%
20 o 1
Titanium wt%
0
.,'.;
5
Gadolinium
.t
.
10

o 25 50 0 15 30 0 20 40
Silicon wt% Iron wt% 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, arago- porosity-free basis, as recently advocated by


nite, gypsum, anhydrite, hematite, barite and K a t a h a r a (1995). The relationship for core
opal. Total clay is the sum of kaolinite, illite, samples from 12 data sets is presented in Fig.
smectite, chlorite and glauconite. Although there 3. A line connecting the origin with 100% clay
are significant amounts of mica, another layered and 250 API is included for visual reference. As
silicate, they are not included in the total clay expected, gamma ray content generally increases
fraction. At high clay concentrations there is as clay content increases. However, there are a
sometimes interference between illite and mica number of characteristics in the clay-gamma ray
phases. plots that highlight the weaknesses inherent in
this approach; many of these have been recently
Exploring elemental relationships discussed by Bhuyan & Passey (1994) and Hurst
& Milodowski (1994).
The most complex aspect of the new lithology The first major feature is the large range of
interpretation is the computation of the clay slopes in the gamma ray versus clay plots which
mineral fraction. In the logging world, clay, or demonstrates the necessity for local calibration.
more often volume of shale, is most frequently For example, in Well 1, a linear trend predicts a
estimated from the gamma ray log. However, maximum gamma ray value of about 100 API
there are many type of clay minerals with widely for the pure clay end member, whereas Well 2
differing compositions and log responses, so would predict 500 API. For Well 12, a pure clay
shale estimates often carry large uncertainties. would have a gamma ray of only about 150 API.
The estimation is further degraded by the many In several wells, either the data or an extrapola-
non-clay minerals which contribute significantly tion of the data to zero clay indicate a near zero
to the gamma ray. minimum gamma ray, but Well 4 has a minimum
gamma ray of 30 API, and in Well 12 an
Gamma ray and clay extrapolation points to 70 API for minimum
gamma ray. The difference between evaluating
With the core database, it is possible to evaluate these plots and using only log data is that with
the relationship between total clay determined by core calibration the amount of clay is known,
FT-IR and the computed gamma ray on a and it is possible to accurately extrapolate to
QUANTITATIVE LITHOLOGY 85

~5o
~r1~176" i I.=. .
O0 e~ II
9

~ Jp

_r --
o;=~'.;'10"< " " 20 0
w
5 10 0 2.5
Thorium ppm Uranium ppm Potassium wt%
100,

} 50 .=" -o-'
I.,../)--:-.
06
100,
lb
Aluminum wt%
20 o i
T'aanium wt%
0
_.
5
Gadolinium
10

f
I
-
-% ~149 ~ O~ 9

0 25 50 0 15 30 0 20 40
Silicon wt% Iron wt% 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"
:- ~ 9 ~0
I.%',,
10 20 0 5 10 0 2.5 5
Thorium ppm Uranium ppm Potassium wt%

k,..'
100, ,
o
o
~5oI '~:~"""
ee 9 9
9 e~po
d.'t' t
~~' lb
Aluminumwt*/,
20 0 1
Titanium wt*/,
2 0 5
Gadolinium
10

100 ....
i .

9 00
u." "4. ~
0 25 50 0 15 30 0 20 40
Silicon wt% Iron wt% Calcium wt%

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

zero clay. With only log data, one must choose a comparison of clay with aluminium. In Well 3,
minimum and maximum gamma ray value aluminium displays a strong relationship with
without knowing the correspondence to real clay total clay. The remaining elements in this figure
concentrations, and the picture is further com- are some that can be obtained by prompt
plicated by porosity variations. For Well 11, the thermal neutron capture spectroscopy logging
minimum gamma ray value observed on the log devices. The two elements remaining in the
is about the same as the 50 API minimum second row of the figure are titanium and
computed for the core data. This value would gadolinium. These elements are commonly en-
normally be assigned to zero clay instead of the riched in shales, but they show only a loose
actual 25 wt% clay. Clearly, such a log correlation with total clay.
interpretation would severely under-estimate The third row holds the key to a new
the clay content in the well. technique for estimating clay. It begins with
The second dominant feature in Fig. 3 is the silicon, which is a major constituent of rock
scatter in the data, particularly in Wells 1-10. In forming minerals. Although silicon is commonly
these wells, even if the observed correlation associated with quartz, it is actually the second
between gamma ray and clay were known, the most abundant element after oxygen in both
scatter in the data would produce an uncertainty sandstones and shales. Because it is a major
of as much as +20 wt% clay or more. For Wells element, its abundance is not affected by trace
3, 5, 7 and 9, at levels of about 20% clay, minerals, and concentrations form a smooth
observed gamma ray values span almost the full continuum between high silicon sandstones and
range from clean sand to shale. The relative error medium silicon shales. For reference, quartz has
is particularly large in sands. 46.8 wt% silicon. The next element is iron, which
A third and less common feature is that some has numerous associations, including heavy
wells exhibit a small dynamic range in gamma minerals such as siderite, pyrite, hematite, and
ray while clay content varies considerably. This magnetite and the clay minerals illite, chlorite,
is notable in Well 12 which is a typical offshore glauconite and some smectites. High concentra-
Gulf of Mexico example. It is also true in Wells 4 tions of the heavy iron minerals can interrupt the
and 11. smooth relationship between silicon and clay
In spite of the problems outlined above, it content. The final element is calcium which is
would be possible to make good clay predictions mainly associated with the carbonate minerals
in Wells 2, 11 and 12 if detailed and accurate calcite and dolomite. The low calcium concen-
core data were available. Without such a trations indicate the absence of carbonate
calibration, it is doubtful that the picks for minerals in Well 3.
GRmax and GRmi n from the log data would The same type of comparison between total
match the core calibration parameters. clay and elemental concentrations is presented in
Fig. 5 for Well 5. For this well, none of the
Seeking an elemental alternative individual elements (Th, U and K) contributing
to natural gamma ray is any better correlated
One goal of this study is to identify an with clay than is total gamma ray. In contrast,
alternative, less subjective approach to determin- aluminium again shows a tight correlation with
ing 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
technology exists to measure elemental concen- two data points which clearly deviate from the
trations from natural radioactivity (Th, U and major trend. These two samples contain 13 and
K), neutron activation (AI), and capture gamma 38 wt% siderite (FeCO3) as reflected by the two
ray spectroscopy (Si, Ca, Fe, Ti, Gd and S). high iron points. As in Well 3, the near absence
Figure 4 shows a comparison of clay content of calcium reflects the absence of calcite and
with all available logging elements (except sulfur) dolomite.
for Well 3. The three components of natural A final example of the element-clay compar-
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.
wells in the field reveals that this correlation Aluminium again shows a strong positive corre-
breaks down entirely in sands containing less lation with clay. Silicon again shows a negative
than 25% clay. correlation with clay, but the impact of carbo-
The second row of Fig. 4 begins with a nate minerals on the silicon--clay curve is much
QUANTITATIVE LITHOLOGY 87

100 w a r

Well I .="/

_~ s o

100
Well5o Well6 o/ Well8 /
/
.'.s/
-./ . ; Y

/
100
I Well9 / Well1 2 ~
.

00
Aluminum
wt% Alumi10num
wt%
20 0 10
'
Aluminumwt%
20 0
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 The improvement of aluminium over gamma
calcium reflecting calcite concentrations that ray is marginal in Well 8, but it is significant for
range from 0 to 85 wt%. This mineral assem- the cleanest sands. In Wells 11 and 12, the
blage produces a ternary composition diagram in aluminium and gamma ray are comparable clay
the silicon-clay plot with the vertices represent- indicators if the core calibration is known.
ing pure carbonate, clean sand, and shale. However, a log interpreter who equates the
Summarizing the observations in Figs 4 minimum gamma ray response with zero clay
through 6, it appears that aluminium is the best introduces a 20 to 25 wt% error in the clay
single elemental indicator of clay. Silicon shows estimation.
a complementary anti-correlation to clay con- Aluminium has an even more striking relation-
tent, but the simple linear relationship between ship with the sum of clay plus mica. This is
silicon and clay is distorted by carbonate demonstrated in Fig. 8. Improvements in the
minerals. The carbonate content is chemically correlation with aluminium are most notable in
represented by calcium and/or iron. These trends Wells 7 and 8, and the effects are most obvious in
are typical of those observed in the other data the shales. The lines drawn in Fig. 8 represent a
sets. slope of 6.4 and the relationship for the first 10
Having observed the strong relationship be- wells has a correlation coefficient of 0.98. It is
tween aluminium and clay, it is useful to examine possible that some of the differences between
the data for all 12 wells, as shown in Fig. 7. In 10 Figs 7 and 8 are due to analytical interference
of the 12 wells, the slope of the aluminium-clay between illite and mica phases in shales 9 The
plot is nearly constant. In 9 of the 12 wells, the decision to include or exclude mica from the clay
intercept of the aluminium-clay linear relation- fraction depends on the application. Since micas
ship is essentially zero. Comparison between Fig. do not contribute significantly to clay counter-
7 and Fig. 3 shows that aluminium is a much ion conductivity, they are not generally included
better clay estimator than gamma ray in most in saturation interpretation. On the other hand,
wells. This is true even when a porosity-free core like clays, micas can be detrimental to formation
calibration is available for gamma ray, and it is productivity.
especially true in the sands 9 There are several reasons for the strong
88 M.M. HERRON & S. L. HERRON
100

I~
+
5c

100 ....

~;
+ 50 / ,/,
tO
|

lOO
Well1 0 / Well1 / Well 12

~ 50

0 lO
|
/ / , .......

Aluminum wt%
20 0 1'0
Aluminum wt% 20 0 1'0
Aluminum wt% 20 0 Aluminum
10
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 into three plots. For samples containing more
mineral content. Clays are aluminosilicates; than 2 wt% organic carbon, the elemental data
aluminium is a major element in and an integral must be normalized to an organic-free matrix or
part of the chemical composition of virtually all else they will perturb the linear relationship.
clays. This is very different from the case of Earlier, Figs 4 to 6 showed that as clay
thorium and uranium which occur at trace (ppm) increases, silicon decreases. Therefore, as alumi-
levels and are not structural components of the nium increases, silicon decreases. In Fig. 9a,
clays. Of course, the clay-A1 relationship is a silicon is converted to SiO2 (by multiplying by
simplified picture and is not expected to be 2.139) and subtracted from 100. Now, we see
perfect. Different clay minerals have different A1 that as AI increases, 100-SiO2 also increases. In
concentrations and there are important nonclay this presentation, carbonate minerals drive the
minerals that contain aluminium. 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
Relationship between AI and Si, Ca, Mg and Fe (CaCO3) and dolomite (CaMg(CO3)2). Fig. 9b
shows that concentrations of A1 vary linearly
Although aluminium is the best element for clay when plotted against 1 0 0 - S i O 2 - CaCO3 - Mg-
estimation, its measurement in a borehole is CO3 concentrations and that the additional
accomplished by induced neutron activation and terms remove almost all of the disturbance of
currently requires a chemical source, two gamma that major trend. The few remaining outliers are
ray spectrometers, and an independent measure- predominantly siderite or pyrite, and they can be
ment of formation capture cross-section, making removed as 1.99Fe where the coefficient of 1.99 is
it an expensive measurement. Fortunately, an optimized on these data.
alternative exists due to the complementary The trend in Fig. 9c can be used to estimate
relationship between aluminium and the ele- the aluminium concentration from
ments silicon, calcium, magnesium and iron.
This relationship is illustrated in Fig. 9, which A1 = 0.34(100 - SiO2 - CaCO3 -
combines elemental data from all 12 data sets MgCO3-1.99Fe), (2)
QUANTITATIVE LITHOLOGY 89

.r
20[a I' b C

/
O Q

o 'ot ;
~ o 00

06 5"0 100 0 50 100 0 50 100

100 - Si02 100 - Si02- CaCOz- 100 - Si02- CaCOs-


MgC03 MgCOs- 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"' / Well 4 '

"10I f o

~< 0 U ; ,

2~IWe,./,5 "

IlU'
2O

<
e

| |
00 10 20 0 10 20 0 10 20 0 10 20
Aluminum Emulator Aluminum Emulator Aluminum Emulator AluminumE m u l a t o r
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 correla- interpretation. This relationship can be quanti-
tion coefficient of 0.99 and a standard error of fied to estimate clay, and the elements calcium,
0.6 wt% A1. Figure 10 presents a comparison of magnesium, and sulfur can be used to estimate
measured A1 concentrations with those estimated the other major mineralogical components. The
from equation (2) for each of the 12 wells. mineralogical fractions defined here are different
Clearly, this is a robust means of estimating A1 from the lithologies commonly used in log
from Si, Ca, Mg and Fe. interpretation. The main difference is that a clay
fraction rather than a shale fraction is computed.
Quantitative lithology According to Bhuyan & Passey (1994), shales
commonly have about 60 wt% clay minerals and
The strong correlation between aluminium and 40 wt% Q - F - M . Using this ratio, a rock with 60
clay provides the cornerstone of the lithology wt% clay is 100 wt% shale. The other difference
90 M. M. HERRON & S. L. HERRON
100 ......... .,
]Well 1 ,~ / Well 4

01r
ol00[Well5 " / Well 6 " o / IWe"7/ jwey.
./
./
Iwe"
0[/r

'O01we"9/

0 50 1O0 0 50 100
l/. 50 1O0 0 50 1O0
Estimated Clay Estimated Clay Estimated Clay 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 minimum of 1.3 for Well 10.
porosity-free (or matrix) basis, and they are If we solve for clay plus mica (Fig. 8) instead
weight rather than volume fractions. of clay, we obtain the following equation:

Estimating clay Clay + Mica = 2.20(100- SiO2-


C a C O 3 - M g C O 3 - 1.99Fe), (4)
The two major points from the preceding section
are that A1 correlates well with clay content and with a correlation coefficient of 0.97 and a
that aluminium concentrations can be estimated standard error of 6.5 wt%.
from Si, Ca, Mg and Fe. The next logical step is Figure 11 presents measured clay content and
to estimate clay content from Si, Ca, Mg and Fe estimates from equation (3) for all 12 wells. The
using the form of equation (2). The problem is estimated clay concentrations are in good agree-
set up to determine clay content by optimizing ment with the measured values for Wells 1-7, 9
the slope. Samples from Wells 11 and 12 are and 10. They are almost the same as the
excluded from the optimization because, as seen estimates from aluminium shown in Fig. 7.
in Fig. 7, the relationship between aluminium For most of the first ten wells, the clay
and clay differs significantly from the relation- estimates portrayed in Fig. 11 constitute an
ships observed in Wells 1 through 10. The new improvement over those attainable from gamma
clay algorithm is: ray. The scatter in the estimate is drastically
reduced, particularly at the low clay concentra-
Clay = 1.67(100 - SiO2 - CaCO3 - tions where clay estimation is most critical. This
M g C O 3 - 1.99Fe), (3) is especially clear in Wells 1-7 and 9 and 10. In
Well 8, the estimate of clay shows a less
which has a correlation coefficient of 0.94 and a spectacular effect relative to gamma ray, but it
standard error of 6.9 wt% clay. The slope of 1.67 does offer slight improvement in the clean sands.
obtained here is representative of the combined In this well, an estimate of clay plus mica would
datasets. Slopes optimized on individual datasets clearly be superior to gamma ray estimates.
range from a maximum of 2.0 for Well 1 to a Equation (3) is a general algorithm for
QUANTITATIVE LITHOLOGY 91

estimating clay from elemental data. It has broad feldspathic sands, so the application of equation
applicability and does not require picks of
minimum and maximum values. Unlike neu-
t 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
loo ;g/
(5) requires some external knowledge.

optimized on individual datasets, the overall ~, 50


slope of 1.67 in equation (3) produces a good
clay estimate. 5
For Wells 11 and 12 the clay estimated from
0 ~-
equation (3) agrees with the measured clay in the 0 50 1 oo
cleanest samples but under-estimates the clay Estimated Clay2
content of the shales. The cleanest samples in Fig. 12. Clay estimated from equation (5) for feldspar-
these two wells have 20 and 28 wt% clay, and it rich sands and shales vs measured clay for Wells 11 and
is not obvious which way the data would trend in l 2 (o) and Well 4 (+).
cleaner rocks. This is the same trend observed for
these two wells in the comparison of aluminium
Estimating carbonate
versus clay.
The problem with the interpretation of clay The second c o m p o n e n t in this lithological
from A1 or from Si, Ca and Fe in Wells 11 and 12 description is the carbonate fraction. The carbo-
is basically the same as the problem with nate fraction will be determined from calcium,
interpreting gamma ray. Inherent in both inter- but first we need to consider the calcium
pretation schemes is the presumption that non- concentration which we obtain from log data.
clay minerals do not interfere. For most wells, Pure calcite (CaCO3) formations have Ca con-
this is true for aluminium. However, Wells 11 centrations of 40 wt%, and this concentration is
and 12 are characterized by feldspar-rich sands. accurately reflected by log data. A complication
This is true to a lesser degree for Well 4. In fact, arises in dolomites (CaMg(CO3)2) because mag-
for all three of these wells, there is an anti- nesium has not normally been detected by
correlation between clay and non-clay alumino- spectroscopy logs. As a result, the log calcium
silicates (feldspars plus micas). The high feldspar concentration in a pure dolomite is also 40 wt%
content of the sands can be either authigenic as (see Hertzog et al. 1987 and Roscoe et al. 1995
in Well 11 or detrital as in Well 12. for detecting Mg from logs). This is equivalent to
In spite of vast geological differences, Wells 11 saying that the Ca detected by logs equals
and 12 show similar patterns in terms of C a + 1.455Mg, an expression that equals 40
aluminium vs clay. This suggests that a common wt% in either pure calcite or dolomite. Using
algorithm might exist to interpret clay content in the core data base, calcite plus dolomite con-
these wells, and if so, it might be broadly centrations were optimized as a function of
applicable to feldspar- or mica-rich sands. The (Ca + 1.455 Mg) to produce equation (6):
relationship determined by least absolute error
optimization on the combined Well 11 and Well Calcite + Dolomite =

12 datasets is: - 7.5 + 2.69(Ca + 1.455Mg). (6)

Clay2 = -20.8 + 3.1 (100 - SiO2 - Here, the non-zero offset of - 7 . 5 wt% accounts
for the small calcium contribution from plagio-
C a C O 3 - M g C O 3 - 1.99Fe) (5)
clase feldspar in sandstones, and the offset and

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 in-
/ ,/
cluded as different symbols; this well was not
included in the optimization. Equation (5) for
feldspar-rich sandstones gives reasonable results o o 50 1 oo
for clay contents in the reservoir rocks despite Estimated Calcite + Dolomite
the fact that these wells are from very different Fig. 13. Calcite plus dolomite estimated from equation
geological environments. Using the geochemical (6) vs measured calcite plus dolomite for all twelve
data alone, it is not possible to identify such wells.
92 M.M. HERRON & S. L. HERRON

oo[ o

so

0
0 50 100 0 50 100 0 50 100
E~imated Clay wt% Estimated Carbonate wt% 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 esti- Anhl = S/23.55 (7)
mate from equation (6) closely approximates the
sum of calcite plus dolomite from all 12 wells Anh2 = Ca/29.44. (8)
(Fig. 13) with a correlation coefficient of 0.98. A
distinction of calcite from dolomite is possible The final anhydrite estimate is the minimum of
with the inclusion of magnesium (Hertzog et al. these two to account for the possibility of non-
1987; Roscoe et al. 1995). anhydrite sources of either sulfur or calcium. The
anhydrite computation precedes the carbonate
Estimating quartz-feldspar-mica and clay estimate's and the anhydrite calcium is
subtracted from the total calcium prior to the
The third component of the new lithological other lithological computations. When solving
description is the sand fraction composed for anhydrite, the Q - F - M fraction is determined
primarily of quartz, feldspars and micas ( Q - F - by subtracting the clay, carbonate, and anhydrite
M). This fraction is determined by subtracting fractions from 100 wt%. Fig. 15 presents a
the clay and carbonate fractions from 100 wt%. comparison of anhydrite measured by FT-IR
Figure 14 shows the estimated and measured and anhydrite using calcium and sulfur from a
concentrations of clay, carbonate, and quartz- single well in West Texas.
feldspar-mica for all 12 wells. In the reservoir 40
rocks, where clay content is less than 30 wt%,
35
the agreement between measured and estimated
concentrations is remarkably good for all com- 30
ponents. In the shales, particularly where clay E
exceeds 50 wt%, the interpretation tends to ~=25
(3.
under-estimate clay and over-estimate Q - F - M .
Obviously, the clay algorithm could be optimized (-920
+
to give more accurate estimates in shales. The e15
carbonate estimates are good over the entire -E
"O
dynamic range. ~r 1 0
<
5
Estimating anhydrite I I I A

10 20 30 40
This three component lithological description is Estimated Anhydrite wt%
easily modified to accommodate formations Fig. 15. Comparison of estimated and measured
containing significant amounts of anhydrite or quantity of anhydrite on a single dataset.
gypsum. The anhydrite estimate precedes the
carbonate estimate to separate carbonate cal- Application to log data
cium from anhydrite calcium. Two estimates of
anhydrite are made, one from sulfur and one The ultimate goal of this study is to identify an
from calcium, according to stoichiometric rela- objective, robust, and efficient means of estimat-
tionships where the sulfur concentration in ing lithology from spectroscopy logs. The two
anhydrite is 23.55 wt. % and the calcium simultaneous developments that have made this
concentration is 29.44 wt.%. possible are the determination of elemental
QUANTITATIVE LITHOLOGY 93

concentrations from induced gamma ray spec- Well 8 is probably the worst example of the A1-
troscopy logs and the derivation of the lithology clay relationship.
algorithms presented above. The interpretation of the cased hole spectro-
To apply these relationships using the data scopy logs from Well 3 (Fig. 2) is presented in
from Figs 1 and 2 requires that the clay Fig. 17. The agreement between core and log
algorithms be modified to account for the known data is quite spectacular considering that these
aluminium interference in the iron measurement. measurements are made with a ll~in, diameter
Equations (3), (4) and (5) for computing clay or tool through casing and cement.
clay plus mica become:
Conclusions
ClayL = 1.91(100 -- SiO2 - CaCO3 - 1.99FeA1) (9)
The quantitative lithology presented here has
Clay + MicaL = 2.43(100 -- SiO2 - been optimized on core data from numerous
wells from around the world. The lithological
CaCO3-1.99FeA1) (10) fractions of clay, carbonate, anhydrite, and
quartz-feldspar-mica are ideally suited for the
Clay2L = -- 18.5 + 3.34(100-- elemental concentration logs of silicon, calcium,
SiO2-- CaCO3 - 1.99FeA1) (11) iron, and sulfur, which can be acquired by single,
induced gamma ray spectroscopy logs. These
where the L subscript designates the application elemental concentration logs could be available
to log data. FeA1 designates the quantity that in both open and cased hole. The strength of this
would be detected as iron by a spectroscopy elemental approach to estimating lithology lies in
device and is equal to Fe + 0.14A1. the use of major element chemistry as opposed to
The clay, carbonate and Q - F - M fractions trace element chemistry which can be so easily
calculated using the Fig. 1 open hole spectro- impacted by sediment diagenesis, depositional
scopy data from Well 8 are presented in Fig. 16. environment, or the spurious introduction of
Also shown are the core clay, carbonate and Q - small amounts of heavy minerals. The elements
F - M fractions determined from the F T A R used are major element contributors to the rock-
mineralogy. The agreement between core and forming minerals. Their concentrations in a
log data is quite good, in spite of the fact that given mineral are relatively stable, and the

20("

40( =......

w 9

60( E

80(
l_
t-t
A

100( =

120(

1400
F"-"
160C

0 50 1 O0 0 50 1 O0 0 50 1 O0
Clay, wt% Carbonate, wt% Quartz-Feld-Mica, wt%

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 !

1.03

L
[-

1.05 B

L
1.07 ~:" i
0 50 1 O0 0 50 O0 0 50 1 O0
Clay, wt% Carbonate, wt% 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 paper DDD. In: 35th Annual Logging Symposium
abundant. Transactions: Society of Professional Well Log
The S i - C a - F e aluminium emulator gives a Analysts, pp. D1 15.
demonstrably superior clay interpretation com- ELLIs, D. V. 1987. Well Logging for Earth Scientists.
Elsevier, New York.
pared to that available from gamma ray. Its GRAU, J. A. & SCHWEITZER, J. S. 1989. Elemental
strength lies in the near constant slope, small concentrations from thermal neutron capture
degree of scatter, and near zero intercept. It is gamma-ray spectra in geological formations.
also independent of fluid volume, type and Nuclear Geophysics, 3, 1 9.
density, rendering it free from gas or light --, ELLIS, D. V. & HERTZOa, R. C. 1989.
hydrocarbon effects, unlike the neutron-density A geological model for gamma-ray spectroscopy
separation. logging measurements. Nuclear Geophysics, 3,
The calcium log provides an unparalleled 351-359.
carbonate estimation. It provides carbonate HERRON, S. L. 1995. Method and apparatus for
determining elemental concentrations for "/ ray
quantification in complex lithologies. In heavy spectroscopy tools, U.S. Patent 5,471,057.
barite muds, it easily and accurately locates - - & HERRON,M. M. 1996. Quantitative lithology:
carbonate cementation at levels of 10 to 20 wt% An application for open and eased hole spectro-
which were previously undetected by conven- scopy. In. 37th Annual Logging Symposium
tional log interpretation. The sulfur log provides Transactions: Society of Professional Well Log
a very accurate estimate of anhydrite which is of Analysts, pp. E1 14.
greatest value in carbonate/evaporate lithologies. HERTZOG,R. C., COLSON,L., SEEMAN,B., O'BRIEN,M.,
While the relationships presented here have SCOTT, H., McKEoN, D., WRA~GHT,P., GRAU, L,
demonstrated a large degree of universality, each ELLlS, D., SCHWEITZER, J. & HERRON, M. 1987.
Geochemical logging with spectrometry tools,
algorithm can be further optimized on a field or SPE-16792. In. 62nd Annual Technical Confer-
regional basis to give improved lithological ence and Exhibition Proceedings: Society of
estimates. Petroleum Engineers.
HURST,A. & MILODOWSK1,T. 1994. Characterization of
References clays in sandstones: Thorium content and spectral
log data, paper S. In: Sixteenth European Forma-
BHVVAN, K. & PASSEY, Q. R. 1994. Clay estimation tion Evaluation Symposium: Society of Profes-
from GR and neutron~tensity porosity logs, sional Well Log Analysts.
QUANTITATIVE LITHOLOGY 95

KATAHARA, K. W. 1995. Gamma ray log response in tubing carbon-oxygen logging tools, paper QQ.
shaley sands. The Log Analyst, 36, 50--55. In: in 34th Annual Logging Symposium Transac-
MATTESON, A. & HERRON, M. M. 1993. Quantitative tions: Society of Professional Well Log Analysts.
mineral analysis by Fourier transform infrared SCHWEITZER, J. S., ELLIS, D. V., GRAU, J. A. &
spectroscopy, Society of Core Analysts Technical HERTZOG, R. C. 1988. Elemental concentrations
Conference, August 9 11, 1993, SCA 9308. from gamma-ray spectroscopy logs. Nuclear Geo-
RoscoE, B., GRAU, L, CAO MINH, C. (~; FREEMAN, D. physics, 2, 175 181.
1995 Non-conventional applications of through-
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 east-
west. 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 three-
dimensional volume.

Nirex is responsible for the development of a excavations and to allow in situ experiments
deep geological repository for solid, intermedi- on rock and groundwater behaviour. These
ate level and some low-level radioactive wastes. measurements were required to provide infor-
Following preliminary geological investigations mation on ground conditions that could only be
of two sites, an area near Sellafield, west obtained from an underground facility and to
Cumbria, was chosen in 1991 for further study. test models of the geology, hydrogeology and
The Nirex science programme aimed to assess geotechnical characteristics and behaviour of the
the suitability of the Sellafield site as the host for rocks.
the repository. Such an assessment required, In the course of the Sellafield site investiga-
among other things, an understanding of the tions, data at a range of scales from microscopic
geology and hydrogeological characteristics of to regional have been collected. The large
the area. volume of data available from the Nirex
The Sellafield region in west Cumbria, Eng- investigations presents problems with respect to
land was the focus of one of the most detailed the estimation of properties at the very large
site investigations projects ever undertaken. This scales required by performance models. In most
investigation aimed to characterize the geology practical applications, the scale of the sample
and hydrogeology of the site to determine measurements is not directly comparable with
whether the site at Sellafield showed sufficient the scale required for the model estimates needed
promise of meeting regulatory targets to permit for the calculations. It is important to evaluate
Nirex to submit a planning application for a the scale of the sample data and the scale
deep repository. An underground Rock Char- required for the final estimates and to apply
acterization Facility (RCF) had been proposed some correction to the sample scale, if they are
in order to allow more detailed characterization different. This corrective process is generally
of the geology and hydrogeology of the area termed 'up-scaling'. In the context of Sellafield,
using direct observations from underground physical property parameters important to the

KINGDON, A., ROGERS, S. F., EVANS, C. J. & BRERETON,N. R. 1998. The comparison of core and 97
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
98 A. KINGDON E T AL.

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

construction of underground vaults required, on Gosforth. The key task of this study was to
the scale of tens of metres may only be measured define an index of rock properties derived from
on core samples at the scale of centimetres or geophysical log measurements down each of the
from geophysical logging of the boreholes at a boreholes in the PRZ. This index was then used
scale of a few metres. to extrapolate those properties across a volume,
The difficulty of extrapolating properties as sampled by the three-dimensional seismic
using data from varying scales makes it difficult survey, aiming to allow prediction of rock
to use data derived at one scale, for example properties at any location within that volume.
borehole core, to another, such as a three-
dimensional seismic survey. In the context of the Location and geological setting
Sellafield investigations, parameters that are
important to tunnelling, such as indices of rock The Sellafield site is in west Cumbria, England
strength need to be derived at one scale and then and situated between the coast of the East Irish
extrapolated to another scale. Techniques to Sea and the Lake District National Park. A map
allow this to be undertaken must have therefore of the area is shown as Fig. 1. Up to the end of
to be both derived and tested. 1995 twenty-six boreholes were drilled within the
This paper examines the possible techniques area as part of the site investigation. The
for comparing data derived at three separate geology from each borehole has been fully
scales: borehole core (centimetre scale), geophy- described ( Nirex 1993; 1995a,b).
sical borehole logs (metre scale) and three- The regional basement in the Sellafield area is
dimensional seismic survey data (10 metre scale). the Borrowdale Volcanic Group (BVG) which
In particular, mathematical techniques were consists of a complex group of Ordovician tufts,
studied that examine relationships between data lapilli tufts and acidic lavas with local inter-
scales; thus demonstrating the validity of the mediate and basic intrusions, and volcaniclastic
methodology of 'up-scaling'. sediments. (Millward et al. 1994). The BVG was
This study largely concentrated on the Poten- deposited as a largely sub-aerial volcanic system
tial Repository Zone (PRZ) an approximately formed by an island arc on the southern margin
four square kilometre area near the village of of the Iapetus Ocean. The BVG is unconform-
THE NIREX INVESTIGATION OF THE SELLAFIELD REGION 99

ably overlain by a south-westerly thickening penetrated to 1600m below ground level. All
Carboniferous Limestone and Permo-Triassic boreholes have been cored from within the St
succession. The Permo-Triassic forms the sub- Bees Sandstone succession to terminal depth
crop across most of the Sellafield area. within the BVG, with total core recovery in
The BVG and (where present) the Carboni- excess of 95% (close to 100% in some of the
ferous Limestone are unconformably overlain later boreholes). In addition, a high quality trial
by the Brockram, a Permian fluvial breccia three-dimensional seismic survey has been ac-
conglomerate with up to cobble sized clasts. The quired across part of the PRZ area. Data
upper part of the Brockram near the coast acquired within the PRZ area therefore allows
passes laterally into the St Bees Shale and particular scope for both deriving detailed rock
Evaporite (Nirex 1993). The St Bees Evaporite properties and up-scaling between different
comprises dolomite and anhydrite, and the St datasets.
Bees Shale is a laminated sandstone, siltstone
and claystone formation. The St Bees Shale is
conformably overlain by the dominantly fluvial Deriving characteristic rock properties
St Bees Sandstone (Barnes et al. 1994) of
Triassic age. This is a sandstone and claystone, The first stage of this project involved the
with the claystone increasing markedly down derivation of the average rock properties in each
succession, particularly in the North Head borehole, for each of the major formations in the
Member at the base. The St Bees Sandstone is Sellafield area. These average properties were
overlain by the Triassic aeolian Calder Sand- then used to determine whether a particular
stone which forms the subcrop in the PRZ area. formation was essentially constant across the
The easternmost of the Sellafield boreholes area or whether there were significant regional
(9A and 9B) were drilled into outcropping BVG and/or local variations in the rock properties.
with the Permo-Triassic succession outcropping Where significant variations in rock properties
to the southwest of these boreholes. This were found the possible causes for the variation
succession thickens towards the southwest were examined. This exercise was first carried
reaching a thickness of 1700m at the Irish Sea out on a regional scale and then concentrated in
coast. In the PRZ area there is 400 to 500 m of more detail upon the PRZ area.
sedimentary cover overlying the basement. Rock properties were studied by comparison
The Permo-Triassic succession occurs on the of geophysical borehole logs from across the
eastern margin of the East Irish Sea Basin, an area. Geophysical logs were chosen because of
extensional basin associated with prolonged the consistent way that the data was acquired,
east-west extension resulting in the dominantly both in terms of techniques and sampling rates.
north-south faulting seen today (Jackson et al., This allows for easy comparison between bore-
1995). holes some distance apart. The main character-
istics of the rock properties studied were
Data sources identified by statistical and graphical techniques
of data comparison.
High quality geological and geophysical data
have been acquired across the Sellafield region Stud), o f velocity
during the site investigation. All twenty-six
boreholes have been geophysically logged using Although many rock properties have been
comprehensive suites of state of the art tools, measured at various scales, compressional velo-
including borehole imaging. In addition, the city is one of the few to have been measured at
boreholes have been extensively cored, allowing all scales, from core to seismic scale. It was
continuous detailed geological and discontinuity therefore chosen as the most representative
description to be undertaken. Detailed gravity property for analysis as an example of the
and magnetic survey data has been acquired average rock property behaviour. The compres-
across the Sellafield region, as well as two- sional velocity of a rock formation is controlled
dimensional seismic data. by the matrix density, the porosity and the fluid
The geology of the PRZ area has been the composition.
subject of a highly detailed investigation. Up to Compressional velocity data for each of the
the end of 1995 eleven boreholes (Boreholes 2, 4 three data scales were derived by different
& 5; RCF1, 2 & 3, RCM1, 2 & 3; PRZ2 & 3) techniques. Core scale data for each of the main
were drilled within an area measuring only rock types were provided by laboratory testing
1200 m by 800 m across the ground surface. All on core samples. Wireline log scale data were
penetrated to the BVG, the deepest borehole 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 Regional studies of velocity
transit velocities from seismic survey informa-
tion. The regional pattern of velocity was studied
Figure 2 shows a frequency histogram of bulk using graphs of midpoint depth against mean
compressional velocity. This shows the velocity formation velocity for each borehole. The
distributions for the three rock types present in midpoint depth of a formation is defined as the
this area (the St Bees Sandstone, Brockram and point equidistant between the top and base of
BVG) in the PRZ. The statistics are derived the sampled section of a formation, regardless of
from the total geophysical logging measure- whether the borehole had sampled the entire
ments in each borehole from within the area. thickness of the formation. This allows compar-
The mean velocity of the St Bees Sandstone is ison of the compressional velocities between
shown to be between 3.5 to 4.5 kms -l, the boreholes without any overprinting of the effects
Brockram between 4.5 to 5.5 kms -1 and the of velocity changes within the formation. The
BVG between 6.0 to 7.0 km s 1. mean velocities have been derived from geophy-
Variations in velocity within the range for the sical logs of the formation and are expressed on
individual rock type are caused by differences in the graph as kilometres per second.
the properties of the materials. This study aimed Figure 3 is a graph of midpoint depth against
to identify these differences and to attempt to velocity for the St Bees Sandstone in each of the
understand their origin. boreholes in the Sellafield region where the
More detailed studies of the nature of the formation is present. Figure 4 shows the same
velocity distributions have been made for two data types for the BVG. Boreholes are shown
rock types: the St Bees Sandstone and the BVG. using different symbols depending on whether
This was done initially on a regional scale and they fall inside or outside the boundaries of the
subsequently more locally in the PRZ area. The PRZ area. Also plotted on the graph are typical
Brockram is of a fairly consistent thickness core sample data that have been pressurized
across the Sellafield Region (approximately under laboratory conditions to simulated
100m) and shows remarkably homogeneous depths.
properties. As a consequence, no further attempt
has been made to characterize variation in rock Results of studies of velocity against midpoint
properties for this lithology. depths. Figure 3 shows that mean compressional
THE NIREX INVESTIGATION OF THE SELLAFIELD REGION 101

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.
102 A. KINGDON ET AL.

velocity of the St Bees Sandstone increases with the material being sufficiently small that the
midpoint depth in a smooth curve. This ex- compaction effect is not as significant. Of greater
ponential increase in velocity with depth is well significance to the properties of the core samples
documented as being caused by decreasing is the fact that core tests were by definition
effective porosity due to increased overburden carried out on samples of intact rock. The
pressure (Birch 1960). The results from the core properties of these samples therefore varies from
samples also demonstrate that increasing depth the bulk rock sampled by geophysical logs,
results in an increase in compressional velocity. which includes the effects of non-intact and
The velocity results from the core samples is fractured rock. This suggests that the variations
somewhat lower than the velocities from equiva- of the bulk rock properties from those of intact
lent bulk depths derived from geophysical logs rock may be a consequence of the discontinuities
in the boreholes. This is a consequence of the within the rock mass.
samples being tested whilst dry (i.e. the pore space
was air filled rather than saturated with water). Local studies gf velocity
Figure 4 shows a plot of midpoint depth
against compressional velocity for those bore- More detailed analysis of the bulk rock velocity
holes in the Sellafield region where the BVG is properties for the boreholes in the PRZ region
present. This plot shows a relatively complex was carried out using box and whisker plots.
pattern in comparison with the St Bees Sand- Box and whisker plots (Figs 5, 6 and 7 and 11)
stone. are a graphical technique which permits an
Most boreholes show a mean velocity over the overview of a complete data distribution,
BVG interval of between 5.0 to 5.5 km s l.The excluding only anomalous data at the extremes
boreholes in the PRZ area (with the exception of of the distribution. This permits the statistical
borehole PRZ3) showing a distinct clustering of comparison of almost the entire data distribu-
mean velocities. Three boreholes 9B, 8B and tion between all the boreholes. The variation in
PRZ3 show significantly lower mean velocities. the range of data as well as average properties
The first two of these boreholes penetrate a short can therefore be assessed. Figure 5 shows the
distance into the BVG. Detailed logging of the symbols used to describe different parts of the
Borehole 8B core (Nirex 1995a) suggested that data distribution on box and whisker plots.
the top section of the BVG is highly altered, Figure 6 is a box and whisker plot of the
which was the only part of the BVG sampled in velocity distributions of the St Bees Sandstone
this borehole. In the case of Borehole 9B, which for the boreholes in the PRZ. The column on the
was drilled where the BVG outcrops at surface, right-hand side of this diagram shows the
the rocks will have been affected by recent combined properties for all the boreholes. The
weathering. As these effects will have had a quartile range of the velocity for all the PRZ
significant influence on rock properties of these boreholes is between 3.6 and 4.4 kms 1. The
two boreholes, the data are not comparable with bulk rock properties from all boreholes are
the other boreholes and they have not therefore essentially consistent between all the PRZ bore-
been included in the study of the average rock holes. This indicates that there is little regional
properties of the deep BVG. In the case of variation in the bulk rock properties of the St
borehole PRZ3, where the BVG is covered by a Bees Sandstone in this area.
thick Permo-Triassic succession, the difference Two additional datasets are also displayed on
could not be readily explained in this way and this plot, the core properties and the faulted rock
the BVG velocities from this borehole are properties. Core properties were derived from
therefore seen to be significantly anomalous. laboratory tests on intact core samples. Labora-
An independent quality assurance check of the tory testing of the properties of the fault rock
geophysical logging of Borehole PRZ3 did not was difficult as fault rock is by definition non-
indicate any systematic error in the acquisition intact. The fault properties shown here were
and processing stages. Borehole PRZ3 was therefore derived from geophysical log measure-
targeted to intersect a fault, Fault F1, in the ments. The standard geophysical log measuring
BVG and it is likely that the mean velocity increment of 6 inches (15.24cm) means that
results from this borehole reflect a large propor- statistically valid samples are hard to obtain
tion of 'faulted rock'. from faults with limited borehole intersections.
The velocities of the available core samples in Therefore properties were only derived for faults
the BVG are slightly higher than the wireline with borehole intersections greater than 50
derived mean formation velocities, despite the centimetres.
core samples again having been tested whilst The core sample seismic velocities are seen to
dry. This is inferred to be due to the porosity of be much lower (interquartile range for all
THE NIREX INVESTIGATION OF THE SELLAFIELD REGION 103

NUMBER OF POINTS NUMBER OF POINTS NUMBER (')F" POIN'IS

95%

UPPER UPPER UPPER


QUARTILE QUARTILE QUARTILE

MEDIAN MEDIAN MEDIAN


_ -

MEAN MEAN MEAN

LOWER LOWER LOWER


QUARTILE QUARTILE 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 1647 2411 2506 2293 2428 1627 2645 2605 2475 23611
?-7 10 9 22 75 18 9 121 59

4.5-

~'~ 4.0--

r~
I I
3.5-

3.0--
tI
I ,., I ,., I - I ~ i ~ i .~ i .~, i ,~ i ~.~ i ~ i ,0,, i ~. 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 datasets was hampered by the small number of
bulk rock as they were tested on dry samples (see sample points and the sampling bias, it is clear
above). The fault data are also somewhat lower that the properties of fault rock are significantly
(interquartile range for all boreholes 3.3 to 3.9 different to the bulk rock. This can be seen in the
k m s 1). Whilst the interpretation of these two 'total column' which includes the summed data
104 A. KINGDON E T AL.

7.0

7238 5482 4950 168 389 3878 3@66 2911 14188 1480 3046 34896
85 t l i 378 76 104 75 27 77 118 23 43 28 119 1002 262

6.0

5.0"
B
tt
4.0'
ti
3.0

BOREHOLE NAMES
I
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 105

for all boreholes. velocity profile of the St Bees Sandstone largely


As can be seen in Figure 7 most boreholes in reflects the depth of burial and there is therefore
the PRZ area shows a consistent set of bulk rock an increase in average velocity to the south-west
velocity (interquartile ranges between 5.0 to 6.1 where the midpoint depth of the formation is
kms-1). Only borehole PRZ3 displays signifi- greater.
cantly different properties with lower values Figure 8 shows a typical cross plot of
(interquartile range 4.5 to 5.0 kms-l). Compar- percentage porosity against density of the St
ison of the three measurements for each bore- Bees Sandstone from Borehole 2. The depth of
hole provided important evidence to the each point is displayed as the z-axis (the
controlling mechanism for bulk rock properties. colouration of the points in Fig. 8). The diagram
Whilst core derived intact rock properties shows that most of the data plots along a line,
showed somewhat higher velocities than the with density increasing as porosity decreases and
bulk rock properties in almost all cases, the fault depth increases. This distribution is caused by
derived velocity values were significantly lower compaction (due to increasing overburden pres-
than the equivalent bulk rock properties and sure) leading to decreasing porosity with depth.
showed greater variability. Some variations from this simple distribution
The difference between bulk rock and intact are seen. In order to understand the anomalous
rock is, by definition, the discontinuities of points, the effects of variations in lithology had
which fault rock was the only measurable to be quantified. Figure 9 shows the density-
example. Hence the discontinuities must be the porosity cross plot for the same data but with
dominant controlling factor on the bulk rock the z-axis now showing the gamma-ray count
properties of the BVG. In the case of borehole from each point. This clearly shows the effects of
PRZ3 the velocity measurements showed a very lithological change as higher gamma counts
localized anomaly, consistent with the targeting occur in those parts of the distribution which
of the borehole into a faulted zone. do not follow the simple trend. This is because
these zones are not clean sandstone but contain
Causes of variataions of rock properties in a significant proportion of claystone.
the St Bees Sandstone These two diagrams therefore indicate that the
dominant control on the bulk rock properties of
The compressional velocity distribution of the the St Bees Sandstone was depth of burial and
boreholes described above shows that the 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 bore-
holes 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 Fig. 12. A diagrammatic fractal distribution: the
properties of the intact rock. In addition a Sierpinski Gasket.
smaller cluster of points with low resistivity,
higher porosity and high gamma ray counts are
also seen. The higher gamma ray counts are volume, such as those sampled by the trial 3-D
probably caused by the alteration minerals seismic survey are important for successful
found around discontinuities within the rock engineering design of tunnels, shafts etc. The
mass. This data acted as a further indication that P R Z area has eleven deep boreholes drilled by
the dominant control over the bulk rock proper- Nirex, and several drilled previously by British
ties of the BVG are the nature of the disconti- Steel and its precursors, and yet less than one
nuities within the rock mass rather than the five millionths of the total volume of the P R Z
intact properties of the rock itself. has been directly sampled by coring.
In an attempt to quantify the effects of the In order to scientifically justify the up-scaling
properties of the faults on the bulk rock of known properties (derived from either direct
properties a box and whisker plot of the gamma measurements on core or indirect measurements
ray counts for the BVG of the P R Z boreholes from geophysical logging tools) it is important
was produced (Fig. 11). In most of the boreholes to demonstrate that at least over a small but
with a statistically significant sample of fault significant part of the scale, properties are
rock, the gamma ray response is approximately comparable. If this can be done then 'up-scaling'
5 API higher in the fault rock than in the bulk of data can be seen as a legitimate concept,
rock. This can be seen most clearly in the total although it should be treated with caution.
(all boreholes) column on the right-hand side of
the diagram. Fractals
This is not an ideal presentational medium
because a fault in an already low gamma ray A true fractal relationship is a relationship
formation may have a lower count than high between two variables that does not change
gamma background elsewhere in the borehole. with scale. Whilst it is unlikely that any relation-
Variations in the condition of the boreholes will ship is truly fractal in a natural system, if a
also significantly affect the results. The data for relationship could be demonstrated over a
Borehole RCM3 is dominated by a single large number of orders of magnitude then this could
fault close to the top of the BVG where low be used to justify up-scaling of data from one
gamma is recorded because of poor hole scale to another. Figure 12 shows a diagram-
conditions (the caliper increases from 6 to 16 matic fractal relationship, the Sierpinski Gasket.
inches through this fault). Despite these pro- Each size of triangles is related to the next
blems the diagram does clearly show that largest and next smallest size of triangles by the
gamma ray counts are higher in fault zones than same scale and geometric relationships, up to the
in bulk rock. limits of page size in one extreme and print
resolution in the other.
Data scales and up-scaling An attempt was made to study discontinuities
in the Sellafield area at two separate scales:
Relating physical properties derived from one distances between individual discontinuities
scale to another represents a significant problem. measured directly from the core and distances
Prediction of the rock properties through a given 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 distribu- (frequent) events. In this case, classes of 0.1 m to
tions were fractal at each scale and then to see 100m were used as applicable to borehole
whether any link between data at the two scales discontinuity spacing, covering three full log
could be established. cycles. In order to analyse the data an approx-
imate geometric progression was used to divide
Discontinuity separation the data up into frequency classes suitable for
log-log output.
Various techniques have been derived to study The cumulative frequency data wwere plotted
the fractal dimension for a distribution of as a log-log plot. To be considered fractal the
naturally occurring phenomena. This study was distribution had to plot as a straight line
carried out using the Spacing Population Tech- (showing that the relationship is scale invariant
nique (after Harris et al. 1991) which is both and conforms to the following function):
straight forward and applicable to the type of y=ax - D
data to be examined.
The basic dataset for this study was the where: y = probability (cumulative frequency);
borehole discontinuity log, produced for Nirex a = a prefactor;
by Gibb Deep Geology Group (GDGG) from x = the discontinuity spacing;
direct measurement of the core. This lists, for D = t h e line gradient (i.e. the fractal
each borehole, all the occurrences of faults, dimension).
veins, joints and other discontinuities, ordered For a distribution to be considered fractal the
by depth. All discontinuities with a non-struc- data should be linear across at least one order of
tural origin were removed, such as bedding magnitude.
features, stylolites in the Carboniferous Lime-
stone and those fractures in the core that were Results of fractal analysis
induced by the drilling process.
Figure 13 shows a cumulative frequency log-log
Methodology plot for Borehole 8B, which gives the results of
the fractal analysis of borehole discontinuities in
The Spacing Population Technique is based on this borehole. Some of the limitations of the
cumulative frequency distributions derived in- fractal technique are demonstrated by this
crementally from large (infrequent) to small 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 sampled by boreholes, occur only infrequently
magnitude, any single measurement technique and the dataset is statistically insignificant in any
may only characterize a subset of this total one borehole. Therefore another measurement
range. The limits for identification of fractal technique must be used to sample the larger
patterns are often therefore controlled by faults. An obvious alternative method is to
limitations in the sampling method. In this study examine faults identified from a two-dimen-
for instance the core fracture dataset was reliant sional seismic grid. This dataset was used in this
on the human eye to identify individual frac- study for a comparison with the borehole
tures. derived results. It was important to make clear
Very closely spaced fractures lead inevitably at this stage that these were not identical
to highly broken core and poor core recovery, so datasets simply measured at different scales.
that such zones are preferentially under- Seismic reflection profiles (and in particular
sampled. Rock, heavily fractured by localized, widely spaced two-dimensional seismic data)
closely spaced events is essentially indistinguish- tend to resolve only large scale fault zones
able from large fractures and will behave in a rather than distinguishing individual minor fault
similar way. Major fault systems have not been strands such as those which would be delineated
sufficiently sampled by boreholes to permit from borehole core.
fractal analysis at this scale, whereas the three-
dimensional seismic survey only identifies faults Methodology
either by 'significant' offset of marker lithologi-
cal contacts or directly where the thickness of In order to get an acceptable level of coverage of
fault-rock is sufficient to cause a velocity fault features with a common resolution, off-
contrast. Also the Sellafield Site boreholes shore seismic reflection data from near the
cannot be labelled a random unbiased dataset, Sellafield site were used for this study. Unlike
as some of the boreholes were specifically boreholes, which essentially sample a one-
targeted at some of the major faults in the dimensional environment, the interpreted seis-
region. mic fault maps used in this part of the study
The borehole 8B fracture set shows a clear were two-dimensional in character. The dataset
fractal relationship over two orders of magni- in this case were fault maps stored in a database
tude, i.e. at fracture spacings from 20cm to of faults derived from V U L C A N software
10m, with a regression coefficient of modelling of the regional structure.
R 2= 0.9996, very close to a perfect straight line. A different sampling technique was used in
order to develop cumulative frequency data. In
Fractal events at the seismic scale this case a two-dimensional 'box counting'
method was applied. This was done by over-
Large events, such as major faults, although laying 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 Fig. 16. An idealized artifical neural network.
Sea basin were studied (Nirex 1995b); The base
Carboniferous (Fig. 14) and Base Permo-Trias-
sic (Figure 15). Both showed very clear fractal their respective scale ranges. It is also possible to
patterns over the scale range from 200m to debate that if up-scaling is valid over both these
1 km, with regression coefficients of R 2= 0.999 ranges there may be the possibility of establish-
in both cases. ing a relationship between the different fractal
Fractal patterns have been demonstrated over equations and thus defining an up-scaling
two different scale ranges for discontinuity function all the way from micro fracture scale
spacing events. The regression coefficient for to major faulting.
both sets of events was very close to one (i.e. a
completely fractal pattern). However the scaling Neural network modelling
exponents, sometimes known as the fractal
dimension, differ suggesting that fracture spa- Artificial Neural Networks (ANNs) are a
cing is a scale dependent parameter. These computer modelling technique that work in a
relationships provide evidence that up-scaling manner analogous to the processes of a mam-
of discontinuity events from features measured malian brain. They are based on simple linear
in core and at the seismic scale are valid within 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 Neural networks may allow another approach in
'learn' to recognize patterns in data and develop identifying fracture frequency.
their own generalizations. A diagrammatic
model of an idealized artificial neural network Multi-layer preceptron
is shown in Fig. 16.
Fracture frequency measured from borehole This study used a type of A N N called a multi-
core was not easy to predict with any degree of layer preceptron (MLP) to model the relation-
accuracy from conventional geophysical log ship between core derived fracture frequency
measurements. Whilst borehole imaging tools and geophysical log measurements.
go some way to addressing this issue, it was not The MLP consists of a series of simple
always possible to distinguish between features processing elements (nodes) connected to one
such as bedding features and discontinuities. 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 Fig. 19. Comparison of RMR and wireline logs for
frequency was derived from the borehole dis- borehole RCF2.
continuity logging file, binned at intervals of one
metre. The two datasets were not immediately
analogous because fracture frequency had been frequency values for BVG sections in boreholes
calculated per metre whereas geophysical logs RCF1 and RCF2.
conventionally sample at 0.1524m (6 inches). The training dataset consisted of the following
Geophysical logs were filtered using a twelve data segments 840-1015 mbRT (metres below
point moving average filter using BGS WELL- the rotary table datum) from RCF1, 525.5-750
OG software. Data were then interpolated to a mbRT from RCF2 and 732.5-932.5 mbRT from
one metre value and extracted in EXCEL 5.0 for RCF3. The data were run through a network in
manipulation. Six geophysical logs were used as its natural order and then randomized to
input for this study: density, neutron porosity, compare the performance of the network.
gamma ray, shallow resistivity, compressional
velocity and shear velocity.
The fracture frequency data were also subject Results of neural network analysis
to some biasing and were therefore subjected to
a five point moving average filter. These were Figures 17 and 18 show the results of the neural
also exported from BGS WELLOG software to network analysis for boreholes RCF1 and
EXCEL 5.0. RCF2. This dataset shows clearly both the uses
Neural network modelling for this study used and the limitations of using ANNs for model-
Neural Connections V1.0 software from SPSS. ling. Whilst in some parts of both boreholes, the
Various network topologies and statistical test actual fracture frequency had been modelled
were applied to validate the results. The software with some accuracy, in others the modelling had
selected the network topology, usually the X-3-1 not adequately resolved the distribution.
layout (X nodes in the input layer, three hidden The actual fracture frequency was shown at
layer nodes and a single output node). the top of both diagrams with zones of similar
levels of general fracturing marked by a black
Network training line. The bottom diagram shows the A N N
predicted results, again with a black line mark-
Training was highly important to the perfor- ing the zones of similarity. Comparison of both
mance of a neural network. Although it was boreholes shows that the models were good at
possible for ANNs to generalize and infer noise distinguishing the background level of fracturing
obscured properties, the network response was in the boreholes. The biggest problems in the
better where it has been trained by high quality models were at the data extremities. Although
data. In this exercise data from boreholes in the this technique was not perfect it does again give
PRZ area were used to model the fracture clear indications that the fracture frequency data
THE NIREX INVESTIGATION OF THE SELLAFIELD REGION 113

derived from core can be up-scaled to be considerable more research to prove valid. If
modelled by geophysical logs. suitable algorithms can be derived, then extra-
polation of geophysical parameters derived from
Comparison of RMR and conventional wire- geophysical logs or cores, across a three-dimen-
sional seismic grid, should allow detailed pre-
line logs diction of the properties for that volume of the
The rock mass rating (RMR) is an industry rock mass.
standard rock quality and strength index derived
from direct measurement of the physical attri- References
butes of the core and is completely independent
of wireline log measurements. This measurement BARNES, R. P., AMBROSE, K., HOLLIDAY, D. W. &
JONES, N. S. 1994. Lithostratigraphic subdivision
allows an accurate assessment of rock strength,
of the Triassic Sherwood Sandstone Group in
but is labour intensive and therefore expensive West Cumbria. Proceedings of the Yorkshire
to collect. R M R is calculated for whole and Geological Society, 50, 51-61.
partial core runs and is reported for Boreholes BIRCH, F. 1960. The velocity of compressional waves in
RCF1, RCF2 and RCF3 in 3 m intervals. Figure rocks to l0 kilobars, Part 1. Journal of Geophy-
19 shows the results of a comparison of R M R sical Research, 65, 1083-1102.
with two conventional wireline log measure- CHADWICK,R. A., KIRBY, G. A. & BAILY,H. E. 1994.
ments from the same borehole over a given The post-Triassic structural evolution of north-
interval. The degree of correlation between these west England and adjoining parts of the East Irish
Sea. Proceedings of the Yorkshire Geological
two sets of measurements is high despite the
Society, 50, 91-103.
wholly different derivation and supports upscal- CHAeLOW, R. 1996. The geology and hydrogeology of
ing of the wireline log data from a measurement Sellafield : an overview. Proceedings of the
scale of 15cm to at least 3 m by simple NIREX seminar, 11 May 1994. Quarterly Journal
arithmetic averaging. of Engineering Geology, 29, Supplement 1.
HARRIS, C., FRANSSEN, R. & LOOSVELD, R. 1991.
Fractal analysis of fractures in rocks: the Cantors
Conclusion dust method-comment, Tectonophysics, 198, 189-
Whilst neither the fractals nor the artificial 197.
HIRATA,T. 1989. Fractal Dimension of fault systems in
neural network derived models showed exact
Japan: fractal structure in rock fracture geometry
matches with the core derived data from which at various scales. Journal of Geophysical Research.
they were extrapolated, both showed that there 94, 7507-7514
was considerable scope for the belief that using JACKSON,D. I., JACKSON,A. A., EVANS,D., WINGEIELD,
the correct criteria, it is possible to up-scale data R. T. R., BARNES,R. P. & ARTHUR, M. J. 1995.
to match both wireline and seismic scale data. United Kingdom offshore regional report. the
The Rock Characterization Facility (RCF) geology of the Irish Sea. British Geological
proposed at Sellafield requires detailed rock Survey.
properties to be derived from boreholes and MILLWARD, O., BEDDOE-StEPHENS, B., WILLIAMSON, I.
T., YOUNG, S. R. & PETTERSON, M. G. 1994.
extrapolated across a wider area to allow for
Lithostratigraphy of a concealed caldera-related
prediction of the likely tunnelling parameters. ignimbrite sequence within the Borrowdale Vol-
Where three-dimensional seismic survey data are canic Group of west Cumbria, Proceedings of the
available across an area, it should be possible to Yorkshire Geological Society, 50, 25-36.
derive rock properties at a borehole scale and NIREX, 1993. The Geology and hydrogeology of the
extrapolate them across a three-dimensional Sellafield area, Volume 1: The Geology. Nirex
volume to give an accurate prediction of the report 524.
nature of the RCF site. This was dependent NIREX, 1995a. The Geology of the Sellafield Boreholes
upon a detailed knowledge of the rock proper- Nos. 8A and 8B. Nirex report 638.
NIREX, 1995b. Sellafield geological and hydrogeological
ties and accurate correlation of core and seismic
investigations. Factual report-compilation of maps
properties. and drawings, Volume 1 of 2. Nirex report SA/95/
The concept of up-scaling parameters derived 02.
at one scale to another may be feasible but needs
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 Aix-
en-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 high-
resolution 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 palaeocli-
matic 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 context of the Ocean Drilling Program (ODP) at
continuous high resolution records. The logs Oki Ridge in the Japan Sea (Leg 128 Hole
reflect the physical and chemical variability of 798B). Core data are first used to construct a
the drilled sequence. Several logging tools based mineralogical model of the sedimentary forma-
on widely varying physical principles (electric, tion at Site 798. Classical log analysis is then
acoustic, nuclear . . . . ) are used. The logs offer applied to deconvolve the different sedimentary
different perspectives about changes in sediment inputs and to compute a continuous formation
composition. Hence, extracting sediment char- factor (FF), which offers a tool to describe the
acteristics or palaeoclimatic information from sediment pore structure. However, this approach
downhole logs appears as a promising field of is limited by the vertical resolution of each
application. The principal limitation on integrat- sensor, that is half a metre on average for
ing logging data for sedimentary and palaeocli- traditional downhole measurements (e.g. logs).
matic studies is the absence of a generally An original approach combining analytical and
applicable method to transform logging data numerical modelling is proposed here to perform
into reliable sediment physical properties or a small scale analysis of downhole logs through
palaeoclimatic data. the computation of the formation factor.
The major objective of this study is to develop
a new methodological a p p r o a c h using, in Geological setting at Site 798
combination, logging, coring data and high-
resolution electrical images (FMS) to derive the Site 798 (37~ 134~ is located in the
detailed structure of near sea-floor sediments. southeastern Japan Sea, about 160km north of
The method is tested on data obtained in the the western coast of Honshu. The site is

LAUER-LEREDDE,C., PEZARD,P. A., TOURON,F. & DEKEYSER,1. 1998. Forward modelling of the 115
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 sedi-
ments 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 palaeo-
climatology. 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 character-
ized by rythmic changes between dark, lami-
nated, diatom- and organic carbon-rich
conductive intervals, and light-coloured, non-
bioturbated 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
Fig. 2. Formation MicroScanner (FMS) micro-resis-
varies between 3 and 43 Wt% in the upper tivity images from the ODP Hole 798B (from 200 to
320m. DeMenocal et al. (1992) also analysed 300 mbsf). The images are azimuthal traces of the four
contiguous samples over three intervals located pads pressed along the borehole wall. Black represents
between 100 and 320 mbsf (metres below sea low resistivity, and white, high resistivity.
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 117

Core Opal SVL


(Ohm m)
recovery (wt%) 0.45 0.55 0.65
_ o~ ~o 20 30 40 so loo 100

OPAL
(%)
II0 10 . lo ,,0 110

120 120

100
I
130 130

s 140 m 140

8
150 150
200 i
Fig. 4. Correlation between the SFL log and opal
percent measured on core in ODP Hole 798B (after
I DeMenocal et al. 1992).

No data
2so the dominant opaline component throughout
the upper 300m; radiolarians and silicoflagel-
lates contribute in a minor way to the opal flux
(Ingle et al. 1990). The periodicity of the
sedimentary cycles was estimated with stan-
dard-power spectral analysis method (Imbrie et
300 al. 1984): the power spectra of the gamma-ray
Fig. 3. Weight percent biogenic opal versus depth in (SGR) time series showed a peak at about 40 ka,
ODP Hole 798B (after Dunbar et al. 1992). probably a climatic expression of the 41ka
'Milankovitch-type' cyclicity (DeMenocal et al.
1992). This suggested that the earth obliquity
floor). These samples were analysed for major was the driving factor of climate and sedimen-
sediment composition: biogenic opal content tary supply in this region over the last 3 Ma.
varies between 5 and 40%, and terrigenous silts The diatomaceous sediments of the dark
and clays, between 40 and 80%. Core-log facies, and the terrigenous-rich signature of the
correlations were established using ash layers light-coloured lithofacies suggested that these
identified in core photographs and Formation cycles also reflect variations in oceanic surface
MicroScanner T M (FMS) images. High opal productivity combined with continental aeolian
values result in low gamma ray, bulk density, dust from central Asia, as a consequence of
grain densities, and resistivity log values. There palaeoclimatic changes. The terrigenous miner-
is a close correspondance between the SFL and alogy assemblage is similar to that of Chinese
the opal data (Fig. 4). Low opal content is loess, a probable up-wind source of the aeolian
balanced by increases in terrigenous sediment, dust. The Chinese loess deposits may indicate a
and this is recorded by high gamma ray log linkage between glacial climate and Asian
values (DeMenocal et al. 1992). Core-log aridity (Kukla et al. 1988), so the periodic
comparisons therefore demonstrate that log increases in terrigenous concentration may
cycles reflect variations in terrigenous sediment reflect the downwind propagation of this
supply and diatomaceous opal. Diatom tests are signal.
118 C. LAUER-LEREDDE E T AL.

Table 1. Chosen physical properties fop" mah7 components

Phase Component Densit~r PEF CEC


gcm - ba e 1 meq gq

Continental Illit e 2.50 3.5 0.1-0.4


Chlorite 2.60 6.3 0.05-0.4
Kaolinite 2.42 1.83 0.03-0.15
Smectite 2.12 2.04 0.8-1.5
Quartz 2.65 1.8 0

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

Zones* Illite Chlorite Kaolinite Smectite


(mbsf) (%) (%) (%) (%)

Za (200 220) 85 10 0 5
Zb (220~ 225) 80 10 10 0
Zc (225- 260) 90 10 0 0
Zd (260 280) 85 10 5 0
Ze (280-300) 80 5 10 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 and carbonate contents average less than 4%
between 200 and 455 mbsf. In this section,
This work is focused on a 100-m-long interval volcanic ash layers are thin and scarce. The
(from 200 to 300 mbsf) because of the well- continental input is therefore assumed to be
expressed cyclicity over this segment (Fig. 2). composed at this site of four clay minerals (illite,
chlorite, kaolinite, and smectite) and quartz. Six
Mineralogical model components are consequently taken into ac-
count in this study, and characterized by three
In the following study, the sediment physical physical properties: density (g cm 3), photo-
properties and the main mineralogical compo- electric-effect (ba e q ) and cation exchange
nents are used to compute the relative propor- capacity (meq g l). The reference values for
tions for oceanic and continental supplies, and each of these components (Table 1) are chosen
to determine the formation factor (FF). A first from the literature (e.g. Grim 1968; Fertl &
order mineralogical model of Hole 798B is Frost 1980; Juhasz 1981; Caill6re et al. 1982;
deduced from smear slide observations of Drever 1982; Schlumberger 1994).
dominant lithologies, and from previous sedi- Our objective is to obtain information on
ment composition studies. oceanic and continental supplies, rather than on
The oceanic input, essentially diatoms, is the relative fractions of the main mineral
associated with opal, on the basis of Ingle et components.The proportions of each element
al. (1990), DeMerlocal et al. (1992), and Dunbar for the continental phase are, however, needed
et al. (1992) works. The continental one is in order to estimate the physical properties of
deduced from Dersch & Stein (1992) core this phase. A short interval from 200 to 300 mbsf
analyses at Site 798. In order to get information (1.7 to 2.5 Ma) was chosen as a first step of this
about the composition of the terrigenous sedi- analysis. This interval was divided in five zones
ment fraction, Dersch & Stein (1992) determined characterized by average clay fractions (Table
the average amounts of quartz and clay miner- 2), on the basis of previous analyses (Dersch &
als. The entire sequence is characterized by Stein 1992). This division in zones allowed to
quartz amount ranging between 5 and 20%. In simplify the mineralogical model still further,
the upper 413m, the clay fraction is dominated and to reduce the number of unknowns: for
by illite with values between 60 and 88% and example, the continental phase is constituted
chlorite, between 0 and 27%. Calcareous com- with only three elements (illite, chlorite, quartz)
ponents are either absent or poorly preserved, for zone C (Table 2). The sediment averages
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 119
Natural G a m m a Ray Electrical Resistivity B u l k density Photoelectric effect N e u t r o n Porosity
CGR (API) (fl m) (g ce-l) (%)
(ba/e')
20 40 60 0.4 0.5 0.6 0.7 1.4 1.6 1.9 2.2 2,6 6O 70 8O
200 I I I 'T'" I 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 tool (ACT) and the NGT, was run (Ingle et al.
between 200 and 260 mbsf and 60% clay 1990). The depths of investigation are sensor-
between 260 and 300 mbsf (Ingle et al. 1990, dependant, and data are typically recorded at
Dersch & Stein 1992). The percentages of the intervals of 15cm. The quality of the logs
continental phase are then deduced for each obtained is generally excellent: most of the logs
zone. reflect variations in biogenic opal production
(diatomaceous) resulting from glacial-intergla-
Downhole measurements cial changes in surface productivity (Matoba
1984; Zheng 1984; Morley et al. 1986).
At the completion of coring operations at Hole
798B, four logging runs were completed from 70 Downhole geophysical logs (m-scale). The fol-
to 518 mbsf. During the first phase, the phaser lowing analysis focuses on the resistivity (SFL)
dual induction tool (DIT), the long-spacing and natural gamma ray (CGR) logs from 200 to
digital sonic tool (SDT), and the natural 300 mbsf in Hole 798B. These logs were selected
gamma-ray spectrometry tool (NGT) were run because the resistivity is influenced both by the
(seismic stratigraphic tool string). The second clay fraction and diatoms (oceanic productivity
phase consisted of the lithoporosity combination input), whereas the natural gamma ray is mainly
tool string including the lithodensity (LDT), sensitive to the clay fraction (aeolian continental
compensated neutron (CNT-G) porosity, and input in this case). The CGR is often used to
natural gamma-ray spectrometry (NGT) tools. indicate downhole variations in clay minerals
After the Formation MicroScanner T M (FMS) content, because it reflects gamma-ray radio-
had been lowered downhole, the geochemical activity from the decay of potassium and
tool string, including an induced gamma-ray thorium which are common elements in clay
spectroscopy tool (GST), an aluminium 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 Deconvolution o f cont&ental and oceanic
terrigenous aeolian component.
inputs
Throughout the studied depth interval, the The aim is to analyse the downhole measure-
highest CGR value corresponds to the highest ments in order to deconvolve both oceanic and
bulk density values (Pb), the highest resistivity continental inputs. The model of the formation
values (R0), and the lowest porosity values (~b) consists of only two known inputs in unknown
(Fig. 5), reflecting a high terrigenous content proportions. Bulk and matrix densities (Pb, Pma,
relative to the biogenic supply. Terrigenous clays g CC-1) and photoelectric effect (Pef, ba e-1) were
have high K and Th contents and relatively chosen to define proportions of the two compo-
higher density and lower porosity than sediment nents. Whereas Pb responds primarily to poros-
with higher opal content: the clay particles filling ity, the Per responds primarily to rock matrix
the pores induce a lower porosity, hence a higher (lithology). The combination of Pb and the Pef,
resistivity. Porosity is, to a first order, propor- the photoelectric absorption cross-section
tional to the inverse square-root of resistivity (Schlumberger 1994), is:
(Archie 1942). The sediments rich in diatomac-
eous opal commonly have high porosities U = Pef x Pb (1)
because of the intrinsically high porosity of
diatoms themselves. These cycles are also and obeys a linear mixing law such as:
apparent in the recovered sediment record: the
high gamma-ray, high density, high resistivity, U=~ Uf-~-(1--(~) Uma (2)
and low porosity levels correspond to the
massive clay-rich intervals, whereas the low where U, Uf, and Uma are for example the
density, low gamma-ray, low resistivity, and photoelectric absorption cross-sections of the
high porosity units correspond to the darker, media, pore fluid and matrix, respectively.
diatom-rich intervals. As our matrix consists of a mixture of two
inputs (oceanic and continental) with relative
High-resolution (cm-scale) electrical images. The weight fraction (#o and #c) and photoelectric
Formation MicroScanner TM (FMS) creates a absorption coefficients Uo and Uc:
picture of the borehole wall by mapping its
electrical conductance using an array of 16 small Uma =/to Uo + #c U~= Pef • Pma (3)
and pad-mounted electrodes on each of four
pads (Ekstrom et al. 1986; Luthi & Banavar The relation necessary to solve for these two
1988; Pezard et al. 1990). FMS data are recorded unknowns is the closure relation of partial
each 2.5 mm as the tool moves up the borehole. fractions:
The vertical resolution of individual features is
about a centimetre. The tool can, however, l=#o+#c (4)
detect thinner features, provided they have
sufficient resistivity contrast to the surrounding The solution can most easily be seen in terms of
matrix. The images registered with the FMS the matrix representation of the set of simulta-
show qualitative conductivity changes, particu- neous equations:
larly due to the different physical properties of
the beds (for example porosity, resistivity of A=R Y (5)
pore fluid or the presence of clays).
The electrical images obtained at Hole 798B where A is the vector of measurements, R is the
(Fig. 2) resolve the cyclicity of sedimentary matrix of known coefficients, and Y is the vector
processes at the site extremely well. Light (dark, of unknown volumes.
respectively) colour is related to the continental The porosity and density logs are first used to
(oceanic) input. compute the matrix grain density Pma. The wet
bulk density Pb is related to the porosity through
a simple mixing law:
Log analysis
Pb = Pwqb + Pma (1 -- qb) (6)
The downhole logs and core measurements,
associated with the proposed mineralogical where Pw is the density of seawater.
model, are used here to determine variations in On the basis of our preliminary mineralogical
sedimentary inputs, and to compute the forma- model which consists of six major components
tion factor. (opal, illite, chlorite, kaolinite, smectite, quartz),
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 121

Uo and Uc are computed for the five zones Cation exchange capacity. Values of CEC can be
previously determined (Table 2) as follows: measured directly on rock samples, but not
directly in situ. Several attempts have conse-
Uo = Pef(opal) • Pma(opa,) (7) quently been made to derive CEC from existing
logs.
Previously developed CEC or Qv estimates
u~ = L #i • Pefi • Pmai (8) from well-logs based on the spontaneous poten-
i=1,5 tial curve (Smits 1968; Johnson 1978), dielectric
with ~i as weight percentage of the component i constant (Kern et al. 1976), reservoir porosity
in the continental phase. (Lavers et al. 1974; Kern et al. 1976; Neuman
The system resolution leads to the weight 1980) and gamma ray (Koerperich 1975; Clavier
fraction of both oceanic and continental phases. et al. 1977; Johnson 1978) have been discussed.
For this study, the correlation between the
natural radioactivity from K and Th elements
Computation o f the f o r m a t i o n f ac t or (CGR) and the CEC was selected. Most shale
The electrical resistivity of saturated sediments is are radioactive due to the presence of K 4~ in the
usually quoted in terms of a formation factor potassium-bearing clay mineral illite. A correla-
(FF) to remove the effect of the pore-fluid tion between gamma ray counts and CEC may
resistivity, because the grains themselves are then be expected. Johnson (1978) showed such a
considered as insulators (Archie 1942): correlation for formation containing largely illite
and kaolinite where the relatively high gamma-
FF = Ro Rw-1 = Cw Co 1 (9) ray count of the illite corresponded to high
potassium content thus making it an excellent
where Ro (respectively Co) is the resistivity in 12 shaliness indicator. Scala (see Clavier et al. 1977)
m (conductivity in S m -1) of the porous medium, found a strong correlation between gamma ray
and Rw (respectively Cw), the resistivity (con- count rate divided by the porosity and Qv. In
ductivity) of the pore-fluid. other words, gamma ray log can be used in some
The formation factor of the porous medium cases and after calibration on core as a substitute
depends on the intrinsic geometry of the pore of the CEC measurement. On the basis of Scala
channels, and therefore describes the manner in data, we estimated the proportionality constant
which the grains are arranged in a sedimentary between the two quantities as follows:
formation (Winsauer et al. 1952). Archie's CEC = (0.005) CGR (12)
equation is generally considered to apply satis-
factorily to clean sands. The presence of clay The computed values of CEC are then
minerals, however, has a detrimental effect on converted into Qv using (11).
Co computations: the capacity of a clay to To check the validity of (12), an analytical
exchange cations at the pore-mineral interface maximum and minimum CEC are estimated
induces the presence of a surface conductivity with CEC values from Table 1 in each zone
term (Waxman & Smits 1968). A resistivity (Table 2). Using the relative proportions of each
model taking into account the effects of dis- clay mineral, an estimate of the CEC of the clay
persed clays was proposed by Waxman & Smits assemblage can be computed, using a linear
(1968) and Waxman & Thomas (1974): summation.

Co = (Cw+ BQv ) FF -1 (10) Formation factor. Waxman & Thomas (1974)


found that B can be related to an exponential
Qv=pma CEC (1-~)qb -1 (11) function of the conductivity, and Juhasz (1981)
proposed the following expression:
where B represents the equivalent conductance
of clay-exchange cations (S m 2 meq-1), as a B = -(1.28) + (0.225)T-(0.0004059)T 2 (13)
function of salinity and temperature, Qv de- 1 + Rw kz3(0.045T--0.27)
scribes the cation exchange capacity or CEC
where T is the temperature in ~ and Rw the
(meq g-a) per unit pore volume (meq cm 3), and
fluid resistivity in f~m.
Pma (g cm-3) is the matrix grain density of the
The mean value of B obtained for Hole 798B
sediment.
from (13) is 3.8 S m2meq -1. Continuous FF
In the following, the successive stages of the
values versus depth may then be evaluated from
computation of the formation factor are de-
(10).
tailed.
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 Hagelberg et al. 1992). Hence, measurements


on core cannot been compared readily with
Grain density. The computed matrix density downhole logs.
(Fig. 6a) exhibits a high degree of variability.
The matrix density reflects the varying clay and Oceanic and continental fractions. The agreement
diatom contents. Diatoms tend to have low between the reconstructed opal fraction curve
densities, sometimes lower than 2.0g cm -3, and core measurements from Dunbar et al.
whereas clay minerals have densities ranging as (1992) is very good throughout the section
high as 2.80g cm -3 (Johnson & Olhoeft 1984). (Fig. 6b), although more measurements in the
The estimated values and the core measurements upper part would be desirable for a better
are in general agreement over the interval, comparison. The major variations are well in
although fine-scale correlations between the phase: for example, Dunbar et al. measured an
two quantities are difficult. This difficulty results abrupt decrease of opal content at about 269
mainly from the discrepancy between the core and 287 mbsf, and our estimated values present
and log measurements themselves. One of the the same feature. Moreover, fine variations
problem with gaseous sediment is that the core appear in the reconstructed opal fraction curve.
recovery is often fragmented and the section is The amount of computed opal is however over-
expanded and disturbed, leading to differences estimated, especially in the upper part (about
between the core and log depth-scales (e.g. 20%). This last point might originate in the
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 123

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 carbo-
nate 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 time-
series dissolution technique and a one-step
dissolution method. In general, the results from
both techniques are comparable, but the one-
step 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 Fig. 7. Formation factor versus porosity plotted on
computed values ranging between 40 and 90% double logarithmic scale. Jackson et al. (1978) and
(Fig. 6c) are in agreement with the measure- Taylor Smith (1971) results are displayed (A, B, C, D,
ments of Ingle et al. (1990). Moreover, the E; F). The present data (G) define a trend described by:
reconstructed continental fraction curve is char- FF = 1.45qb-2'38.
acterized by a significant increase in clay content
between 278 and 286 mbsf, as suggested by the
increase in gamma-ray, bulk density and resis- and the relationship F F = 1.24qb 2.31 is close to
tivity (Fig. 5). that from Oki Ridge. Taylor Smith (1971)
analysed samples from Mediterranean Sea clays
Cation exchange capacity. The computed CEC and found a m value close to 2.20 (Fig. 7). The
and Qv logs (Fig. 6d) follow the variations of results of Jackson et aI. (1978) show that the
clay abundance and are restricted to analytical exponent m depends entirely on particle shape
boundaries. The proposed proportionality con- for unconsolidated sands (Fig. 7). Similar
stant fits well. Guo (1990) measured the CEC of measurements on assemblages of shell frag-
several loess samples from China. The CEC ments, kaolinite particles, and marine illite clays
ranged between 0.07 and 0.28 meq g 1, which is produce similar values of m (close to 2.0),
within the range of the present results. suggesting that the platey nature of the particles
within clays controls the relationship between
Formation .factor. The formation factor derived FF and qb. The high value of m derived for Oki
from the Archie formula is lower than that Ridge sediments is then in agreement with
derived from Waxman & Smits formula (Fig. similar results in formations with large amounts
6e), particularly in high resistivity zones. The of clays (Jackson et al. 1978), especially illite.
data set can be represented by a regression The results obtained for Oki Ridge sediments
similar to that proposed by Winsauer et al. are also typical with regard to the large spread of
(1952), and such as F F = a ~ .... , with a =1.45 F F values. This spread reflects the change in
and m = 2.38 (Fig. 7). This result is in the range shape of particles in relation to the supplies
typical of marine sediment. In a similar ap- cyclicity. High values of FF correspond to the
proach, Henry (1997) analysed clay-rich sedi- continental input, i.e. clays, whereas low values
mentary samples for CEC from the Barbados of FF correspond to the oceanic input.
wedge (ODP Site 948), similar to those from Oki As a conclusion, the simple mineralogical
Ridge: the electrical resistivity varies from 0.5 to model used here appears as well adapted to the
0.8 f~ m, the porosity is larger than 50%, and the description of sedimentary formations with high
grain densities measured on samples are close to porosities (greater than 60 %) and clay content.
2.80 g cm 3. The CEC measured on core by These first results also demonstrate that the
Henry (1997) ranges between 0.2 and 0.5 meq g-i computation of the oceanic and continental
124 C. LAUER-LEREDDE ET AL.
C G R (API) C G R model S F L (f2 m) Rt model FFs
--- computed (API) ... computed (fi m) ... from raw logs
- - measured - - measured -- from modelled logs
200 30 50 30 50 0.5 0.7 0~5 0.7 2.5 3.5 4,5
tl%[ I I I I ~ I i I II I 4 1 ] I I [ I I t II ;ll=1 [

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 method is also used to model the natural gamma
cross section (U) is in agreement with core ray data (CGR). This study is restricted to a 20
measurements and, so, might be used to predict m-long interval (200-220 mbsf/1.6 to 1.9 Ma).
the core data. Nevertheless, the vertical resolu-
tion of the downhole logs and the computed Numerical modelling
formation factor is rather poor in some zones,
especially in the upper part of the section. A new Modelling code. Resmod2D ~ is a newly devel-
forward modelling method is therefore proposed oped two-dimensional finite element numerical
in the following to improve the vertical resolu- code. It allows modelling of the response of
tion and derive more accurate Rt, CGR and FF electrical resistivity downhole probes, such as
profiles. the Spherically Focused tool (SFL). In brief, a
formation resistivity model composed of hor-
Forward model izontal sedimentary beds (layers) with fixed
thicknesses and resistivities, is entered in the
The aim is to obtain an accurate formation code in order to compute the response of the
resistivity model (Rt) from the numerical mod- probe in front of this formation. The resistivities
elling of the electrical resistivity log (SFL), of the model are referred to as 'true', whereas the
constrained by the high-resolution electrical computed resistivities (so the simulated response
images of borehole surfaces (FMS). A statistical of the tool) are referred to as 'apparent' because
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 125

in an inhomogeneous formation, it depends on made of 76 layers, the smallest one measuring 8


the resistivity of the bed next to the probe and cm and the greatest, 90cm. In order to validate
also that of the adjacent formations. The the models, several precision and robustness
measure of the resistivity by the SFL is therefore tests were run. For example, the error E between
lower than the true resistivity. the downhole logs and the computed theoretical
logs was computed using Whitman (1989)
Res&tivity modelling. Using Resmod2D :t:, a method:
formation resistivity model (Rt) is created on
the basis of cm-scale electrical images (FMS) for E(%) + 100 (10 El~ 1) (14)
bed thickness and m-scale electrical log (SFL)
for individual bed resistivity. The layer bound-
aries are determined from FMS images by (log(mi) -- log(ci)) 2
distinct colour contrasts; the gradual transitions E l o g = ~l Z (15)
are disregarded here. This initial model is then i=l,U N
processed with Resmod2D ~ to simulate the SFL
tool and compute a theoretical resistivity log where N is the number of records for the chosen
(SFLc). By comparison between SFLc and downhole log, and mi (respectively ci) is the
SFLm, the formation model is modified step value of the record i for the downhole log (for
by step (resistivity values and frame). Thickness the computed log). This estimated error is on the
changes as well as layer additions in the order of 1 % for the SFL, and 3 % for the CGR.
iterations are constrained by FMS images. This Due to the integration of high-resolution
process is iterated until the best fit between SFLc FMS results, the models (Figs 8b, 8d) have a
and SFLm is obtained. Each processing lasts much better resolution than the raw logs (Figs
about four hours for an evaluation every 5 cm 8a, 8c). Whereas the logs show essentially three
and over a 20 m-long interval (from 200 to 220 continental events between 200 and 220 mbsf,
mbsf). the models show the same main three events, but
also several small ones. For example, while the
Natural gamma ray modelling. A statistical SFL seems rather linear over the first five metres,
method is used to model the natural gamma- the Rt model brings out an alternance of minor
ray tool in an analytical manner. The aim is to troughs and peaks suggesting little changes in
determine the natural gamma activity of each lithology, such as clay content decreases corre-
layer defined in the resistivity model. This simple sponding to the troughs. Whereas other low-
method is based on an exponential attenuation resistivity units seem to be massive on the SFL,
of the gamma ray flux versus depth (Ellis 1987). they are composed of several fine layers in the
Using the lithological frame established for the model. The high-resistivity units, related to
resistivity model, and the natural gamma ray colder periods, are characterized in the raw log
measurements (CGRm), an initial model is by two or three regular peaks, whereas the
established to estimate the C G R (CGRc). By model displays a much more irregular profile.
comparison between CGRc and CGRm, the The major peaks are more pronounced and the
model is modified step by step and the chosen contrast is greater: the peak at about 211.15
model corresponds to the best fit between CGRc mbsf has a value of 0.58 f~ m for the raw log, and
and CGRm. 0.80 f2 m for the model. Moreover, the transition
from the low-resistivity interval upward into the
Results high-resistivity one is rather gradual for the raw
log, whereas the model seems to show that a
Formation electrical resistivity and gamma ray sharp boundary is present at the base of the
models. The chosen models, obtained after about high-resistivity unit, suggesting an abrupt initia-
80 iterations, correspond here to a near-perfect tion of each glacial period. All these features
fit between the measured and computed values revealed by the models are confirmed by core
(Figs 8a, 8c). The initial Rt model took into observations (Ingle et al. 1990): the vertical
account the major layers seen on the FMS lithologic variations within the dark/light cycles
images, not the discrete ones. The SFLc then are remarkably constant. The dark-coloured
presented the major variations of the SFLm but intervals are either thinly to thickly laminated
not the small ones. In order to reproduce these and finely bedded. These intervals generally
small events and to take the gradual transitions possess a well-defined and sharp base that grades
into account, the basic frame was refined by upward into light-coloured/high-resistivity sub-
adding small layers. Whereas the initial models layer (Ingle et al. 1990); the lower portion of the
were composed of 47 layers, the chosen ones are light-coloured intervals is commonly gradational
126 C. LAUER-LEREDDE ET AL.

and mostly obliterated by bioturbation. The for example the importance of cm-scale resistive
determination of the boundaries in the model beds on the response of m-scale logs. This case
seems therefore to be relatively accurate. study, which now requires additional measure-
ments on core to further improve the precision
Formation factor, The formation factor is of the method, could become a parallel method
computed for the modelled logs (Rt and CGR) to obtain meaningful physical properties and
on the basis of the method presented in the log palaeoclimatic data from future sites.
analysis section.
The results for the F F model are similar to This work was carried out by the main author with
those for the Rt model. The F F model has a financial support from the French 'Minist~re de la
better resolution than the F F computed from Recherche et de l'Enseignement Sup6rieur'. The
authors wish to acknowledge the contribution of GaYa
logs, as the former displays small events not
Entreprises (Marseille) for the use of the forward
detected by the latter (Fig. 8e). The major peaks modelling code Resmod2D". They also wish to thank
are also more pronounced: the peak at about S. Brower (LDEO) for providing the logging data, C.
211.15 mbsf has a value of 3.48 for the raw log, Robert (COM, Marseille) and the two anonymous
and 4.62 for the model. Changes between glacial reviewers for their helpful comments.
and interglacial periods also present the same
characteristics as the Rt model (sharp transition
References
from dark laminated interval upward into light-
coloured interval, and gradual transition from ARCHIE, G. E. 1942. The electrical resistivity log as an
light interval upward into dark one). aid in determining some reservoir characteristics.
Petroleum Transaetions of the AIME, 146, 54-62.
CAILLERE, S., HENIN, S., RAUTUREAU, M. 1982.
Conclusions Min{ralogie des argiles. 2. Classification et nomen-
clature. Masson, Paris.
The first results of the method using the raw logs CLAVIER, C., COATES, G. • DUMANOIR, J. 1977. The
show that the proposed mineralogical model is theoretical and experimental bases for the dual-
well representative of the sediment from ODP water model for the interpretation of shaly sands.
Hole 798B. The photoelectric absorption cross- Society of Petroleum Engineers 52nd Annual Fall
section (U) allows differentiation and computa- Technical Conference of AIME, Denver.
tion of the oceanic and continental fractions. DEMENOCAL, P. B., BRISTOWJ. F. & STERN,R. 1992.
The vertical resolution of the raw logs and of the Paleoclimatic applications of downhole logs:
computed formation factor curve being however Pliocene Pleistocene results from Hole 798B, Sea
poor in some zones, a new forward modelling of Japan. Proceedings of the Ocean Drilling
Program, Scientific Results, 127/128, 393407.
method is proposed in this paper to improve this DERSCH, M. & STEIN, R. 1992. Pliocene-Pleistocene
study. fluctuations in composition and accumulation
The m-scale electrical log (SFL) and the rates of Polo-marine sediments at Site 798 (Oki
natural gamma ray log (CGR) are modelled Ridge, Sea of Japan) and climatic change:
using cm-scale electrical images (FMS) to define preliminary results. Proceedings of the Ocean
and map high-resolution layers, The first results Drilling Program, Scientific Results, 127/128 (1),
show that the Rt, C G R and F F models are more 409-422.
precise than the Rm, C G R and FF obtained DREVER,J. I. 1982. The geochemistry of natural waters,
from log analysis, insofar as the models bring Prentice-Hall, New Jersey.
DUNBAR, R. B., DEMENOCAL, P. B. & BURCKLE, L.
out small layers not detected by the raw logs. 1992. Late Pliocene-Quaternary biosiliceous sedi-
The modelling approach allows the study of mentation at Site 798, Japan Sea. Proceedings of
changes from oceanic to continental supply the Ocean Drilling Program, Scientific Results,
hence from interglacial to glacial periods: the 127/128 (1), 439~455.
continental input tends to increase abruptly EKSTROM, M. P., DAHAN, C. A., CHEN, M.-Y., LLOYD,
from warm periods to colder ones suggesting P. M. & Rossl, D. J. 1986. Formation imagining
abrupt initiation of glacial cycle, whereas it with microelectrical scanning arrays. Transactions
seems to decrease gradually from cold to warmer of the Society of Professional Well Log Analysts,
27th Annual Logging Symposium, Paper 88.
periods.
ELLIS, D. V. 1987. Well logging/or earth scientists.
This study provides a continuous description Elsevier, New-York.
of changes in intensity of the different sedimen- FERTL, W. H. & Frost, E. 1980. Evaluation of shaly
tary sources within the analysed interval. While clastic reservoir rocks. Journal of Petroleum
FMS images reveal the presence not only of Teehnology, 31, 1641 1646.
large-scale layers but also of thin (short) events, FOLLMI, K. B., CRAMP, A., F(SLEMI, K. E., ALEXAN-
the numerical modelling enhances the non- DROVlCH, J. M., BRUNNER, C., et al. 1992. Dark-
linearity of m-scale logging devices, stressing light rhythms in the sediments of the Japan Sea:
FORWARD MODELLING OF OCEANIC SEDIMENT PHYSICAL PROPERTIES 127

preliminary results from Site 798, with some salinity, shaly sand reservoir. Journal of Petroleum
additional results from Sites 797 and 799. Technology, 27, 1204-1208.
Proceedings of the Ocean Drilling Program, KUKLA, G., HEELER, L., MING, X., CHUN, X. T.,
Scientific Results, 127]128 (1), 559-576. SHENG, L. T. & SHENG, A. Z. 1988. Pleistocene
GRIM, R. E. 1968. Clay mineralogy. Second Edition, climates in China dated by magnetic susceptibil-
Mac Graw Hill Book Company, New York. ity. Geology, 16, 811-814.
Guo, Z. T. 1990. Succession des paleosols et des loess du LAVERS,B. A., SMITS,L. J. M. & VANBAAREN,C. 1974.
Centre-Ouest de la Chine: approache micromor- Some fundamental problems of formation evalua-
phologique. PhD Thesis, University of Paris 6. tion in the North Sea. The Log Analyst, 12, 1-10.
HAGELBERG, T., SHACKELTON, N., PISlAS, N & The LiiTH1, S. M. & BANAVAR,J. R. 1988. Application of
Shipboard Scientific Party 1992. Development of borehole images to three-dimensional geometric
composite depth sections for Sites 844 through modelling of eolian sandstone reservoirs, Permian
854. Proceedings of the Ocean Drilling Program, Rotliegende, North Sea, American Association of
Initial Reports, 138, 79-85. Petroleum Geologists Bulletin, 72, 1074-1089.
HASSAN, M., HOSSIN, A. & COMBAZ, A. 1976. Funda- MATOBA, Y. 1984 Paleoenvironment of the Sea of
mentals of the differential gamma-ray log. Trans- Japan. In: OERTLI, H. J. (ed.) Benthos '83 2nd
a c t i o n S P W L A , 17th A n n u a l L o g g i n g International Symposium on Benthic Foraminifera,
Symposium, Paper H, 1-18. 409-414.
HENRY, P. 1997. Relationship between porosity, MORLEY, J., HEUSSER, L. & SARRO, T. 1986. Latest
electrical conductivity and cation exchange capa- Pleistocene and Holocene paleoenvironmnet of
city in Barbados Wedge sediments. Proceedings of Japan and its marginal sea. Palaeogeopraphy,
the Ocean Drilling Program, Scientific Results, Palaeoclimatology, Palaeoecology, 53, 349-358.
156. NEUMAN, C. H. 1980. Log and core measurements of
IMBRIE, J. J., HAYS, J. D., MARTINSON, D. G., oil in place. Journal of Petroleum Technology, 32,
MCINTYRE, A., MIX, A. C., MORLEY, J., PISIAS, 1309-1315.
N. J., PRELL, W. L. • SHACKLETON,N. J. 1984. PEZARD, P. A., LOVELL, M & The Ocean Drilling
The orbital theory of Pleistocene climate: support Program Leg 126 Shipboard Scientific Party 1990.
from a revised chronology of the marine oxygen- Downhole images: electrical scanning reveals the
isotopic record. In: BERGER,A., IMBRIE,J., HAYES, natur of subsurface oceanic crust. EOS, 71, 709
J., KUKLA,G. & SALTZMAN,B. (eds), Milankovitch SCHLUMBERGER 1994. Log interpretation charts.
and Climate, Part L Dorrecht, Holland, 269-305. Schlumberger Wireline & Testing, Houston.
INGLE, J. C., JR., SUYEHIRO,K., VONBREYMANN,M. T., SMITS, L. J. M. 1968. SP log interpretation in shaly
et al. 1990. Proceedings of the Ocean Drilling sands. Society of Petroleum Engineers Journal, 8,
Program, Initial Reports, 128. 123-136, Transaction of AIME, 243.
JACKSON,P. D., TAYLORSMITH,D. & STANFORD,P. N. TAYLOR SMITH,D. 1971. Acoustic and electric techni-
1978. Resistivity-porosity-particle shape relation- ques for sea-floor sediment identification. Pro-
ships for marine sands. Geophysics, 43, 1250- ceeding Symposium on engineering properties of
1268. sea-floor soils and their geophysical identification,
JOHNSON W. 1978. Effect of shaliness on log response. Seattle, Washington.
Canadian Well Logging Society Journal, 10, 29- WAXMAN, M. H. & SMITS, L. J. M. 1968. Electrical
57. conductivities in oil-bearing shaly sands. Society
JOHNSON, G. R. & OLHOEFT, G. R. 1984. Density of of Petroleum Engineers Journal, 8, 107-122.
rocks and minerals. In:: CARMICHAEL,R. S. (eds) - - & Thomas, E. C. 1974. Electrical conductivities
CRC Handbook of Physical Properties of Rocks. in shaly sands--I. The relation between hydro-
Boca Raton, FL (CRC Press, Inc.), 3, 1-38. carbon saturation and resistivity index; II. The
JUHASZ, I. 1981. Normalised Qv - - the key to shaly temperature coefficient of electrical conductivity.
sand evaluation using the Waxman-Smits equa- Journal of Petroleum Technology, 6, 213-225.
tion in the absence of core data. Transaction WINSAUER,W. O., SHEARtN,H. M., MASSON, P. H. &
SPWLA 22nd Annual Logging Symposium, WILLIAMS, M. 1952. Resistivity of brine-saturated
Paper Z. sands in relation to pore geometry. Bulletin of the
KERN, J. W., HOYER, W. A. & SPANN, M. M. 1976. American Association of Petroleum Geologists, 36,
Low porosity gas sand analysis using cation 253-277.
exchange and dielectric constant data. Transac- WHITMAN, W. 1989. Inversion of normal and lateral
tion of the Society of Professional Well Log well logs. The Log Analyst, 30, 1-11.
Analysts, 17th Annual Logging Symposium, ZHENt, H. H. 1984. Paleoclimatic events recorded in
Paper PP, 17 p. clay minerals in loess of China. In."PECSI, M. (ed.)
KOERPERICH,E. A. 1975. Utilization of Waxman-Smits Lithology and Stratigraphy of Loess and Paleosols.
equations for determining oil saturation in a low- 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 3

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 not k n o w the exact sub-bottom depth of the core
c o m m o n application of d o w n h o l e logs is to samples that are recovered, though we may
c o m p l e m e n t and calibrate measurements m a d e assume their relative original positions are
on cores and to fill in the gaps left by incomplete preserved. Thus, at the limit, a continuous
core recovery. Below the range of the Advanced 1.5 m section recovered from a 9.7 m core may
Piston Corer (typically about 200 m sub-bottom) have originally come from the top or the b o t t o m
core recovery is rarely m o r e than about 50% for 1 . 5 m interval. A g r i n i e r & A g r i n i e r (1994)
sediments and 40% for basement rocks ( O D P showed that the best estimate of the position
1990). D o w n h o l e logs make measurements at within finite limits of any arbitrary length of core
ambient temperatures and pressures and sense a sample is given by Euler's Beta distribution. This
volume a r o u n d the borehole greater than the can be given as a probability density function of
core itself. Such measurements provide a con- position in terms of the lengths of core, section
tinuous stream of in situ data on the wall rocks and the n u m b e r and positional order of core
and borehole fluids. sample. Therefore, d o w n h o l e logs play an even
The above figures for typical core recovery are more important part in filling the gaps in our
aggregate values. Our knowledge of the exact knowledge of rock sequences for which there is
depths of recovered core samples is, in general, incomplete core recovery.
worse than these figure imply. In the ODP, If we can identify the characteristic ranges of
coring advances in steps of about 9 . 7 m (the combined log values that correspond to different
length of individual core barrels). Unless core lithologies penetrated by the hole then we have a
recovery for this 9.7 m interval is complete we do means of assigning lithological class labels to the

WADGE,G., BENAOUDA,D., FERRIER,G., WHITMARSH,R. B., ROTHWELL,R. G. & MACLEOD,C. 129


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
130 G. WADGE E T AL.

total interval from the downhole logs. We show foraminiferal). This nomenclature is applied by
how this can be done for ODP data when we can the shipboard petrologists when the cores are
assign the lithological classes from those sections split and each 1.5m section of recovery is
where we do have 'complete' core recovery and described individually on the Visual Core
extend the classification to the full hole. Our Description sheets in freehand. This description
approach has been to develop software that may show some variation from one petrologist
supports a general user in completing this task. to another. From this initial detailed visual
There are a number of methods by which such description a more generalised sequence of
a classification scheme can be driven (e.g. graphical codes (from a total of about 50) are
discriminant functions, principal components assigned to each interval of core to denote a
analysis and cluster analysis; Doveton 1994). lithological class label (e.g. T6=sandstone;
We have chosen to use classifiers based on C34= foraminiferal chalk) so that a composite
artificial neural networks, principally because of graphical log (Barrel Sheet) of all the core
their ability to cope with complex non-linear recovery can be drawn for publication in the
problems. Also, neural network classifiers have Initial Report series of ODP publications.
been shown to be of value for lithological Beyond this the shipboard scientists can, and
classification of downhole logs in hydrocarbon sometimes do, erect other, non-standard but
exploration wells, in many cases with superior complementary, classification schemes, perhaps
results to other techniques such as discriminant based on local variation of the sediment. Within
analysis (Baldwin et al. 1990; Rogers et al. 1992; the ODP there is no official archived digital
Wong et al. 1995). Goncalves (1995) reports version of the lithological classification of core.
similar findings for ODP data. Hence working on core classification off the ship
This paper has three main sections. Firstly we requires digital recoding of the shipboard
present the main factors involved in the classi- scheme(s).
fication task. These are how the problem is The neural network approach is a supervised
defined, the constraints imposed by the data classification scheme. It requires that a sufficient
available and how to validate the results and number of training examples of the logs from
assess performance. Next we describe how we each separate lithological class be made avail-
have implemented the neural network method able for the algorithm to learn the character of
on our computer system. Finally, we present that class. Two general rules apply here. First, if
classification results from ODP Hole 792E. The there are too few samples within a class then
succession represented in this hole is a complex those samples will not fully represent the
sequence of island arc lithologies that is a good distribution of values that the classifier might
test of the general usefulness of the technique. meet through the whole borehole. Second, the
number of samples presented to the classifier
ODP lithological classification from each class should be approximately the
same. If the number of samples from one
The lithologies encountered in ODP holes are lithology presented to a neural network classifier
usually deep-sea sediments and oceanic crustal is much greater than for the other classes then
rocks and are generally distinct from those the network will tend to bias its classification in
encountered during drilling in sedimentary favour of this class. This problem becomes
basins underlain by continental crust. Deep sea serious whenever the logs do not provide a clear
sediments are often relatively unconsolidated separation of the lithological classes.
and rich in carbonates and/or silica or composed The quality of log samples used to train the
of terrigenous or volcanic detritus; they may be classifier is also important. In addition to
underlain by a basaltic basement. Porosity is checking for spurious outliers we use three logs
typically high, with ubiquitous saturation by for quality control. Samples exceeding any of the
sea-water. The classification framework of Maz- threshold values for the caliper, density correc-
zullo et al. (1987) is widely employed for tion and geochemical factor logs are not used for
sediments by ODP. The highest-level division is supervising or testing the classifier.
into granular and chemical sediments. Granular
sediments are subdivided into pelagic, neritic, Classification choice
siliciclastic, volcaniclastic and mixed sediments
and chemical sediments into carbonaceous, The choice of what classification to attempt is of
evaporites, silicates/carbonates and metallifer- vital importance. The ideal is to have a set of
ous sediments. Below this level classes are based lithological classes that best represents the
on principal names (e.g. ooze, chalk) together geological information required from the hole
with major or minor modifiers (e.g. nannofossil, and which produces distinctive responses in the
ODP LITHOLOGY USING NEURAL NETWORKS 131

suite of available downhole logs. The usual sub-populations are used to train and test the
starting point will be the principal-names level of performance of the classifier (e.g. discriminant
shipboard classification of the core. Classes with analysis) and the classification rates of the two
very few member samples may need to be classifiers can now be compared.
amalgamated with other classes. The lithological
information required from the hole may not be Implementation of a neural network method
the most obvious. For example, a hole may
penetrate a succession of oozes above siliciclastic System design
rocks lying on a basaltic basement. Classifying
such a tripartite division should be trivially easy The quite complex processing chain implicit in
and the real problem of interest may be in the the above discussion is best handled by a
second-order variability, say, distinguishing vol- computer system designed for the job. We have
canic from non-volcanic rocks in the siliciclastic designed such a system, the essential elements of
sequence. In this case the classification task can which are shown in Fig. 1. The computing
be constrained by choice of depth interval. platform is a Sun Sparcstation and the graphical
Alternatively, the need to change the class user interface is designed using PV-WAVE
labelling given to the core samples to best fit visualization software. There is a separate
the problem may only become apparent after an development environment for designing the
initial attempt at classification. Merging and neural networks that the user does not see, but
splitting of classes may be required. There is a which can create portable networks (as C code)
clear general need for more than one classifica- that can be retrained. The user must define the
tion scheme to be tested and for the editing problem by choosing appropriate depth inter-
facilities to support that need. vals, lithological classes, logs and a neural
There is no guarantee that the recovered core, network. The results of running the network
and hence any classification scheme based on it, are displayed graphically and in terms of relative
is fully representative of the lithologies in the performance of the classification rate. There are
missing intervals. Examples of preferential three main functional components to the system.
recovery of, say, clays relative to sands are These are shown in Fig. 2 and are described in
well-known. At the extreme a relatively common detail in the following sections.
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 explora-
tory 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.

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 sub-
population of core-classified samples from the
same general population and classify it indepen-
dently 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 Fig. 1. Schematic structure of our computer system to
another classification technique. Again, the same derive lithological logs from ODP core-log data.
132 G. WADGE ET AL.

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, nomin-
ally 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) sedimen-
tary bed would be of uniform lithological
character with sharp boundaries and have
contrasting neighbours giving the logs the
Fig. 2. Functional schematic of the way data is character of step functions across the bound-
explored, selected and classified in our method. aries. This ideal is the basis of log segmentation
algorithms (e.g. Vermeer & Alkemade 1992)
Log data exploration which seek to segment the borehole into uniform
intervals to which single (lithological) labels can
The purpose of log data exploration is to enable be attached. This can be helpful in constraining
the user to become familiar with the data before classes in intervals of incomplete core recovery.
making explicit choices. The ability to plot logs The data selection process creates a Log-Class
against depth and cross plots of one log against File (Table 1), whose values are used directly by
another is standard. In our system there is also the neural network classifier.
the ability to display simultaneously, selected
sample populations both in terms of their depth Neural network classification
and their position in log space. This is done by
manipulating a graphical cursor. It is a valuable Samples from the Log-Class File are separated
facility for deciding whether some 'extreme' into class populations and counted. The total of
values in a cross plot, say, correlate with a the smallest population is then used to select
specific bed or are scattered throughout a samples for training and testing. Classes with
sequence. Principal components analysis and larger p o p u l a t i o n s are subsampled evenly
unsupervised cluster analysis are also available throughout their range to give a total equal to
for any selection of samples and logs. These that of the smallest class. Each same-size class
exploratory analytical functions help decide the population is then split into two sub-populations
following: by alternate sampling to give training and testing
sample populations for each class. The log
(1) what is the effective dimensionality of the values of these populations are then examined
log data (i.e. how many distinct classes will graphically to check that; the distributions of the
the data support)?; training and testing samples are similar, and that
(2) at what depth intervals do the most the distributions of the core-classified samples
representative samples lie?; are representative of the whole interval under
(3) are any samples obviously not represented investigation. If these conditions are not met
by core intervals with good depth match- then other data selections must be made.
ing? The neural network used is the feed-forward
back-propagation type which is standard for
Data selection classification problems. The selection of logs and
classes constrains the structure of the network.
Explicit selections of depth interval, logs and The input layer consists of one node for each log
classes must be made. As we show later, the 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 Clay Sst Cong 1 Cong2

26.6793 24.7285 21.4907 1 0 0 0


26.8207 25.1096 20.9049 1 0 0 0
26.6378 25.287 - 20.0691 1 0 0 0
12.6669 9.3789 - 23.0037 0 1 0 0
12.1269 8.8892 - 21.9625 0 1 0 0
11.7988 8.7609 21.4303 0 1 0 0
12.1286 8.1325 - 19.5016 0 0 1 0
11.8547 7.9513 - 18.8318 0 0 1 0
11.7594 7.9992 17.9526 0 0 1 0

each class. There is at least one other, 'hidden', outcome of the training and testing cycle is
layer of nodes between the input and output considered satisfactory then the network can be
layers with weighted connections between nodes applied to the full interval under consideration.
of different layers. The number of hidden nodes
is not externally constrained and can be changed
to suit a particular problem. The other network Application to Hole 792E
parameters are rnainly related to the weightings
applied to the connections. These can be tuned Hole 792E of the ODP was drilled in the Izu-
to improve performance, but detailed optimiza- Bonin foreare sedimentary basin in 1989
tion of neural networks is a complex issue. Our (Taylor et al. 1990). The primary objective of
strategy is to make available a limited number of drilling at Site 792 was to understand the
default networks initially and optimize indivi- stratigraphy of the forearc and the temporal
dually applied networks later. variations in sedimentation and volcanism that
Once the network has been selected it begins controlled it. About 800m of sediment were
training with the prepared training data by drilled above volcanic (andesite) basement,
selecting samples, at random, and presenting including rocks of Pleistocene, !ate Pliocene,
their log values to the input layer of the network. Miocene and early Oligocene age. The lower
The effect of these values propagates through to part of this section is a more volcaniclastic-rich
the output layer where the 'error' between the sequence. The interval between 482-732 m below
network node values and the 'correct' values is sea floor (mbsf) is the focus of this study and
then propagated back through the network, comprises a large part of Unit IV, a sedimentary
thereby changing network weightings. In this succession of early Oligocene age (Fig. 3).
way the network, after hundreds to thousands of Overall, Unit IV is composed of vitric sandstone
learning cycles, improves its ability to recognize (58%), sandy pebble-granule conglomerate (
classes until no further improvement is achieved 10%), silty claystone (9%), nannofossil-rich silty
and the network is said to be trained. The claystone (5%), claystone (4%), siltstone (4%),
trained network can now be tested by presenting nannofossil c l a y s t o n e ( l % ) , clayey siltstone
each of the test samples to the network once, (1%), sandy mudstone (1%) and sandy siltstone
and recording how close the network result is to (1%). This unit is interpreted by Taylor et al.
the correct classes. The network gives propor- (1990) as a rapidly deposited turbidite blanket in
tional values for each class within the range 0-1, an oversupplied basin or distal fan. The pumice
whereas the core-derived class labels are binary clasts of some of the coarse sandstones and
(0 or 1). The network results are thus essentially conglomerates indicates contemporary volcan-
probabilistic. We express the test result as a ism but most of the andesite and dacite clasts are
thresholded classification rate. For example, at a probably the result of erosion from the arc
user-defined threshold of 70% probability, a test volcanoes.
result of 0.76 sandstone, 0.20 siltstone and 0.04
claystone for a sample with class values of 1,0,0 Core recovery
would count as a correct classification, though
not at a threshold of 80%. 92 such correct Hole 792E had a recovery of 48.2% of total
results out of 100 test samples, for example, possible core length. Recovery for the 482-732
would give a classification rate of 92%. If the mbsf interval was 79.1%. However, as discussed
134 G. WADGE E T AL.

Downhole 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), Geo-
chemistry (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 gam-
ma, radioactive potassium, thorium and ura-
nium, deep, medium and shallow resistivity,
density, photoelectric effect, sonic velocity and
the oxide contents of calcium, silicon, iron,
titanium, potassium and aluminium. Unfortu-
nately, no density correction and photoelectric
Fig. 3. Summary graphical log of the stratigraphy of effect measurements are available below about
the 482-732 mbsf interval of ODP Hole 792E (after 550 mbsf. We removed low-quality log data that
Taylor et al. 1990). The symbols represent: Inverted
'T'= nannofossil ooze, dashes=claystone-siltstone, exceeded any of the following thresholds for the
check = sandstone, dot-ellipse=conglomerate. Grain three quality-control logs: Caliper (29.5cm
sizes are represented by c=claystone, s=siltstone, Density Correction (0.1 gm cc l) and Geochem-
fs = fine sandstone, cs = coarse sandstone and g = grav- ical Factor (800). These thresholds were deter-
el/conglomerate. mined empirically by e x a m i n i n g the log
distributions. This reduced the number of
samples available by about 3%.
earlier, many individual cores contained only a Within the 482-732 mbsf interval Taylor et al.
fraction of the full core length and hence cannot (1990) and Pratson et al. (1992) observed the
be matched to the logs with confidence. We have following relationships between log and core:
used only individual cores that have greater than
90% recovery to form the basis of our classifica- (1) Natural gamma spectrometry shows that
tion. These 13 out of 26 cores (Fig. 3) therefore potassium is the dominant radioactive
represent a useable recovery of 50%. Each of source mineral and is inversely related to
these cored intervals have had their depth values of resistivity, velocity and density,
assignments normalized to 100% recovery after and in some cases, grain size (482-500 and
closing any gaps between core sections. Notice 555-585 mbsf). Uranium and potassium
that we have no matched core for the interval contents are generally negatively corre-
617-655 mbsf. lated. Beds with high mud contents have
Five lithological classes based on shipboard high values of potassium content and
visual description were used in the 482-732 mbsf natural gamma.
interval: claystone, silty/sandy claystone, muddy (2) Above about 515 mbsf the sequence has a
siltstone/sandstone, siltstone/sandstone and bimodal character with high resistivity/
gravel/conglomerate. One of these five classes velocity/density - low natural gamma beds
was initially assigned to each (0.15m) log alternating with beds of opposite character.
interval. A number of thin (< 0.15 m) , mainly Below this depth the logs lose their high
claystone, beds were ignored in this process. frequency nature and generally have raised
However, the claystone and muddy siltstone/ resistivity/velocity/density values.
sandstone classes had so few samples (8 and 37, (3) There is correlation between upward-fining
respectively) that they were both merged with sandstone/conglomerate beds and a saw-
the silt/sandy claystone class. For simplicity the tooth response of resistivity in the 540-590
remaining three classes were called claystone, mbsf interval. Of particular importance is
sandstone and conglomerate. the reduction in resistivity at 587 mbsf,
where the base of a large conglomerate bed
ODP LITHOLOGY USING NEURAL NETWORKS 135

Table 2. Networkperformance results


Network Network Depth Classification Rate(%)
Number Structure Interval
(mbsf) Total Clay Sst Cong (1) Cong(2)

1 1-717-3 482-550 88.4 87.0 78.3 100


2 15-15-3 482-550 85.5 87.0 73.9 95.7
3 15-15-3 482-732 75.8 83.5 59.3 84.6
4 15-15-4 482-732 90.0 92.0 80.0 87.5 100
5 4-8-4 482-732 78.1 85.0 47.5 82.5 97.5
6 5-15-4 482-732 80.0 82.5 52.5 85.0 100
7 6-15-4 482-732 81.9 87.5 67.5 72.5 100
8 11-10-4 482-732 86.3 85.0 65.0 95.0 100
9 15-02-4 482-732 71.9 87.5 92.5 12.5 95.5
10 15-04-4 482-732 82.5 90.0 47.5 92.5 100
11 15-06-4 482-732 83.1 85.0 60.0 87.5 100
12 15-08-4 482-732 86.1 87.5 85.5 75.0 100
13 15-10-4 482-732 86.9 90.0 62.5 95.0 100
14 15-12-4 482-732 86.9 90.0 72.5 85.0 100
15 15-14-4 482-732 88.1 85.0 72.5 95.0 100
16 15-16-4 482-732 86.9 92.5 70.0 85.0 100
17 15-18-4 482-732 89.4 92.5 72.0 95.0 97.5
18 15-20-4 482-732 88.1 90.0 77.5 85.0 100
19 15-25-3 482-732 78.0 86.8 57.1 90.1
20 15-30-3 483-732 76.6 79.1 69.2 81.3
21 15-25-10-3 482-732 80.2 84.6 72.5 83.5
22 15-25-20-3 482-732 78.8 82.4 65.9 87.9

also marks the d o w n h o l e increase in There are 17 logs available for the interval
smectite concentration and magnetic sus- 482-550 mbsf, but only 15 (no density and
ceptibility. photoelectric logs) were considered for the full
interval between 482-732 mbsf. Using the basic
Performance of different neural networks rock type classification, the 482-550 mbsf
interval has 98 claystone, 164 sandstone and 46
There is no one single neural network that will conglomerate samples available for training and
work for all classification problems. The basic testing. Network 1 (Table 2) then, is a 17-17-3
type used here, whose structural variants we now network used for 482-550 mbsf that gives a total
discuss, is a back-propagation network with a classification rate of 88.4%. Using only 15 logs
single hidden layer and a sigmoidal activation for the same interval in network 2 gives a
function at each node. As part of the pre- reduced rate of 85.5%. Hence density and
processing for neural network training the data photoelectric effect logs do have some extra
from each log that are input to the network are capability to discriminate between these rock
normalized to the range 0-1. Logs with large classes in addition to that present in the other 15
potential ranges, such as the resistivity tools, logs. However, when we use the same type of
may be best converted to a logarithmic scale network for the 482-732 mbsf interval, but
first. However, the resistivity values of Hole trained using the samples from a 183 claystone,
792E did not show a very great range and this 407 sandstone and 185 conglomerate pool of
was not done. As was discussed earlier, the samples (network 3), the classification rate falls
number of nodes in the input and output layers to 75.8%. This means that the claystone-
is at least partly determined by the nature of the s a n d s t o n e - c o n g l o m e r a t e classification that
data and the problem to be solved; the number worked well from 482-550 mbsf is much less
of nodes in the hidden layer is chosen to appropriate below 550 mbsf. There is a distinct
optimize performance once the input and output fall in the capability of the network to recognize
layers are fixed. In the performance results the conglomerate and sandstone class samples.
reported in Table 2 we use a convention in Exploring the log data over this wider interval
which a 15-10-4 network means 15 input nodes suggests that core samples classed as conglom-
(downhole logs), 10 hidden nodes and 4 output erate may be usefully split into more than one
nodes (lithological classes). type. For example, principal components analy-
136 G. WADGE E T AL.

100 60
x

90 9 50

80 40
~
o
x
x
=

O
i
70 30
9 Classification Rate
x Computing Time

60 ! n ! i I l 20
2 4 6 8 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 9 m ra

86

84

82

80
o

78

76
.--4

74
72
70 9 i 9 i 9 i 9 i 9 i , i 9 i 9 1 9 i 9 i

0 2 4 6 8 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 erate 2. Conglomerate 1 includes those samples
gamma, computed gamma and potassium oxide classed as conglomerate in networks 1 and 2;
logs shows that some of the conglomerate class conglomerate 2 (the samples described above)
samples give much lower values of principal corresponds to core which is conglomeratic but
component 3 and higher values of principal has higher proportions of large ( > 5 cm) clay-
component 4 than the other conglomerate stone clasts than conglomerate 1. In the logs,
samples. Thus we have created a second conglomerate 1 has distinctly lower spectral
classification scheme with four classes: clay- gamma and potassium values than those of
stone, sandstone, conglomerate 1 and conglom- 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 justification for the 15 nodes in the hidden layer
distinct. With four classes, network 4, otherwise is provided by a systematic test of performance
equivalent to network 3, gives a much improved in a series of networks with variable numbers of
classification rate for the whole 482-732 mbsf hidden nodes (networks 4, 9 to 18 inclusive; Fig.
interval of 90.0%. 5). The optimum configuration of the hidden
We saw an example of improvement in layer of 15 nodes is given by the maximum
performance with increased numbers of logs performance value. In this case the number of
(from network 1 to 2) for the 482-550 mbsf hidden nodes equals the number of input nodes.
interval. For the full interval with 4 classes it is This would be a useful rule-of-thumb for initial
also generally true that the more logs input to network configuration but it does not guarantee
the network the better the performance (net- the optimum solution. For instance, as can be
works 4, 5 to 8 inclusive; Fig. 4). There is a seen from networks 3 and 19 to 22 inclusive,
penalty to pay in terms of increased computing greater numbers of hidden nodes (and even a
time (Fig. 4) but this is not too great a burden. second hidden layer) can give improved perfor-
Hence it makes sense to use as many useful logs mance, but at the cost of increased computing
as are available at the outset. time.
Network 4 has a 15-15-4 structure. The
138 G. WADGE ETAL.

Computed lithological log results spondence is more difficult to assess. For some
cores such as 39 and 55, correspondence could
Figure 6 shows the output of network 4 when be achieved by appropriate expansion of the
applied to the whole 250 m interval of downhole core lithology down section to match the
logs from 482-732 mbsf. The aggregate thick- network lithological log results. For other
nesses of the lithologies according to this intervals no such a c c o m m o d a t i o n can be
classification are: c l a y s t o n e = 4 4 m , sandsto- achieved (e.g. core numbers 45, 52 and 59).
ne = 130 m, conglomerate 1 = 42 m, conglomer- The two main possible explanations of this are
ate 2 = 34 m. that the network is 'overtrained' and has lost its
Some sample intervals are classed as 100% of ability to generalize when exposed to new data,
a particular lithology but many are classed as and that the classification scheme is not optimal.
mixtures, with one dominant lithology and one The way ahead in our system would be to
or more minor components. This is particularly explore the second possibility by choosing
noticeable for mixtures of sandstone and con- another classification scheme. Because the sand-
glomerate 1, but much less so for mixtures stone group of samples is the largest and shows
involving the other two lithologies. In Fig. 7 the poorest general classification performance
the results from this network have been recast figures (Table 2) this is the most likely group for
such that the major ( > 5 0 % ) lithology in the possible splitting into two or more classes. Some
output is attributed to that depth interval. This of the geochemical logs such as TiO2 and CaO
gives a columnar plot that mimics a traditional show clear evidence of alternating high and low
Lithological log and allows direct comparison valued sandstone horizons (e.g. 670--682 mbsf)
with the visual core classification. that could form one of the criteria of such a new
The change in character of the rocks at about classification scheme, though this is not pursued
515 mbsf noted above is apparent but is over- here.
shadowed by the decrease in claystone which
occurs about 20 m higher. Lower down the hole
the occurrence of conglomerate 2 is restricted to Discussion
3 zones where there is an apparent association of
claystone-conglomerate 2. The base of the We have chosen to ignore a number of major
shallowest of these zones corresponds to the issues of core-log-driven classification including
major break noted at 587 mbsf, though there is graded bedding, the differences in spatial resol-
no obvious change in general lithological char- ving power of the logs and the use of segmenta-
acter below this. The log gives the impression of tion, in order to emphasize the value of quality
three major cycles of mixed conglomerate and control of the data and careful consideration of
sandstone sitting above conglomerate 2 and the optimal structure of the neural network. In
claystone (515-587, 587-660, 660-732 mbsf). particular, we wish to stress the need to have a
The second thickest interval classed as conglor- flexible mechanism for changing the classifica-
nerate 2 is from around 640 mbsf, where there is tion scheme of rock types based on the
no core recovery. In fact there is no sense of this information content in the logs and the ship-
second cycle in the recovered core. The lowest of board-derived classification scheme. Such an
the three cycles is richer in sandstone at the approach is inherently hole-specific. It lies at
expense of conglornerate 1. an intermediate position between a totally
As discussed earlier, the rocks recovered from empirical approach, driven solely by the log
only 50% of the full 482-732 mbsf interval were data, and one that might use a universal library
used to train and test the neural network. Figure of log responses derived from fundamental core
7 displays these core (core numbers 37-39, 40, components (e.g. sand, carbonate, sea-water
42-43, 46, 48-50, 56-57, 60, 61) together with etc.). If the data can support it, the refinement
the additional recovered core, that comprised of the classification scheme in a hole will be
29% of the total interval, that was not used in essentially hierarchical. However, there is no
training and testing the network because of poor guarantee that the way that the log data can be
depth control (core numbers 39, 41, 44-45, 47, optimally divided will correspond to the classi-
51-55, 58-59, 61). For many of the intervals used fication scheme that the geologist wants or
in the training and testing there is a high degree expects. This sort of approach should be of
of detailed correspondence between the core and value to ODP scientists both on and off the ship.
the network classifications (e.g. core numbers Results of our work using data from other ODP
37-38, 50, 62) as we would expect. For a few, the holes will be presented elsewhere. We also
correspondence is weaker (e.g. core 43). For the envisage that this technique might be of value
core intervals with < 90% recovery any corre- 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.
140 G. WADGE ET AL.

piston cores from which horizontal-track logs GONCALVES, C. A. 1995. Characterisation of formation
have been collected o n - b o a r d ship. heterogeneity. PhD Thesis, University of Leice-
ster.
MAZULLO, L, MEYER, A. & KIDD, R. B. 1987. A new
sediment classification scheme for the Ocean
This work is funded by a grant (GST/02/993) to RBW
Drilling Program. ODP Technical Note, 8.
and GW under the NERC Special Topic--UK ODP
ODP 1990. Wireline Logging Manual, Ocean Drilling
Science Programme. ESSC work is supported by
Program. Borehole Research Group, Lamont-
NERC grant F60/G6/12/02. We are very grateful to
Doherty Geological Observatory.
our collaborators Drs P. Harvey and H. Grubb, for
PRATSON, E. L., REYNOLDS, R., LOVELL, M. K.,
their help and the Borehole Research Group at LDEO
PEZARD, P. A. & BROGLIA,C. 1992. Geochemical
and ODP/TAMU for supplying data.
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.
References ROGERS, S. J., FANG, J. H., KARR, C. L. & STANLEY,D.
K. 1992. Determination of lithology from well
AGRINIER, P. & AGRINIER, B. 1994. A propos de la logs using a neural network. American Association
connaissance de la profondeur a laquelle vos of Petroleum Geologists Bulletin, 76, 731-739.
echantillons sont collectes dans les forages. TAYLOR, B., FUROKA, A. & OTHERS 1990. Proceedings
Comptes Rendus de la Academie Sciences de Paris, of the Ocean Drilling Program, Initial Results, 126.
318, serie II, 1615-1622. VERMEER, P. L. & ALKEMANDE, J. A. H. 1992.
BALDWIN, J. L., BATEMAN,A. R. M. & WHEATLEY,C. Multiscale segmentation of well logs. Mathema-
L. 1990. Application of neural networks to the tical Geology, 24, 27-43.
problem of mineral identification from well-logs. WONG, P. M., JIAN, F. X. & TAGGART, I. J. 1995. A
The Log Analyst, 3, 279-293. critical comparison of neural networks and
DOVETON, J. H. 1994. Geologic log analysis using discriminant analysis in lithofacies, porosity and
computer methods. Computer Applications in its permeability predictions. Journal of Petroleum
Geology, 2. American Association of Petroleum Geology, 18, 191-206.
Geologists, Tulsa.
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 tory measurement of static and dynamic
like shear wave velocity (Vs), permeability (k) elastic constants as a calibration reference;
and elastic constants (Young (E), shear (G) and (3) obtain empirical correlations, for each well,
bulk (K) modulus) have been the subject of between Vp, k and q~, thereby yielding a
various works in geophysics (Wendt et al. 1986; calculated permeability log.
Castagna et al. 1993; Bastos et al. 1995; Tang et
al. 1996). In this paper we present, for three Methodology
Brazilian offshore wells, a generation procedure
for Vs, k and elastic constants logs calculated Ultrasonic P and S wave velocities were mea-
from laboratory data: Vp, Vs, porosity (~), (k) sured in about 120 samples of limestone from an
and static and dynamic elastic constants on offshore Brazilian field. These samples were
cores. The importance of the generation of retrieved from three vertical wells at depths of
unmeasured in situ logs includes the possibility about 2350m to 2550m and vertically cut as
of obtaining more accurate information about right cylindrical plugs with diameter 2.5 cm and
lithology and fluid content in reservoir rocks 3.75 cm and length 3.75 to 5cm. The measure-
and, in this way, contributing to generating ment frequency was 500 kHz for both Vp and Vs
more reliable AVO and seismic models, and also and over a range of confining pressure of 1000
optimizing drilling and hydraulic fracturing psi to 5000 psi at room temperature. The
programes. For reservoir development, these porosity and permeability range were 5% to
kind of data are also helpful for generating 35% and 0.1 mD to 1800mD, respectively. The
correlations between seismic attributes and same measurements were made under dry and
petrophysical properties and for monitoring formation water saturated conditions. However,
subsurface fluid flow. So, our main goal in this the results showed only small variations due to
work was to: saturation, as noted by Bastos et al. (1995).
Simultaneous measurements of static and
(1) obtain empirical correlations between Vs dynamic elastic constants were made on some
and Vp from laboratory data in order to samples of diameter 5cm and length 12.5cm.
generate unmeasured Vs logs from mea- These samples were placed in a triaxial cell and
sured Vp logs; subjected to an in situ confining stress of about
(2) generate logs of static and dynamic elastic 5000 psi, and to a deviatoric stress which was
constants using the simultaneous labora- 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 141
& 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
142 A . C . BASTOS E T A L .

4000
I tion related to the increasing deviatoric stress
allowed us to determine the static constants. The
Vm - 0 , 5 5 V p + 4 1 , 6 0 c c " 0,96 dynamic constants are obtained simultaneously,
by monitoring changes in transit time.

Procedure and results

I Calculated logs o f Vs and Elastic constants


The three wells that are the subject of this work
do not have in situ Vs logs. Therefore, a
laboratory relationship was obtained between
F
Vs and Vp in order to generate a pseudo Vs log.
1000 j ~ I ,
Figure 1 shows the linear fit to the Vs-Vp cross
2OOO 4O0O 600o plot (equation 1). As shown in this figure, an
Vp (m/s)
excellent correlation was obtained with a corre-
lation coefficient of 0.96. For the case of these
Fig. 1. Vs vs Vp for limestone plugs. samples, this linear fit was better than the

WELL - A WELL - B WELL - c

Velocity (m/s) Velocity (m/s) Velocity (m/s)


1000 2000 3000 4000 5000 6000 1000 2000 3000 4000 5000 6000 1000 2000 3000 4000 5000 6000
2380

2390

2400

2410

242O

2430

2440
A
E
.....

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 143
2O
' I ' I ' I ' ' I ' I ' I '

(A) (B) (c) /

..y
6O

i
,-. 4O ,..,.

o (P

f "
I11 9 _
20
2O

"2d,:, , , , , J I i
0 0 0
20 40 60 80 20 40 60 80 0 2O 40
E d y n (GPa) K d y n (GPa) G d y n (GPa)

Fig. 3. Static and dynamic elastic constants for sedimentary rocks obtained from simultaneous laboratory
measurements.
WELL - C

O (GP=) K (GPa) E (GPa)


0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 70 80 0 10 20 30 40 50 60 7O 8 0
2390 v } i
, , i ~ i , i , i
I E t I I - g_%, , , I -

2400 I b

- i ~ 1 I I - 1 I I I
2410 9 I I I t

I I t I -

2420 , ', I I
I I I 1 I I -

2430 I l , i I

I I I I I I -

2440 i" i I I I I
I I I t I t -

•. 2450 I 1 I I J i
i I I I I _

I I I I i t I I i I
I~ 2460
I I I I I
I I I I I I
2470
I I I 1
,i,},,
I I I I i :l ~ I I
2480
~ '' t I I -! I ,~ I I t
I I I
2490
I I I I I

b ~ ~ L i I I I ~ )l L I
2500
I I I !, ~, , ,
, I I I N ~1 i i
2510

2520
L

,k, >
L

I,I,
| I

i,l,I,
I I

, ,
)I

,t,t,
I I I

I,

Fig. 4. Calculated static and dynamic elastic constants logs for weU-C.

polynomial fit proposed by Castagna et al. 2). There is good agreement between pseudo-
(1993) even for values of Vs close to 1500ms -1. logs and laboratory data. With the Vp, Vs and
The regression algorithm is: density (p) logs and the following elastic theory
equations:
Vs = 0.55 Vp + 41.60 (1)
+ 4#/3
Using this relationship and the in situ Vp logs
v~ = d
/ K (2)
v Pb
it was possible to calculate Vs pseudo-logs (Fig.
144 A.C. BASTOS E T AL.
WELL - A

6000F . i ' i ' J ' i ' I '


1o L , I ' 0116 '10 I ' I ' I '/,
I , Vp - 5868 --e
~176 cc ,, 0.88 8 ~ K'~0"05e ;cc=0"90 9 /

5000 I

4000
j ../
3000

2000 1 I ~ I , I , I , I ,
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Porosity (%) Porosity (%)

WELL - B
6000 I ' I ' i ' r ' I ' 5 -- ' I ' I ' I ' I ' I
-o.026
t4 e ; cc ,, 0.93 0.179

~o~
4 K" 0.024 e ; cc = 0.77
5000 -

,ooo t-
1
9 9 9

2000 1 , I , I J I ~ I , I ,

5 10 15 20 25 30 35 5 10 15 20 25 30 35
Porosity (%) Porosity (%)

WELL - C

..
6000 ' i ' 1 i ' i ' i 2000 ' I ' I ' r ' I ' T

B 0.4 6 El
V p =, 5 8 2 2 e0.021 ; c c ,, 0 . 8 9 K = 0.0002 e ; cc =0.93
5000

.~4000
D,.

3000
i i000 9

2000 , I , I , I , I , I J
5 10 15 20 25 30 35 5 10 15 20 25 30 35
Porosity (%) Porosity (%)

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

~/~ obtain logs of the static and not the dynamic


V, = # (3) elastic constants. For this purpose, we use
simultaneous laboratory measurements of static
where: K is the bulk modulus, # is the shear and dynamic constants in order to transform the
modulus and Pb is the Bulk density, pseudo-logs dynamic to the reference static. Figure 3 shows a
of dynamic elastic constants have been calcu- cross plot between static and dynamic constants
lated, for sedimentary rocks and illustrates strong
In fact, to optimize drilling and/or hydraulic empirical relationships which are expressed
fracturing programs, it is often necessary to mathematically as follows:
CORE-DERIVED COMPUTATION OF PSEUDO-LOGS 145
WELL - A WELL- B WELL-C
Permeability (mD) Permeability (mD) Permeability (mD)
0 1 10 100 o 1 10 o 1 10 100 10o0 10000
2380 [ 1 1 1 1 ~ I I1111llI I Illllll~ I Illllll

2390

2400

2410

2420 I
2430

2440

2450
I
2460

g
r,,
2470

2480
_s
Q 2490

2500

2510

2520

2530

254O

2550

2560

2570

2580 J , ,,,,id L ,,,,,,,I , ,,,


v4
~,,,,,,l q~,,,,d ..... ,.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 laboratory permeability, porosity and velocity
coefficient --- 0.95 (4) data. Figure 5 shows core data, the cross plots of
velocity against porosity, and permeability
Kstat = 0.992 Kdyn -- 8.82; correlation against porosity. From these plots it has been
coefficient---- 0.89 (5) possible to deduce a relationship between Vp, ~b
and k. As can be seen in Fig. 5, an exponential fit
Gstat = 0.621 Gdyn -- 0.95; correlation was the best one obtained for both the Vp-q$ and
coefficient = 0.94 (6) the k-~b relationships for the three wells. Thus,
with the equations obtained (equations (7), (8),
where the subscripts 'stat' and 'dyn' denote (10), (11), (13) and (14)) we can isolate qb from
static and dynamic moduli, respectively. Vp-q) and k-q5 relations and then obtain k-Vp
relationships (equations (9), (12) and (15)) which
Figure 4 shows the calculated log of static and can be used to calculate the k-log shown in Fig.
dynamic constants obtained from equations (4) 6. In order to check these relationships we
to (6) for well C. As expected, the logs of the include some points in well A (cross points in
dynamic elastic constants show higher values Fig. 6) which were not used to obtain equations
than their static equivalents. (7) to (15). Again, it can be seen that there is a
good correspondence between these points and
Calculated logs of permeability the obtained log:

The next step was to calculate permeability logs Well A


for the three wells using algorithms based on Vp = 5868e~~ cc = 0.88 (7)
146 A.C. BASTOS E T AL.

k = 0.05e~ cc = 0.90 (8) of this relationship give us the static mechanical


behaviour, characteristic of production engi-
k : Vp 73.e6~ (9) neering, but, with the continuous character of
geophysical logs.
Well B (4) Cross plots between Vp-qb and k-~b
Vp = 6214e-~176 cc -- 0.93 (10) indicated good exponential fits for the three
wells that formed the subject of this work;
k = 0.024e~ cc -- 0.77 (11) (5) The good experimental relationships ob-
tained between k-qb and Vp-qb (see correlation
k = g p -6"54. e 53"35 (12) coefficients in equations (7) to (15)), together
with available logs give us an additional method
Well C to estimate permeability.
Vp = 5822em~ cc = 0.89 (13) (6) There is good agreement between labora-
tory permeability measurements and synthetic
k = 0.0002e~ cc = 0.93 (14) permeability logs from velocity data, even for
points that were not used in the generation of
k : Vp-184.e152.5 (15) these pseudo-logs.

Figure 6 shows a good correspondence for References


wells A and C, but not for well B. In this well the
good correlation coefficient for the k-qb relation- BASTOS, A. C., DILLON, L. D., SOARES, J. A. &
ship, 0.77, was lower. VASQUEZ, G. F.. 1995. Estimativa dos perils de
constantes elfisticas em carbonatos pouco per-
mefiveis a partir de dados laboratoriais. 4th
Conclusions International Congress of the Brazilian Geophy-
sical Society and the 1st Latin American Geophy-
(1) The large number of limestone samples sical Conference. Volume II.
gave us a precise empirical relationship between CASTAGNA, J. P., BATZLE, M. L. 8r KAN, T. K. 1993.
Vs and Vp for limestone, and this differs from Rock Physics: The link between rock properties
the earlier work of Castagna et al. (1993), even and AVO response. In: CASTAGNA, J. P. t~
for Vs close to 1500 ms -1. BACKUS,M. M. (Eds) Offset-dependent reflectivity:
(2) Good relationships between static and SEG, 124-157.
dynamic elastic constants were obtained for TANG, X. • CHENG, C. H. 1996. Fast inversion of
sedimentary rocks, and these have allowed us formation permeability from Stoneley wave logs
using a simplified Biot-Rosenbaum model. Geo-
to generate logs for these constants. As expected, physics, 61, 639-645.
dynamic constants are greater than static ones. WENDT, W. A., SAKURAI, S. t~ NELSON, P. H. 1986.
(3) The capacity to obtain a relationship Permeability prediction from well logs multiple
between static behaviour of rocks from dynamic regression. In: LAKE, L. W. & CARROLL, H. B. Jr
properties combines the advantages of both (eds) Reservoir characterization. Academic Press,
methods in one. Thus, the resultant properties 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 of impedance is mathematically expedient. For


results from its conductive and dielectric re- example, when the real and imaginary compo-
sponse in the presence of an electric field; the nents are paralleled, it is better to use admittance
former is related to the transport of free charge (Y*),
carriers and the latter is due to geometrical, Y* = 6+ jB (4)
interfacial and electrochemical mechanisms (Sen
1980, 1981). A complex impedance vector (Z*) where G is the conductance and B is the
consists of a real part ( in-phase or resistance, R) susceptance. The complex conductivity or* can
and an imaginary part (out-of-phase or reac- be calculated from Y*,
tance, X). Using the rectangular-coordinate
or* = Y* L/A = or'+jcr" (5)
form, the complex impedance can be expressed
as follows, where a' and or" are the real and imaginary
Z*= R + jX (1) conductivities, respectively.
where j = v / - 1 is the complex operator. The
phase angle (0) by which current and voltage are Background
shifted is given as:
Complex electrical impedance measurement is a
0 = tan I(X/R) (2) non-invasive technique where an electrical cur-
The complex resistivity p* can be calculated rent flows through the sample at different
from Z*, frequencies. Experimental measurements of the
p* = Z* A/L = p ' + j p " (3) electrical properties of rocks, when submitted to
an alternating electrical field at different fre-
where A is the cross-sectional area of the sample quencies, have shown that both the resistive and
and L is its length, and p' and p" are real and reactive components of the complex impedance
imaginary parts of the complex resistivity, vary over the frequency spectrum. These two
respectively. In some cases, using the reciprocal features (complex quantity and dispersion or

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

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

Sample Density Porosity Kair Saturation Chargeability Chargeability


grams cm-3 % mD Sw(%) Partial Sat. Full Sat.

Z1 2.66 22.4 1760 33 0.74 0.53


Z3 2.66 14.1 38.7 39 0.81 0.53
Z4 2.64 12.2 3.51 61 0.68 0.54
Z5 2.65 20.1 163 47 0.55 0.52
Z7 2.71 25.8 23.1 77 0.54 0.53
Z8 2.69 21.5 101 82 0.52 0.54
Z9 2.64 29.1 819 44 0.53 0.50

Table 2. List of synthetic shaley samples


Sample Clay Type Clay Length Area Grain Porosity Kair
Content cm cm2 Density % mD
grams cm-3

SZ1 clean 0 6.41 10.75 2.65 29.1 337


SZ2 montmorillonite 5 6.49 10.75 2.66 28.1 235
SZ3 montmorillonite 10 6.53 10.75 2.66 27.4 146
SZ4 montmorillonite 15 6.37 10.46 2.66 27.9 105

frequency dependence) can be used to estimate layer by an oscillating electrical field is usually
rock petrophysical properties, such as specific accepted (Lima & Sharma 1992) as the main
surface area and permeability. mechanism for the frequency dependence of
The origin of the frequency dependence can be rocks. Therefore, this interface polarization may
related to geometrical effects of the clay particles provide a link between complex resistivity data
(Sen 1980) or electrochemical phenomena at the and pore-scale attributes, such as pore geometry
fluid-grain (Rink & Schopper 1974) and/or and specific surface area, which in turn can be
fluid-fluid interface (Knight & Endres 1991). related to rock permeability through a Kozeny-
The interface region between matrix and the Carman type of relationship ( Borner 1995;
fluid-filled pore space is complicated due to the Denicol & Jing 1996). Since the frequency
existence of the ionic double layer. The concept dependence is reflecting interface phenomena,
of the electrical double layer forms the theore- salinity of the pore water also influences the
tical basis for understanding the electrical dispersion due to the variation in the double
properties of rocks, especially shaley sandstones. layer thickness and ion mobility. Furthermore,
Electrochemical theory suggests that the surface fluid saturation also plays a role due to addition
of clay minerals carries excess negative charges of the water/oil interfacial area, an increase in
as a result of the substitution of certain positive the tortuosity of the brine-phase distribution
ions by others of lower valence. When the clays and the presence of a non-ionic fluid. The main
are brought in contact with an electrolyte, these objective of this paper is to investigate, experi-
negative charges on the clay surface attract mentally, the effects of brine salinity, fluid
positive ions and repulse negative ions present in saturation and clay minerals on the complex
the solution. As a result, an electrical ionic impedance of different rock samples with vary-
double layer (or diffuse layer) is generated on the ing porosity and permeability. The samples
exterior surface of particles. Typical distribution include outcrop cores, oil-field reservoir rocks
for ionic concentration and electric potential can and synthetic shaley rocks (Tables 1 and 2). A
be predicted by the Guoy (1910) theory. The brief geological description of all sandstone
Gouy theory also predicts that the double-layer samples is given in the Appendix.
thickness (Xd) is reduced as the concentration of
the bulk solution increases. The region outside Experimental apparatus and procedures
the electric double layer (distance > Xd) is called
the free-water region. Complex impedance measurements were per-
Ionic double layers exist between rock and formed using a multi-sample rock testing sys-
fluid interfaces. The perturbation of the double tem. The apparatus can accommodate five
Fig. 1. Schematic representation of the experimental apparatus.

samples simultaneously under varying hydro- measured continuously at 2KHz frequency,


static confining pressure, temperature and in- showed no significant variation (i.e. < 1%
dependently controllable pore pressure. Since all change over a period of 12 h) with brine
the samples are under the same conditions of displacement. The RCL meter was then con-
pressure and temperature, it eliminates experi- nected and a frequency sweep performed on
mental comparison errors due to fluctuations each sample. After the frequency measurement,
during the period of testing (Jing et al. 1992). 5% brine solution was injected through the
The experimental system is shown schematically samples to displace the original brine. The
in Fig.1. Complex impedance measurements resistance was observed continuously. A sharp
were made using the frequency response analy- decrease was observed during the first few pore
ser (QuadTech Model 7600 RCL) in the volumes of displacement, when the more con-
frequency range of 10 Hz to 2 MHz. The ductive brine became continuous. Then, the
instrument is capable of compensating for the decrease was less accentuated and reached
residuals of test fixture and cables based on the equilibrium after about 20 pore volumes of
open/short circuit compensation technique in injection. A frequency sweep was then repeated
the whole frequency range. The instrument is at 5% brine salinity.
equipped with four coaxial BNC terminals on its In order to study the frequency dependence of
front panel which locate its calibration plane. partially saturated rocks, the desaturation tech-
The calibration plane is the position where the nique using semi-permeable capillary dia-
instrument measures within its specified accu- p h r a g m s has been used f o l l o w i n g the
racy (0.05%). In our experiment, test fixture and laboratory procedures described by Elashahab
cables were used to interconnect the sample to et al. (1995). The main advantages of the method
the instrument in a four-terminal configuration are the reduction of capillary end effects and
(4T). The parasitics related to test fixture, cables uniform saturation distribution along the core
and connections are frequency dependent and length. These improvements are achieved by
they were minimized using the 4T configuration using highly hydrophilic ceramic membranes
and by applying the open/short compensation positioned between the sample and the end
technique in the whole frequency range similar plate. The resistivity distribution along the core
to the technique used by Taherian et al. (1990). is monitored by six potential electrodes equally
The effect of salinity was investigated for two spaced along the rock sample so that resistivity
brine concentrations: 20 g and 50 g of sodium measurement can be taken at pairs of electrodes
chloride (NaCI) per litre of solution (i.e. 2% and (four-electrode configuration) and also between
5% NaC1). The solution is made up of NaC1 the top and base current electrodes which give
dissolved in de-aerated and de-ionized distilled the total resistivity (Fig. 2). The resistivity
water. Initially, the rock samples were fully measurements for saturation monitoring based
saturated with 2% brine solution and loaded in on the Archie type of equations are taken at a
the test cell. The samples were considered fully frequency of 2 kHz. The volume of brine
saturated when the resistance of the samples 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 tmorillonite) and SZ4 (15% montmorillonite).
average sample saturation by material balance. The sand and clay mixtures were mixed uni-
The effect of clay minerals on the complex formly to achieve homogeneous samples. Table
resistivity was investigated using synthetic shaley 2 lists the petrophysical characteristics of the
samples following the method established by synthetic samples. After loading the samples in
Jing et al. (1992). According to this technique, the high pressure cell, they were saturated with
mixtures of sands with different ranges of grain 5% by weight of NaCI brine and the consolida-
sizes and different clay types and contents can be tion process was started. Repeated loading and
prepared and consolidated through cycles of unloading cycles were performed with confining
loading/unloading and heating/cooling in a high pressures varying from 500 psi to 4000 psi until
pressure and high temperature cell. The main sample consolidation.
advantage of the technique is full control of the
sample preparation so that the desired variation Results and discussion
of clay type, content and distribution can be
systematically obtained under laboratory condi- Frequency effect
tions.
Five synthetic samples of different clay con- Figures 3 and 4 show the real component and
tents were prepared , namely SZ1 (clay free), phase angle versus frequency for two reservoir
SZ2 (5% montmorillonite), SZ3 (10% mon- core samples. This plot of resistivity and phase
THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES 151

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

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
Fig. 5. Argand diagram with the critical frequency (fc) sharp change in impedance and phase angle. The
separating electrode polarization and bulk sample transition between the intermediate and high-
response for sample Z1. frequency region is characterized by the relaxa-
tion 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.
Fig. 6. Argand diagram with the critical frequency (fc)
separating electrode polarization and bulk sample
response for sample Z1. Salinity dependence
The general frequency behaviour of the complex
angle against frequency can be divided into impedance is shown in Fig. 7 for the reservoir
polarization and sample response regions. The sample Z7 at two different brine concentrations.
electrode polarization region (e.g. < 10 KHz) is The effect of increasing the pore electrolyte
strongly influenced by polarization at the rock- salinity on the frequency behaviour of the
electrode interface and can be identified from 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 favour the expansion of the double layer. The
brine salinity increases in the whole fre- frequency dependence, as expressed by the slope
quency range; taken from the semi-log plot of the normalized
(2) The complex impedance decreases with impedance in the frequency range from 10 to 100
frequency for both brine concentrations; kHz, is found to increase from 5% to 2 % NaC1.
(3) The rate of decrease is more pronounced Similar results were reported by Kulenkampff et
for the lower brine concentration, that is, al (1993) and Kulenkampff & Schopper (1988).
the lower the salinity of the brine the higher This salinity dependence is also related to the
the frequency dependence. This behaviour relative mobility of the ions in the pore space
is best illustrated when the normalized from the free water to the double layer near the
impedance is plotted against frequency in solid surface. In the free water region, the charge
the range from 10 to 100 kHz (Fig.8); carriers are free to move and therefore follow the
(4) As the salinity of the brine increases, the alternating electrical field. On the other hand, in
relaxation frequency increases. the double layer region, the movement of ions is
partially restricted by the electrostatic potential.
The frequency dependence as a function of The result is a delayed oscillation of the diffuse
salinity variations is related to the electrical layer when compared to the free ions of the bulk
double layer, the thickness of which varies with solution that react promptly to the alternating
the brine concentration. High solution concen- electrical field. Consequently, a phase lag is
trations are associated with the compression of established between the input voltage and the
the double layer whilst low concentrations corresponding current flowing through the pore
THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES 153

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)
154 P.S. DENICOL & X. D. JING
0.85
space. If the concentration of NaCl decreases, Z3
Zl 9
the double layer thickness increases and the 080
phase lag is more accentuated. Additionally, an E_075 Z4
increase of the diffuse layer thickness favours the ~"070
blockage of ions, especially at narrowing pores, ~ 065
with consequent accumulation of charges and ="0.60
local concentration gradients.
0.55

0.50
0.3 0.4 0.5 0.6 0.7 0,8 0,9
Saturation dependence Sw (fraction)

The saturation dependence was studied by Fig. 12. Correlation between chargeability (m) and
comparing the frequency spectrum of the real water saturation.
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 metallics, also plays an important role in
dielectric constant of 2.02 at 25 ~ So far, only increasing the complexity of the frequency
water-wet samples have been tested. The fre- dispersion. However, the general behaviour of
quency dependence of the in-phase resistivity the saturation dependence is characterized by an
and phase angle are shown in Figs 9 and 10 for increase of the frequency effect in response to the
sample Z1 at 33% and full brine saturation, oil saturation, as distinguished by the phase
respectively. The saturation dependence be- angle and chargeability results.
comes clearer when both resistivity curves are It is important to note that in two-phase
displayed in a log-log plot (Fig. 11). The fully systems, the frequency dispersion due to the
saturated curve is almost flat for the whole polarization at the solid-liquid interface (pore-
frequency range. On the other hand, the grain) may be added to by polarization at the
partially saturated curve is flat in the low liquid-liquid interface (oil-water). As the water
frequency range and shows clear frequency saturation decreases, there is an increase in the
dependency above the relaxation frequency. water-oil interfacial area and an increase in the
The frequency effect can be better analysed by complexity of the brine phase topology. For any
the empirical parameter chargeability (m) de- rock-fluid systems, wettability plays a significant
fined as follows (Siegel 1959): role in controlling fluid distribution at the pore
scale. Therefore, it might be possible to derive
m = R1/(R1 +R2) (6) wettability information based on the frequency
dispersion measurements of reservoir rock-fluid
where R1 and R2 stand for the low and high systems. However, further research is needed in
resistivity asymptotes, respectively. Table 1 this area.
summarizes the results obtained for the charge-
ability of the samples. For a given sample, there Clay effects
is a consistent increase in m when brine is
displaced by oil. The correlation between m and Synthetic shaley samples with controlled clay
the water saturation is shown in Fig. 12 for all type, content and distribution were used to
the samples. Although a general trend of higher investigate the effects of clay minerals on
m for lower saturation can be observed, the complex impedance measurements. Figure 13
correlation is weak. Samples Z3 and Z9 showed shows the results for the synthetic sample SZ4.
a more remarkable deviation from the trend, The low-frequency region from 10 Hz to ~10
possibly due to the high iron oxide content in the kHz indicates strong dispersion in both impe-
former and dispersed glauconite in the latter. dance and phase angle which is attributed to
Knight & Nur (1987) also observed that a electrode polarization. In the intermediate fre-
sample with high iron oxide content (Indiana quency range (1 to ~100 kHz) the impedance
Dark sandstone) had an anomalous dielectric decreases monotonically while the phase angle
exponent apparently due to the effect of the reaches a minimum and then starts increasing.
magnetic susceptibility on the dielectric re- The high frequency region is characterized by
sponse. the relaxation frequency at N800 kHz where the
The interpretation of the saturation depen- phase angle reaches a maximum and the
dence upon frequency is difficult due to the impedance decreases more drastically. All the
intricate geometry of the pore space and its effect synthetic samples present the relaxation fre-
on distribution of fluids within the rock. Miner- quency at around the same position. However,
alogical complexity, mainly related to clays and 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, applications for the evaluation of low resistivity
varying from 1 degree for the clay-free sample to and low contrast pay formations.
5 degrees for the 15% montmorillonite sample. The frequency dispersion consistently in-
This observation suggests that a polarization- creases with the amount of clay in the sample.
like process is being caused by the clay presence This effect is better illustrated when the normal-
although the classical induced polarization effect ized impedance is plotted in the frequency range
would be expected at lower frequencies. A from 10 to 100 kHz . The impedance slope is
possible explanation for this frequency effect relatively flat for the clay-free sample (SZ1) and
may be related to the electro-osmotic coupling increases with the content of montmorillonite
due to the accumulation of charges at narrowing for the shaley samples. A plot of the clay content
pores (Marshal & Madden 1959; Dankhazi versus the frequency dependency clearly shows a
1993). Although its effect is found to be very relationship.
weak, this type of polarization is expected to
increase with a reduction of the sample perme- We would like to thank Petrobras S.A. for sponsoring
ability. Indeed, the synthetic samples show a P.S. Denicol and for providing reservoir rock samples.
decrease in permeability with increasing amount We also wish to thank M. S. King for many valuable
of clay (Table 2) that leads to the narrowing and discussions.
reduction of effective pores and hence the
electro-osmotic coupling.
The slope of the impedance curve in the range References
from 10 to 100 kHz is also found to correlate
with the clay content of the samples. Figure 14 BORNER, F. D. 1995. Estimation of hydraulic con-
shows the normalized impedance versus fre- ductivity from complex electrical measurement.
quency for the samples containing montmor- International Symposium of the Society of Core
illonite. The graph indicates a consistent increase Analysts, paper 9523.
COLE, K. S. & COLE, R. H. 1941. Dispersion and
in the frequency slope from sample Z1 (clay- absorption in dielectrics. Journal of Chemistry and
free) to sample Z4 (15% montmorillonite). A Physics, 9, 341.
plot of the rate of impedance decrease with DANKHAZI, G. 1993. A new principle approach to
frequency versus clay content is shown in Fig. induced polarization in porous rock. The Log
15, where the clay effect appears to decrease at Analyst, 34, 54-66.
higher clay contents. DEBVE, P. 1929. Polar molecules. Chemical Catalogue
Co.
DENICOL, P. S. & JING, X. D. 1996. Estimating
Conclusions permeability of reservoir rocks from complex
resistivity data. Society of Professional Well Log
The frequency effect in the intermediate fre- Analysts, 37th Annual Logging Symposium,
quency range (10-100 kHz) increases when the paper X.
solution concentration is decreased from 5% to ELASHAHAB,B. M., JING, X. D. & ARCHER,J. S. 1995.
2% NaC1. This salinity dependence may be Resistivity index and capillary pressure hysteresis
explained by variations of the double layer for rock samples of different wettability charac-
thickness and ion mobility. At high salinity, teristics. SPE paper No. 29888, the 9th Middle
the double layer is compressed to the pore East Oil Show and Conference, March, Bahrain.
surface and gradually expands with decreasing Gouv, G. as discussed in HUNTER, R. J. 1988. Zeta
brine concentration. As a consequence, the Potential in Colloid Science, Academic Press.
JING, X. D., ARCHER, J. S. 8r DALTABAN,T. S. 1992.
mobility of the ions in the diffuse layer is Laboratory study of the electrical and hydraulic
reduced at high salinity preventing them from properties of rocks under simulated reservoir
following the alternating field as opposed to the conditions. Marine and Petroleum Geology, 9,
free ions in the centre of the pore. Additionally, 115-127.
the expansion of the double layer supports the KNIGHT, R. & NUR, A. 1987. Geometrical effects in the
blockage of ions particularly at the smaller pores dielectrical response of partially saturated sand-
with subsequent electro-osmotic polarization stones. The Log Analyst, 28, 513-519.
due to the accumulation of charges. KNIGHT, R. & ENDRES,A. 1991. Surface conduction at
The frequency effect is found to increase for the hydrocarbon/water interface. Society of Pro-
fessional Well Log Analysts, 32nd Annual Log-
the whole frequency range when brine is ging Symposium, paper I.
displaced by oil (Isopar H). A variation in water KULENKAMPFF,J. M. & SCHOPPER,J. R. 1988. Low
saturation from full to partial saturation re- frequency conductivity--a means for separating
sulted in a dramatic increase in the frequency volume and interlayer conductivity. Society of
dispersion and a clear shift of the relaxation Professional Well Log Analysts, 12th European
frequency. This observation may have potential Formation Evaluation Symposium, paper P.
THE COMPLEX RESISTIVITY OF SANDSTONE SAMPLES 157

--, BORNER, F. D. & SCHOPPER,J. R. 1993. Broad logically mature, texturally submature to
band complex conductivity lab measurement mature.
enhancing the evaluation of reservoir properties.
Society of Professional Well Log Analysts, 15th
European Formation Evaluation Symposium,
Sample Z3 (Block 18-2)."
paper A. Lower Permian "Penrith Red Sandstone", pre-
LIMA, 0. A. L. t~ SHARMA, M. M. 1992. A grain dominantly quartz grains cemented by quartz
conductivity approach to shaly sandstone. Geo- over-growths with iron oxide petina, sub-
physics, 55, 1-10. rounded-rounded, mineralogically and textu-
LOCKNER, D. A. & BYERLEE, J. D. 1985. Complex rally sub-mature.
resistivity measurements of confined rock. Journal
of Geophysical Research, 90, 7837-7847. Sample Z4 (Block 16-2):
MARSHALL, D. J. t~ MADDEN, T. R. 1959. Induced Upper Carboniferous sandstone, grain size : 0.1-
polarization, a study of its causes. Geophysics, 24,
790-816. 0.3 mm, 85% quartz, 0% alkali feldspar, 5%
RINK, U. • SCHOPPER,J. R. 1974. Interface conduc- mica, very irregular, poor sphericity, texturally
tivity and its implications to electrical logging. immature and mineralogically submature.
Society of Professional Well Log Analysts, 15th
Annual Logging Symposium, paper J. Sample Z5 (Block 19-4):
SEN, P. N. 1980. The dielectric constant and conductiv- Lower Triassic 'Bunter' sandstone, fine to
ity response of sedimentary rocks. Society of medium grain sizes (< 0.5 mm), 95% quartz,
Petroleum Engineers, paper 9379. % alkali feldspar and calcite, sub-rounded, poor
SEN, P. N. 1981. Relation of certain geometrical sphericity, texturally and mineralogically ma-
features to the dielectric anomaly of rocks.
Geophysics, 46, 1714. ture.
SEIGEL,H. 0. 1959. A theory for induced polarization
effects (for step excitation function). In: WArr, J. (b) Reservoir rocks
R. (ed.) Over Voltage Research and Geophysical
Applications. Pergamon Press Inc., 4-21. Sample Z7:
TAHERIAN, M. R., KENYON, W. E. & SAFINYA, K. A. Glauconitic sandstone, semi-friable, grains are
1990. Measuremen of dielectric response of water- sub-rounded with regular to good selection.
saturated rocks. Geophysics, 55, 1530-1541. Mineralogy also includes quartz, feldspar and
mica.
Appendix: geological description of sand-
stone rocks Sample Z8."
Sandstone with pseudo-argilaceous matrix (27
(a) Outcrop rocks %), quartz (31%), K-feldspar (18%), glauconite
(6%), plagioclase (5%), others (2%). Cements
Sample Z1 (Block 15-8): include dolomite and pyrite.
Lower Carboniferous sandstone, average grain
size 0.2 ram, 95% q u a r t z , alkali feldspar, clay, Sample Z9."
biotite, alcite cement (5%) with some chert, Sandstone with pseudo-argilaceous matrix,
angular to sub-angular, poor sphericity, minera- 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 cracks has been extensively studied by, amongst
and fractures reduces the acoustic velocities of others, Crampin (1984, 1985a,b) and Crampin &
P- and S-waves in rocks. When the principal Atkinson (1985), who are of the opinion that S-
stresses are altered on a rock that initially has a wave velocities are more sensitive to the presence
random distribution of cracks, the crack dis- of aligned cracks and that they provide a better
tribution no longer remains randomly oriented. quality of information on anisotropy effects than
The effect of an applied non-hydrostatic stress is does the P wave. Crack orientation, when cracks
to close cracks in some directions and leave are aligned vertically, can easily be determined
cracks open in others (Sayers 1988). Those by the splitting of vertically propagating polar-
cracks with their normals lying close to parallel ized shear waves. This splitting occurs as a result
to the new major principal stress will tend to be of azimuthal anisotropy induced by the micro-
closed more than those with their normals sub- cracks and fractures. A knowledge of seismic
parallel to the new minor principal stress (Sayers anisotropy can provide useful information about
1988). The elastic and transport properties of the the mineralogy, the orientation of cracks and
rock then become anisotropic in their behaviour, pores, the degree of cracking and crack geome-
with the degree of anisotropy depending on the try, orientation of the in situ stress field, and the
magnitude of the principal stress differences, the possible proportion of gas and liquid within the
type of fluid filling the cracks (Xu & King 1989, inclusions in hydrocarbon reservoirs (Crampin
1992; King et al. 1995a,b). 1985a).
Seismic anisotropy was studied more than 40 Thomsen (1986) has derived a set of three
years ago by Postma (1955) and Uhrig & Melle dimensionless anisotropy parameters (e, 7 and 8)
(1955), but for a long time its effect was ignored to describe weak to moderate transverse iso-
or considered insignificant, due to the fact that tropy of a medium. These parameters are
most of the seismic surveys carried out were for defined in terms of the five components of the
P-wave reflection and conducted at small angles stiffness tensor (Cll , C33 , C13 , C44, C66) relating
to the vertical. However, for seismic surveys stress and strain for the transversely isotropic
conducted with large angles of the incidence medium as follows:
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 Cll -C33 V 2 1 - V22
-- - - -- (1)
1985a,b). Seismic anisotropy due to aligned 2C33 V 22

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

C66 - C44 V 21 - V 22 Imperial College of Science and Technology


2C44 V 22
(2) London, has been used for testing 51mm-side
cubic rock specimens. The system, described in a
preliminary technical note by King et al. (1995a)
(~ ~__ (C13 + C44) 2 - (C33 - C 4 4 ) 2 and in detail by Shakeel (1995), consists of a
2C33 (C33 - C44) (3) loading frame in the form of an aluminium alloy
ring within which two pairs of hydraulic rams
The parameters are all zero for an isotropic and ultrasonic transducer holders are mounted
medium and their deviation from zero represents to provide orthogonal stresses on the cubic rock
the degree of anisotropy. The value of ~, which is specimen in the horizontal plane. Each of the
always positive, represents the relative difference three principal stresses may be varied indepen-
between the P-wave velocities propagating per- dently in the range 0 to 115 MPa in the
pendicular (Vp1) and parallel (Vp2) to the axis of horizontal principal directions and to over 750
symmetry. The general term 'anisotropy' of a MPa in the vertical major principal direction.
rock usually refers to the quantity e, calculated The horizontal principal stresses may be servo-
using the following equation for small values controlled using facilities associated with a
o f e, Schenk compression testing machine. The ver-
Vp 1 B Vp 2 tical major principal stress is provided through
E -- - - (4) ultrasonic transducer holders mounted in a
re2 Schenk 160-tonne closed-loop servo-controlled
The parameter 3' describes the S-wave aniso- compression testing machine. Stress is trans-
tropy of a transversely isotropic medium. It is mitted to each of the six faces of the cubic rock
the relative difference between the faster S-wave specimen through 5 ram-thick magnesium face-
velocity (Vsl) and the slower S-wave (Vs2) plates matching approximately the elastic prop-
velocity travelling in a transversely isotropic erties of the rocks being tested. Deformation of
medium. Thus, for small values of 7, it can be the rock specimen is measured by pairs of
used to define 'S-wave anisotropy' of a medium extensometers (LVDTs) mounted in each of
(Thomsen 1986) as the three principal directions. An isometric
view of the polyaxial loading frame is shown in
VS1 m Vs 2
7 -- - - (5) Fig. 1.
Vs2 Each of the three pairs of transducer holders
contains stacks of piezoelectric transducers
where Vsl and Vs2 are S-wave velocities
capable of producing or detecting pulses of
propagating parallel to the plane of cracks with
compressional (P) or either of two shear (S)
their polarization parallel and perpendicular to
waves polarized at right angles propagating in
the plane of cracks, respectively. The parameter
one of the principal stress directions. The
dominates the anisotropic response when the
transducer holders have a bandwidth in the
acoustic wave propagates in a plane which is
range approximately 450 to 800 kHz for P-wave
parallel or approximately parallel to the axis of
and 350 to 750 kHz for S-wave pulses.
symmetry. It is independent of the seismic
Loading in the 1-direction is characterized by
velocities of the medium perpendicular to the
the major principal compressive stress (cq)
axis of symmetry and can take either positive or
direction and that of 2- and 3- as the inter-
negative values.
mediate (or2) and minor (a3) principal stress
As shown by Thomsen (1986), the parameters
directions, respectively. The wave type nomen-
~, 3" and g are less than 0.2 in magnitude for
clature employs two suffixes 'i' and 'j' (as with
weak-to-moderate anisotropy. Furthermore,
Vij) where T refers to the propagation direction
Thomsen (1986) states that elliptical anisotropy
of the wave and 'j' to the polarization (particle
will be observed if 6 = ~. Since the parameters ~,
motion) direction. Thus V33 is the P-wave
3' and ~ are easily interpretable and can be
velocity propagating in the minor principal
calculated from the five elastic constants ob-
stress direction and V13 is the S-wave velocity
tained from Nishizawa's (1982) theory, they are
propagating in the major principal stress direc-
used here to model and study the variation in
tion with polarization in the 1-3 plane. A total
anisotropy as a function of aspect ratio, crack
density and stress. of nine components of velocity are measured:
three compressional VPll , VP22 and VP33 and
six shear VS12, VS13, VS21, VS23, VS31 and VS32
Experimental system Both the P- and S-wave velocities are measured
with an accuracy of +1% and a precision of
A polyaxial stress loading system, developed at +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

9 , ,

TRANSDUCER HOLDERS
HYDRAULIC PRESSURE,
2-DIRECTION
4. HYDRAULIC RAM,
2-DIRECTION
5. HYDRAULIC PRESSURE,
3-DIRECTION
6. HYDRAULIC RAM,
3-DIRECTION
7. "fRANSDL~ER HOLDERS
8. CUBIC ROCK SPECIMEN
9. REACTION RING

Fig. 1. Isometric view of the polyaxial loading system.

Results and discussion of 1.5• -4 GPa, and liquid-filled inclusions


with a fluid bulk modulus of 1.5 GPa. The
First a numerical example is provided to enable isotropic background material is the same as
a better understanding of the effects of the that used by Nishizawa (1982): matrix density
different parameters, such as crack aspect ratio, 2.7 g cm -3 and Lame's constants A = # = 39 GPa.
crack density and type of saturating fluid on the Figures 2 and 3 show results of the Thomsen's
anisotropy parameters and on the acoustic anisotropy parameters as a function of aspect
velocities of such a cracked solid permeated ratio for dry and liquid-filled inclusions, respec-
with aligned ellipsoidal inclusions. Finally, the tively, for four different crack densities ~= 0.01,
theory is used to study the anisotropy as a 0.05, 0.10 and 0.20. It is clear from these figures
function of stress for a solid progressively that the values of anisotropy parameters in-
permeated with a system of aligned cracks. crease as the crack density is increased from ~ =
0.01 to 0.20. They all become zero for an aspect
Numerical results and discussion ratio a = 1, corresponding to the isotropic
situation.
In this numerical example, P- and S-wave Note that for dry inclusions (Fig. 2) all the
velocities and Thomsen's (1986) anisotropy anisotropy parameters have a non-zero constant
parameters are calculated as a function of aspect value for a large range of small aspect ratios and
ratio for a solid permeated with aligned ellipsoi- that they only tend to zero for large aspect ratios
dal inclusions. Nishizawa's (1982) theory is used approaching a = 1. Hence, for a large group of
to calculate the elastic constants. The aspect small aspect ratios the resultant anisotropy is
ratio of the inclusions is varied from a =0.0001 hardly affected by a change in aspect ratio for
(almost flat cracks) to a = 1 (spheres). Four the case when solid is permeated by dry
crack densities are studied, ~=0.01, 0.05, 0.10 inclusions. The non-zero constant values of the
and 0.20. Both types of inclusions are investi- parameters ~ and 7, for a large group of small
gated: dry inclusions with a fluid bulk modulus aspect ratios for dry inclusions, indicate that
176 A. SHAKEEL & M. S. KING

0.04 . . . . . . . . I . . . . . . . . I' . . . . . . ' . . . . . . . I . . . . . . .

: C r a c k density = 0 . 0 1

:,= o.o2 i / Y 87" ""


o7
o.01

O.OO I 9 * ' I llll i . . . . . . . I . . . . . . . . I

O.OeO o.ool O.Ol 0.1


(a)

0~0 . . . . . . . . I . . . . . . . . | . . . . . . . . I . . . . . . . .

Crack density = 0.05 ~-. r


0.15
"2.'.'_"2~: "-'."L\~ 22:" 2"-".'2"2:: - " 2 " . ' - " . 2 : ~ 22".222:: -"2.-'_"".'.:: -':;.':.z-~. ~,.,..
>~ o.lo y b
, \
o.o$ -~- "..,~
0.00 ~ " ' " I I''| . . . . . . . . | ' ' ' ' ' ''']

0.04}01 0.001 0.01 0.1


(b)

0.5 . . . . . . . . I . . . . . . . . i . . . . . . . . I . . . . . . . .

Crack density = O. I0 ~)
0.4 9 ~"
"~ "..--.'.'T ." . - : -" . " . ~. - - - : . ' T . . " . ' : " -".:. 7----.-.-: : .--.'..--:.--. - .-"" -- "~.'.'T 2 2": : .".-:" - -~....: : :"; "'~" ". ~

o.3 ~';r
O~ y '~.~
/
0.1

O.O . . . . , , ..! 9 , , i , , ,r I . . . . . . . . I . . . . . . . .

0.0001 O.OOl 0.01 0.1


(c)

. . . . . . . . I . . . . . . . . I . . . . . . . . I . . . . . . .

Crack density = 0.20

l.o .". . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . - 8

p..

/ Y "'-..

. . . . i i,,| . . . . . . . . [ L , . . . . ..l , , L i " "


0.0
O.OeOI O.OO! O.OI O.t
Aspect 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 for dry inclusions of crack density of 0.01. In
two P- and two S-wave velocities propagating this figure (also in Fig. 5) the slower acoustic
both parallel and perpendicular to the plane of velocities (VP2 and VS2) are represented by
cracks for the same range of aspect ratios. This dotted lines and the faster (VPi and VS1) by
conclusion corresponds to Fig. 4 which shows a solid lines. A study of Fig. 2 indicates that the
very small variation in P- and S-wave velocities anisotropy parameters follow a certain pattern
propagating both parallel and perpendicular to for dry inclusions, i.e. for low crack densities
the plane of cracks as the aspect ratio is changed < 0.1, e > 6 > 7, (Figs 2 a-b) and for high crack
ANISOTROPY IN CRACKED ROCKS 177

0.03 9 '' ...... I ...... ''I ........ I .......


Crack~ density = 0.01 p t ' Y..... . . - ' " " ..- .-.. . -. . . -. ( _ : . ~ " :~.-'=" = : - ' : - : k : . ,..
0.02

+r O.Ol

0 ..... -+"~
o-,"
-0.01 ..................... .~-"*"

-0.02 ........ I ...... ill i i i i. .... I . . , ....


0.0001 O.OOl O.Ol 0.I ]

(a)

O.IS ......... I . . . . . . . . I ~ . . . . . . . I

0.1
C r a c k density = 0.05 ........... ","U:':u:-':~.:. ,,
/ ...... . .... _.-j.....-
0.05
s . ...o..-o'~ ~- j-
.................................. ~
,,J 0
. o~"
-0.05 ................ .---"
, . , , .... I , , , , , ,.,1 i n i i i '''I
-0.1
0.0001 0.001 O.OI 0.1

(b)
0"~ ........ ! ........ 1 ........ i .......
0.2 C r a c k density --- 0.10 + ..)~ ~~ ~~ -'~" = : z.-;~..
/ ........ . - \
.o 9
,r 0,1 ~ ~ .......... .....~ .*" f . /

0 ........................ ~ o ~ o "" J~l~

-0.1 . . . . . . . . . . . . . . . . . . . . . . . . "~'~"

-0.2 . . . . . . . . i . . . . . . . . i . . . . . . . . I . . . . . . . .

0.000 0.001 0.01 O. !

(c)

0,6 9 " ...... I ........ i ........ i ..... .:~. . . . . . . .


0.45 C r a c k denmty = 0.20 ~ ..-" I" "%

0.3 I o
~ - i -/" \ 9

0.15 E ....o~ .,.o


.s

-0.15
-0.3 1.. . . . . . . I i . . . . . +,il I i . . . . . . l , , , ....
0.O~i 0.001 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). result corresponds to Fig. 5a which shows a
However, for a solid permeated with liquid- significant variation in V P 2 a s the aspect ratio is
filled inclusions, a large variation in anisotropy changed for liquid-filled inclusions of crack
parameters ~ and 6 is observed as the aspect density ~=0.01. It can be seen from Fig. 3 that
ratio is changed (Fig. 3). The changes in e are the value of ~ tends to zero for very small aspect
related to P-wave velocities, especially VP2 ratios for all the crack densities, indicating that
which is strongly influenced by the liquid-filled the difference between the P-wave velocities in
inclusions as the aspect ratio is changed. This both the directions parallel and perpendicular to
178 A. SHAKEEL & M. S. KING

6.60 . . . . . . . . i . . . . . . . . I . . . . . . . . J

6.55

6.$0

6.45

6.40 ......................................................................... ~

6.35 ,,.

Vr2
6.30 Crack density = 0.01

, , , , , , , I , , t . . . . . i
6.~g . . . . . . . . t .

0.0001 0.001 0.01 O. I

Aspect Ratio
(a)

3.85

3.80 Vs~

3.75

3.70

3.65 Crack density -- 0.01

3.60 , , . . . . . . . I . . . . . . . . I , , , . . . . . I

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 both dry and liquid-filled inclusions, the acoustic
is clearly shown in Fig. 5a. The changes in ~5are velocities VS1 and VS2 are not strongly affected
related to those P- and S-wave velocities which by aspect ratios (Figs 4b and 5b for a crack
are either propagation or polarization perpendi- density ( = 0.01).
cular to the plane of cracks (3-direction). For all the cases studied for dry and liquid-
A comparison of Figs 2 and 3 with each filled inclusions, the values of e, 7 and 6 are
respective crack density, indicates that there is always positive, indicating that the P- and S-
hardly any difference in the value of anisotropic wave velocities (VP 1 and VS1) propagating
parameter 7 for a large range of aspect ratios perpendicular to the axis of symmetry are always
when either dry or liquid-filled inclusions are greater than those propagating along (VP2 and
used. This behaviour corresponds to the fact VS2) the symmetry axis (Figs 4 and 5). Finally,
that S-wave velocities are not affected much as Figs 2 and 3 show that the parameters e and 6
the condition of inclusions is changed from dry are equal for aspect ratios lying between a = 0.4
to liquid-saturated. Since the parameter 7 is and 1 for both dry and liquid-filled inclusions.
constant for a large range of aspect ratios for This suggests that the resultant anisotropy is
ANISOTROPY IN CRACKED ROCKS 179

6.60 ........ i ........ i . . . . . . . . i

Vp l
6.55

6.50

6.45

6.40

6.35

Vp2
6.30 Crack density = 0.01

........ I ........ I ........ i .......


6.25
0.0001 0.001 0.01 O.t
Aspect Ratio
(a)

3.85 . . . . . . . . 1 . . . . I . . . . [ ~ . . . . . . . I . . . . . . . .

3.80
Vs i

E
.'a 3.75

3.70
i
3.65
Crack density = 0.01

. . . . . . . . l 1 1 . . . . . . I . . . . . . . . I , [ . . . . .
3.60
0.01.| 0.001 0.01 0.1

Aspect 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 involves measurements on the flesh, uncracked
saturated cases in this limited range of aspect rock specimen during the application of an
ratios. increasing hydrostatic stress. The second cycle
involves measurements while a system of aligned
Experimental results and discussion cracks, with their normals parallel to the minor
stress direction, is formed in the rock specimen
Tests have been performed on five dry sandstone (cracking cycle). The third cycle involves mea-
specimens, in which a system of aligned cracks surements during the application of a further
has been first introduced by increasing the major increasing hydrostatic state of stress to close the
(el) and intermediate (~2) principal stresses in cracks formed during the cracking cycle (crack
unison to near failure, while keeping the minor closing cycle).
(or3) principal stress constant at some low level. Discussed here as being characteristic of the
The aligned cracks are then closed by the studies made on five sandstones tested in this
application of a hydrostatic compressive stress. research programme will be that of Penrith
The nine components of velocity are measured sandstone. This is a fine-to-medium grained
throughout three separate stress cycles. The first sandstone of lower Permian age having a low
180 A. SHAKEEL & M. S. KING
S.O . . . . [ . . . . i I . . . . I . . . . i . . . . [ . . . .
. . . . i . . . . i 9 -

o3 = 3 M P a (constant)

4.5 :I-+1% ,~ I -*1%

4.0 4.0

3..~

3.0
:~- 3.0

ft. 2..~

2.0 ~
= Pll : P22 a P33
1.5 . . . . ' . . . . ] . . . . I . . . . J , , 9 . , ....
l.$ . . . . . . i . . . . [ 9
0 25 SO 75 100 12S ISO
O 50
01 = o 2 = 0"3 ( M P a ) ;~-- 01 = O2 ( M P a ) ~:~
o l > o 2 = IOOMPa

3.2 . . . . i . . . . i . . . . i . . . . ~ . . . . i . . . . "

3.2 . . . . i . . . .
03 - 3 M P a (constant)

2.9 . r -*1%
2.9 -'r -*1%

~ 2.6

2.3
!~
~ 2.,
g 2.0
1.7
$31 ~ $23---o--- $21
1.7 t S13 ~ $32 9 S12
I S13 ~ $32 ! S12 1.4 ,,. i .... i .... i .... i.,,. i ....
50 75 100 125 l.gO
1.4 , , , , I , , , , ~ , ,

50
i~ o I = 02 ( M P a ) =',~
o 1 = 0 2 = 0 3 (MPa) o 1 > o 2 = IOOMPa

Fig. 6. P and S-wave velocities as a function of 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


hydrostatic stess during the initial stress cycle for during the cracking cycle for Penrith sandstone
Penrith sandstone sample, (a) P-wave velocities and (b) sample, (a) P-wave velocities and (b) S-wave velocities.
S-wave velocities.

principal stress was increased in steps to


clay content (3%), an effective porosity of 13%, 132 MPa until the specimen was near failure.
a permeability of ~150 mD, a grain density The acoustic velocities propagating in the 3-
2.6 gcm -3 and a bulk density of 2.26 gcm -3 in its direction show an initial increase with stress due
dry state. to the closure of pre-existing cracks with their
Figure 6 shows changes in the three P- and six normals in the 1- and 2-directions, followed by a
S-wave velocities plotted as a function of decrease as dilatant cracks with normals parallel
hydrostatic stress on the fresh uncracked rock to the 3-direction begin to form and open up. It
specimen. Although loading in the 1-, 2-, and 3- is concluded from Fig. 7, with VPll~VP22 and
directions is identical there is a small difference VS12~,~VS21 all increasing monotonically, that
between the changes in P- and S-wave velocities the majority of the cracks formed are aligned in
which is due to the differences in the initial the 1-2 plane, perpendicular to the 3-direction.
elastic properties between these three directions. Shear wave birefringence occurs in all directions
It will be observed that the sandstone exhibits of propagation except along the symmetry axis
behaviour that is close to being isotropic, with (3-direction) for obvious reasons of symmetry.
both sets of P- and S-wave velocities increasing This effect was also observed in the experiments
in magnitude with increasing stress and lying of Nur & Simmons (1969).
within + 1% error bar, except at the lowest stress The cracking cycle velocity data plotted in
level. Fig. 7 indicate that the magnesium plates match
Figure 7 shows changes in the three P- and six the sandstone well in elastic properties up to
S-wave velocities during the cracking cycle. The stresses of o1 = cr2 = 100 MPa, when the majority
minor principal stress (o'3) was kept constant at 3 of the aligned cracks are formed. As o'] is further
MPa while o']--o'2 were increased in unison in increased (with o'2 constant), the platens cause
steps from 3 to 100 MPa. Then, while maintain- confinement and the S-wave velocities propagat-
ing the intermediate principal stress at 100 Mpa ing in the 1- or 2-direction and polarized in the
(limited by the experimental system), the major 3-direction (V13 or V23) become higher than the
ANISOTROPY IN CRACKED ROCKS 181

7Z:L~ ,--]
5~ 9 9" * 9 I '' " * " " ' 'I . . . . 2.4 . . . . i . . . . i . . . . , . . . . i . . . . i . . . .

4J I 7-O

~ 1.6
3,$ ,o

~ 3.0
G a m m a , ,

O.8

2.0 & 0.4

0 25 SO 75 50 75 1o0 12~ 150

a I *, 0" 2 - 0 3 ( M P a ) '= O 1 = 02 ( M P a ) -'~ o l 9 o2 = 100MPa "4''I


(a)

'-4
3.2 ,~., . ~ . A L

Aspect Ratio ct = 0 . 0 0 ~ /
2.0
2.9 -Z • ~
I~lta.

1.6
A 2.6

/ ,,
I..*

0J

o.4
1.7 .--o--- $31 ~ $23 ~ $21
. . _ , . ~ - - I - .
4 ..... SD ~ S32 x S12
o.o
1.4 . . . . ~ " , , , f , * , , 0 0.1 0.2 o.3 0.4
5O 7$
Crack Density
01 - 0 2 - 0 3 (MPa)

Fig. 8. P and S-wave velocities as a function of Fig. 9. Thomsen's (1986) anisotropy parameters for
hydrostatic stress during the crack closing cycle for Penrith sandstone sample during the cracking cycle as
Penrith sandstone sample, (a) P-wave velocities and (b) a function of (a) stress and (b) crack density of the
S-wave velocities. aligned cracks.

velocities that are propagating in the 3-direction described in detail by Shakeel (1995), who found
(V31 or V32). This behaviour suggests that the excellent fits (within 4-1% at all stress levels) in
aligned crack density towards the extremities of comparing the theoretically modelled and the
the specimen in the 3-direction is higher than in the laboratory measured velocities during both the
centre for values of stress greater than 100 MPa. cracking and crack closing cycles.
Figure 8 shows changes in three P- and six S- As the Penrith Sandstone was tested in its dry
wave velocities plotted as a function of hydro- state, a value of 1.5x 104 GPa was chosen for
static stress during the subsequent crack closing the fluid bulk modulus. A range of aspect ratios
cycle. As the stress is increased, both sets of P- (0.0005 to 0.002) was employed during each of
and S-wave velocities appear to be approaching the stress cycles to obtain the best match
asymptotic values that are only slightly lower in between the modelled and experimental velo-
magnitude than those shown in Fig. 6 for the cities. It was found that a value of aspect ratio
preliminary uncracked cycle. Upon removal c~--8.0xl0 ~ provides the best match between
from the loading frame after completion of the the modelled and the experimental velocities
tests, the specimens all showed signs of through- during the cracking cycle and for the crack
going fractures aligned close to normal to the 3- closing cycle for most of the stress levels.
direction. Figure 9a shows changes in the anisotropy
The nine components of velocity determined parameters e, 7 and 6 as a function of (71 and 0"2
as a function of stress during the cracking and during the cracking cycle, during which o.3 was
subsequent crack closing cycle have been used to kept constant at 3MPa. All the anisotropy
evaluate the Thomsen's (1986) anisotropy para- parameters increase as the stress is increased
meters and crack density for the cracks aligned due to an increase in crack density (Fig. 9b). The
perpendicular to the symmetry axis (3-direction). rate of increase in the value of these parameters
The procedure, employing Nishizawa's (1982) is lower as the stresses o-~ =o-2 are increased
theory, first to model the velocity data, is initially from 2 to 100 MPa, but it becomes much
182 A. SHAKEEL & M. S. KING
. . . . I , . . . . . . . . i . . . . i . . . . i . . . . i . . . . i'.

Aspect ratio a = 0.001~; Aspect ratio a = 0.0005 /

o 3 - 2 MPa (Constant) Delt~

1.5

Epsilon, 9 ~
1.0

Gamma. y ~ - . . . . i . "~. . ,~. . . , . . . . n . . . .


, , , , t , , , , I , , , ,
25 50 75 100 1~ 13o
25 50 75
I~ 0 1 = 02 ( M P a ) -' ~ o I > o2= I00MPa
c; 1 = 02 = 03 (MPa) (a)
(a)

. . . . t , J 9 - J . . . . L . . . . L . . . . i . . . .

Aspect Ratio a = 0.0015 Dolts, 6 "--'~/~

U)
0.02 0.04 0.06 0,(]~ 0.I 0.12 0,05 O.l 0. I$ 0.2 0.25

Crack Density Crack Density

(b) (b)

Fig. 10. Thomsen's (1986) anistropy parameters for Fig. 11. Thomsen's (1986) anistropy parameters for
Penrith sandstones sample during the crack closing Crosland Hill sandstone sample during the cracking
cycle as a function of (a) stress and (b) crack density of cycle as a function of (a) stress and (b) crack density of
the aligned cracks. the aligned cracks.

higher at higher stresses. The higher rate of The anisotropy parameters follow the same
increase of anisotropy parameters for stresses pattern during the cracking and crack closing
o'1 >a2 = 100MPa is due to the nucleation and cycles, i.e. at each stress level ~ > 7 indicates that
coalescence of the majority of the aligned cracks, the anisotropy in P-wave velocities is greater and
which is also clear from the sharp decrease in more sensitive to the crack density than the
V33 and S-wave velocities propagating or polar- anisotropy in S-wave velocities. Furthermore,
ized perpendicular to the plane of cracks (Fig. for higher crack densities (~> 0.08), ~ is greater
7). than ~, which is in accordance with the predic-
Figure 10a shows changes in the anisotropy tion of Nishizawa's theory as shown in Fig. 2. A
parameters e, 7 and 6 as a function of hydro- study of Figs 9 and 10 also indicates that
static stress during the crack closing cycle. All elliptical anisotropy is only possible in the weak
the anisotropy parameters decrease as the stress anisotropic region (anisotropy p a r a m e t e r s
is increased due to a decrease in crack density < 0.2) which occurs only at low crack densities
(Fig. 10b). Cracks close very quickly during the (~ < 0.06). These results are true for all the other
initial loading, resulting in a consequent rapid sandstones tested in this research program. As
decrease in the value of the anisotropy para- an example, Figs 11 and 12 show similar
meters. When the stress is increased further, a anisotropy results for the Crosland Hill (low
major fraction of the crack surface area comes clay content [< 1%], effective porosity 6% and
into close contact which slows down the closure permeability < l mD) sandstone specimen dur-
of cracks and the anisotropy parameters de- ing the cracking and crack closing cycles.
crease much more slowly than before. Results in
Fig. 10 show that the anisotropy becomes weak- Conclusions
to-moderate and elliptical (e = 6) for hydrostatic
stresses >10 MPa. (1) The results of the experimental study for
ANISOTROPY IN CRACKED ROCKS 183

. . . . i . . . . i . . . . i . . . .

Aspect ratio ct 0.0005 = 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
t.6
~ 171tt, 8
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.
25 50 75 ioo
oI= o z = c 3 (MPa)
(a)

References

. . . . . . . i . . . . 0 . . . . i . . . . CRAMPIN, S. 1984. Anisotropy in exploration seismics.


Aspect Ratio cc= 0.0005 #~ First Break, 2, 19-21.
/ - - 1985a. Evaluation of anisotropy by shear wave
"4 / Delta, 8
/ splitting. Geophysics, 50, 142-152.
1985b. Evidence for aligned cracks in the
:,= 1.6 Earth's crust. First Break, 3, 12-15.
d Ion , - - • 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 permeabil-
ity for sandstones with aligned cracks. Interna-
o.0 i /7 tional Journal of Rock Mechanics and Mining
o.1 0.2 o.3 0.4
Crack Density
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
Fig. 12. Thomsen's (1986) anistropy parameters for sandstones with systems of aligned cracks. Pre-
Crosland Hill sandstone sample during the crack sented at the Geophysical Society of London,
closing cycle as a function of (a) stress and (b) crack Borehole Research Group, Conference on Devel-
density of the aligned cracks. opments in Petrophysics, Sept, 1995.
NISHIZAWA,O. 1982. Seismic velocity anisotropy in a
medium containing oriented cracks--transversely
dry sandstones suggest that the anisotropy isotropic case. Journal of the Physics of the Earth,
30, 331-347.
parameter 6 is the most sensitive to the
NUR, A & SIMMONS,G. 1969. Stress-induced velocity
crack density. Moreover, for the dry rocks, anisotropy in rocks: An experimental study.
the anisotropy in P-wave velocities is Journal of Geophysical Research, 74, 6667-6674.
greater and more sensitive to the crack POSTMA,G. W. 1955. Wave propagation in a stratified
density than the anisotropy in S-wave medium. Geophysics, 20, 780-806.
velocities. SAYERS, C. M. 1988. Stress induced ultrasonic wave
(2) The results of the numerical study suggest velocity anistropy in fractured rock. Ultrasonics,
that the anisotropy in P-wave velocities is 26, 311-317.
greater when the saturating fluid is very SHAKEEL, A. 1995. The effect of oriented fractures on
elastic wave velocities, attenuation and fluid perme-
compressible (gas) and when the cracks are
abilities of sandstones. PhD Thesis, Imperial
flat (small aspect ratios), while the aniso- College of Science, Technology and Medicine,
tropy in S-wave velocities is almost un- University of London.
affected by the nature of the saturating THOMSEN, L. 1986. Weak elastic anisotropy. Geophy-
fluid. sics, 51, 1954-1966.
(3) For dry inclusions over a large range of Xu, S & KING, M. S. 1989. Shear-wave birefringence
aspect ratios less than 0.1 the resultant and directional permeability in fractured rock.
anisotropy is hardly affected by a change in Scientific Drilling, 1, 27-33.
the aspect ratio. & - - 1992. Modelling the elastic and
hydraulic properties of fractured rocks. Marine
(4) The results of the numerical and experi-
and Petroleum Geology, 9, 155-166
mental studies suggest that elliptical aniso- U H R I G , L. F & VAN MELLE, F. A. 1955. Velocity
tropy will be observed in a m e d i u m anisotropy in stratified media. Geophysics, 20,
containing aligned ellipsoidal inclusions of 774--779
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 is that although they are associated, varying one
employ a system of transducers making similar of them does not vary the other. For example,
measurements, spaced out along the logging invasion depth and Rt (resistivity) are orthogo-
tool. nal, because varying Rt does not affect the
The main reason for this is to provide invasion depth, and vice versa.
measurements with multiple depths of penetra- However, if we measure resistivity using a
tion in order to compensate for the effects of dual induction tool, the deep and medium
some disturbance to the measurement. This measurements are not orthogonal since varia-
disturbance can have a multitude of origins, tion in R t affects both deep and medium logs.
such as the borehole itself, its size, fluid nature, The tornado chart shown in Fig. 1 is an
caliper fluctuations, etc., or near-borehole effects attempt to transform the measurements into an
such as invasion. orthogonal set, the tornado being a skewed
The compensation relies on the disturbance orthogonal co-ordinate system. All dual or
being common to the array of transducers, and multiple measurements are intended to achieve
requires a model to describe the disturbance (e.g. similar objectives. It will, however, be noticed
a step invasion profile). that the transformed orthogonal pair has two
The multiple spacings employed have differing parameters--we get depth of invasion as well as
vertical resolutions, and much effort in recent Rt. This value of depth of invasion is an example
years has been spent optimizing this resolution of the orthogonal Complementary Parameter.
by ensuring that boundary information is not
lost. The V E C T A R (Vertical Enhancement by Linear perturbation and sharpness
Combination and Transformation of Associated
Responses) computational technique is one such Let us consider an observation O, looking at a
method (Elkington et al. 1990). true value V, subject to a perturbation P. Let us
In this paper, we will consider the converse of suppose we perform all the chart book correc-
this combination method and develop equations tions we can (borehole size, etc.) but we are still
which are not dependent on a pre-imposed left with a perturbation we cannot measure
model but are very general. In the process of directly. Let us further assume that the pertur-
doing this, we will see that another parameter is bation is reasonably small and that it disturbs
revealed that is orthogonal to the true value that the observation from its true value in a linear
is being examined. way. We can then write:-

Orthogonalization 0 = V+ KP (1)

A definition of an orthogonal pair of parameters where K = t h e proportionality constant i.e. the

SAMWORTH,J. R. 1998. Complementary functions reveal data hidden in your logs. 159
In." HARVEY,P. K. & LOVELL,M. A. (eds) Core-Log Integration, Geological Society, London,
Special Publications, 136, 159-171
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 cant, as the ratio of perturbation rates is easier
with different transducers but subject to the to calculate, and additionally the rates can
same perturbation we get:- change their absolute values without invalidat-
ing equation (4) as long as the rate ratio is
0 1 = V-t- K1P (2) unchanged.
We can also solve equations (2) and (3) for the
02 = V+ K2P. (3) perturbation P by eliminating V:-

We can eliminate the perturbation P from the /9__ 0 2 -- O1 (5)


two equations and solve for the true value V. K2 -- K1
With some re-arrangement, we then get:
This parameter is the Complementary Parameter
and is orthogonal to the true value. It can often
V -- O 1 ~- (O1 - 0 2 ) (4) be numerically scaled into some useful unit but
is frequently ignored.
K1 If the perturbation is due to a variety of
different effects they become lumped into a
This is arranged in the following form:- correction that cannot be assigned an explicit
physically meaningful value, so P becomes
(True value) = (Observed value) + (Correction). Unsharp. This is the price of getting a good
assessment of the true value, V, which is Sharp.
It is important to note that the correction It is, however, often the case that many of the
depends on the two observations and the ratio lumped parameters are constant over the length
of the perturbation rates K2/K1 and not the of the borehole. If these values can be ascer-
individual rates themselves. This is very signifi- tained independently, and any one of the
HIDDEN DATA IN LOGS 161

Fig. 2. Mudcake thickness from Density logs.

perturbations varies significantly over the bore- is usually one of optimization. We can encapsu-
hole, this variable parameter can be derived late this principle thus:-
explicitly and a curve plotted. That is, it becomes
Sharp. a computational process on a measurement can
only be justified if the result after the process is
Principle of betterness better than the original.

Before considering examples of the application Alternatively:-


of Linear Perturbation it is prudent to consider
our objectives. The main objective is to improve if a measurement can be improved by applyNg a
the quality of a measurement, not necessarily to computational process it is usually worth doing.
make it absolutely correct, because this depends The result does not have to be correct, only better.
on the quality of our assumed model. We can
often become unnecessarily obsessed with cor- An example of this principle at work is the
rectness, whereas the log interpretation process Compensated Density Log. In the presence of a
162 J.R. SAMWORTH

0 API 250 2.0 GM/CC 3.C


1 inches 11 -0.25 0 0.25
i
GM/CC
'%

_.~~176 h

C~ sJ

,~rill-pipe) Densi ~' Density


~,/Correction

-[ ~_.Gamma Ray

6,5O (

_i
_I
I
_i
I
-I
I
-I
1
-It
DEPTH BA~ED DATA - MAXIMUN S A M P L I N G INCRE~4ENT I O C N . P J E ~ ON AT 20:48

FILENANE: .CIB RUN I D : PLOTTED ON 0 7 - J U N - I ~ AT 0 9 : 3 7

Fig. 3. Compensated Density through casing.

mudcake, the corrected log is usually more We then get:-


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
pt=pn+ 11 ]~ (PL--PS) (6)

often occur in practice, especially in slim wells.


We can see that we must, therefore, apply the where pt = True Density
technique circumspectly. PL = Long Spacing Density
Ps = Short Spacing Density
Application of linear perturbation Ks & KL = Perturbation rates.

We will now consider several applications of the This equation is identical to that derived by
theory. applying the geometric factor theory to density
logs (Samworth 1992).
Density logs If we wish to explore the complementary
parameter we need to set up the original
We can apply equation (4) directly to the long equations. Density logs can be expressed in
and short-spacing density logs. 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 (7) A log of this mudcake thickness is shown in Fig.


2. It is, of course, similar in character to the
Pmc = mudcake density. density correction but is scaled in inches.
Rearranging;
PA = Pt + J(Pmc -- Pt). (8)
Density log through drill pipe
If we approximate J to a straight line function of
standoff, d, Occasionally, circumstances arise when the
borehole stability is so poor that it is not
i.e. J = Kd, (9)
possible to leave the hole open for conventional
we get:-
logging. It is then possible to run a variant of the
PA = Pt + Kd (Pine-- Pt). (10) density tool inside the drill pipe to log the
This is the linear perturbation equation from density of the formation outside the pipe. In this
which (6) can be derived. case, there is no mudcake and we have no
We can now use equation (5) to derive the knowledge of the borehole caliper.
standoff, d A special form of density tool is employed
i.e. which has no preferential circumferential colli-
mation, i.e. it looks all round the hole. The long
Ps--PL (11) and short detectors are calibrated for the
d=(Pmc-P,)(Ks-KL)" 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 comple- assume that areas of high corrections are areas
mentary functions are not now associated with of hole enlargement. If we pursue this assump-
mudcake. tion, we can compute a caliper log using
Figure 3 shows a compensated density log equation (11). This caliper log is shown in Fig. 3.
obtained in this manner. If the original sharp
value required is the bulk density, we do not Caliper from array induction logs
have to assign an explicit meaning to the density
correction; it can remain largely unsharp. Induction logs can be combined in a similar way.
Although unsharp, it is still probably safe to 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 example, that there are three permeable zones,
equations (2) and (3). This has been previously the invasion of the lower two being uniform
explored for slimline array tools (Samworth et since the invasion indicator is of a trapezium
al. 1994). shape. However, the upper zone shows a graded
Figure 4 shows an example of this application. form on its top edge, probably indicating a
A difficult horizontal well was logged with a slim gradation in permeability.
array induction tool, without an opportunity to
run any other log. The borehole fluid was saline Borehole fluid salinity from neutron logs
(.07 52 f2m) and since no caliper was available to
correct the logs, they were apparently quite We can adapt linear perturbation to the dual
poor. (The right hand set of curves in Fig. 4). neutron tool. Dual neutron tools are usually
Since we know the mud resistivity, by assuming designed so that the ratio of the count rates from
that the two shallowest measurements see no near and far detectors is related to the formation
further than the invaded zone we can use linear porosity. The design is normally such that the
perturbation to calculate an apparent caliper. sensitivity of this ratio to such things as borehole
This caliper, shown in the left-hand track, was fluid salinity is minimized. However, a comple-
then used to correct the deeper reading measure- mentary function can be calculated specifically
ments. A much more systematic log then results to be sensitive to this salinity.
(in the centre track of Fig. 4). This is a form of This can be seen in Fig. 6 where we have cross
optimization of the induction logs, and it leads plotted the short spaced count rate against the
to a better product without necessarily being long spaced. On this plot, all points on a straight
absolutely correct. line through the origin have the same ratio, and
this represents the same porosity. A line can also
Invasion indication from induction logs be drawn for constant salinity but with porosity
varying. This mesh is a skewed orthogonal
Figure 5 shows what can be achieved with a system, as described earlier.
simpler slim dual induction tool. Only two In setting up the linear perturbation equations
conductivities are measured here, but unlike in this case, we shall use the two count rates V1
the previous example, the caliper is known as and V2, which are the unperturbed values. So we
well as the mud resistivity, and the appropriate now get:
corrections can be applied.
When linear perturbation is applied here, as O1 = V1 +K1P (12)
well as deriving Rt, we get a lumped comple-
mentary function associated with both Rxo and 0 2 = V 2 + K2P. (13)
invasion diameter. This is shown as the shaded
curve in Fig. 5 and gives some indication of the We can establish a relationship between 01 and
character of the invasion. It can be seen, for 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 167

!0.0 (SalinityIndicator) 1.0


0 API I00 5 inches 1~ 0 B/E 10 30 LST % - 10
L

6600

6700

9 ) 6800
Gamma \
J Caliper
fk 6900

! i~ vooo
Bit Size 7100

: k "7:~00 ~ Ne r--~"
r 7~0o
\ ? ,r .
I" 7400 Salinity • ~.~ =
r Indicato~ .~.,.~.s
r
~, 7gO0
?
I
~'- ~ 7600
I
\) 77OO
,L ; "~
,,.&~ 9

b- r
.-.,~ I
d
-,ooo
7900
e-J
_N 2~
i)' 8100

8400-

)I 8~00
-g
iI 8600
2~."
~ bi) 87oo

f f ~ '8800
%-N
i
~" . Y j ogoo

Fig. 8. Borehole salinity from Neutron logs (2).


168 J.R. SAMWORTH

I
0.0 (SalinityIndicator) 1.0 I
API 100l inches 1 0 B/E 10]30 LST% -loI
6600

6700

6800

6900

7"Caliper 70OO

7100
~-Bit Size
7200 sNeor PE
tm~. 1

7300
Salinity
7400

7~00

7600

7700 t',

7800
7
7900

8000

8100

8200

8:S00
<
8400

8~00

8600

8700

8800

18900
I

Fig. 9. Borehole salinity from Neutron logs (3).


HIDDEN DATA IN LOGS 169

API 150 3 inches 13 0 (SalinityIndicator) 10.2 ohm-m 2000


,/
......

, ~-.~ " ' B i t Size

I5 2 0 0 ~~ ~ ~ ~ ~ Medium

i ' _- ~ "Salinity t~Dee p


, A..~Caliper
~' Gamma 5300 _-- ,~ Indicator # ~

154oo E

5,5oo

-!
b 5600 -

5"700 :i
b

,58oo --

OE]>THBA~s DATA- NAXIMUHSAMPLINGI ~ T 1004. RECOROEDON AT 14:2:5


FTLENANE: .CZB RUINI0: PLOTTEDON19-,JUN-1996AT 14:07
4'

Fig. 10. Borehole salinity in a horizontal well (1).

K1 Kx V2). previously been explored in some detail (Sam-


0 1 "-" ~22 O 2 "~- ( Vx - - ~ 2 (14)
worth, 1991). If, however, we can identify the
positions of O] and 02 on the ratio line, we can
This is a straight line equation of the form identify which line of constant salinity we are on,
and we can then estimate the borehole salinity.
O1 = m 0 2 + (constant). (15) Some examples now follow to show the
For the particular case of ratio processing: usefulness of this complementary parameter.
Figures 7, 8 and 9 show sets of logs, on a
K1 01 V1 compressed vertical scale in a dolomite reservoir.
m = / ( 2 - 02 -- V2" (16)
The field was being produced by an injected
Since we actually observe Ol and O2, m can be waterflood, and the wells were close to each
calculated and we can migrate along the line other. The reservoir section is the whole of the
until we reach V] and V2. This method has lower halves of the wells where the gamma ray
log activity increases.
170 J. R. SAMWORTH

) API 150 3 inches 13 0 (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 - MAXIWKJM SAMPLING INCREMENT I OCM. RECORDED ON AT 1 3 : 3 8

FILENAJWE : .ClB RUN I D : PLOTTED ON l O - J U N - 1 9 9 6 AT 1 1 : 1 ' : ;

Fig. 11. Borehole salinity in a horizontal well (2).

Figure 7 shows a well where the well fluids Figure 10 shows a horizontal well where there
were static. The salinity indicator shows low are several intervals with an anomalous Array
salinity, i.e. oil, above a high salinity sump in the Induction response (e.g. at 5810-5910). The
reservoir section. Figure 8 shows a log taken neutron based salinity indicator shows high
with the well flowing, i.e. the injector had not levels at these points indicating water plugs in
been turned off. Here the salinity profile is an otherwise oil-filled well. The nuclear logs are
inverted, but the inversion starts several hundred shown in Fig. 11 for reference. Comparison of
feet into the reservoir. the logs with a plot of the hole trajectory shows
Figure 9 is similar to Fig. 7, but with a blip at depressions at these points, the salinity indicator
a similar place to where there is a change in Fig. showing that these are full of water.
8.
The conclusion from these logs must be that Conclusion
the waterflood is breaking through at the top of
the reservoir and not efficiently sweeping the There is much information to be had from well
lower levels. logs by interpreting them in a slightly unconven-
HIDDEN DATA IN LOGS 171

tional way, so it is imprudent to discard any References


data, especially raw data.
The linear perturbation technique is comple- ELKINGTON,P. A. S., SAMWORTH,J. R. & ENSTONE,M.
tely general, and does not rely on any particular C. 1990. Vertical enhancement by combination
physical model. It is applicable to a wide variety and transformation of associated responses.
of logs where multiple measurements of a similar Transactions of the 31st Annual Logging Sympo-
sium, SPWLA, Paper HH.
type are made. The method also produces a SAMWORTH, J. R. 1991. Algorithms for compensated
complementary parameter which can be very neutron logging--57 varieties. Transactions of the
useful in revealing effects not apparent on the 14th European Logging Symposium, SPWLA,
normal logs. Paper A.
- - 1992. The dual-spaced density log, character-
istics, calibration and compensation. Log Analyst
This paper illustrates some of the work carried out by 33, 4249.
the Research and Development Department of Wire- , SPENCER, M. C., PATEL,H. K. & ATACK,N. A.
line Technologies Ltd, and grateful thanks are given to 1994. The array induction tool advances slim hole
that company for permission to publish. 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 ex- taken from great depth which contain stress
ploitation of petroleum and geothermal energy relief microcracks. The main objective of this
resources require a knowledge of the in situ state study is to contribute to the development of the
of stress. Such data are required for determining technique and to compare the results obtained
borehole stability in the drilling phase, for with those from other proven methods.
avoiding solids production problems and for For the purpose of this study, specimens were
hydraulic fracturing stimulation in the produc- prepared from core samples taken from various
tion phase, and for reservoir characterization in petroleum wells in the Irish and North Seas. The
reservoir engineering. A common method for samples comprised ten sandstone specimens
obtaining in situ stress data from great depth is from depths lying between 1361 and 1422 m
indeed by hydraulic fracturing or, alternatively, and between 3232 and 3304 m, eight of which
sleeve fracturing (Desroches et al. 1995). These (SSA) were oriented and so could be used for
techniques, however, possess certain disadvan- determining the actual orientation of the in situ
tages with regard first to cost and second to stresses. The other two samples (SSB) were not
technical considerations in fractured formations, oriented; they are, however, considered valuable
deviated well bores and high pressures and high for further comparison studies. Descriptions of
temperature formations. Results obtained from the rock samples are provided in Table 1.
these methods are often influenced and biased by
stresses close to the well bore and hence do not Differential stain analysis (DSA)
reflect the governing in situ stress field. To
overcome some of these problems, the technique Technique
of differential strain analysis (DSA) has been
developed. Since it was first suggested by DSA is a method based on the existence of
Strickland & Ren (1980) as a tool for in situ oriented microcracks generated within core
stress determination, DSA has been used fre- samples as a result of stress relief processes
quently; successful applications have been re- which occur during the core process and
ported by various investigators (Dey & Brown recovery of the cores from depth. Evidence of
1986; Dyke 1988; Oikawa et al. 1993). the existence of this type of microcracks has
Nevertheless, considerations of the length of been reported by various investigators using
time required for a DSA test have led to efforts different approaches (Kowallis & Wang 1983,
to find alternative methods. Early studies (Yale amongst others), such as comparing images
& Sprunt 1989) utilized the phenomenon of from scanning electron microscopy (SEM),
ultrasonic S-wave splitting on rock specimens studying P- and S-wave velocities adjacent to

WIOARSONO,B., MARSDEN,J. R. & KIN6, M. S. 1998. In situ stress prediction using differential strain 185
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
186 B. WIDARSONO ET AL.

Table 1. Description of sandstone samples

Sample Depth Grain size Comments


(m) (mm)

SSA-1 1361 0.5 Well cemented, very weak bedding


SSA-2 1363 0.1-1.0 Well-cemented, poorly-sorted, weak bedding
SSA-3 1364 0.5-1.0 Fairly well-cemented, strong bedding
SSA-4 1366 0.54).75 Well-cemented, no sign of bedding
SSA-5 1374 0.5 Well-cemented, weak bedding (possible micaceous laminae)
SSA-6 1399 0.14).5 Dark red, very weak bedding
SSA-7 1409 0.14).5 Well-cemented, weak bedding/laminae
SSA-8 1422 0.5 Well-cemented, strong bedding
SSB-1 3232 0.14).5 Well-cemented, no bedding/laminae
SSB-2 3304 0.14).5 Well-cemented, no bedding/laminae

the borehole and in the laboratory, and employ-


ing differential strain analysis itself. For exam-
ple, Teufel (1983) observed anisotropy of P-wave
velocity measurements and correlated this with
results from the anelastic strain recovery techni-
que (ASR).
Basic assumptions of the technique are pre-
sented 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. Conse-
quently, when the sample is compressed hydro-
statically, the resulting strains will show
preferential orientation in magnitudes which
are proportional to the microcrack densities. A
further assumption is that all microcracks Fig. 1. Axes system and strain gauge orientations.
existing within a tested sample are of stressrelief
type, or at least that all pre-existing microcracks
do not affect sample deformation significantly. gauge readings are made during the test.
Hydrostatic pressure on the specimen is first
Experimental procedure increased in steps of 200 psi to 21 MPa (3000
psi), then in steps of 500 psi to 55 MPa (8000
Three basic steps are followed in specimen psi). During this loading procedure, a transition
preparation: machining the specimen, attaching from microcrack closure to intrinsic elastic
the strain gauges and encapsulating the speci- deformation is generally found to occur. From
men. Each step in specimen preparation must be 55 MPa (8000 psi) to the maximum pressure
performed carefully in order to prevent genera- (usually around 83 MPa-12000 psi) 1000 psi
tion of new microcracks in the rock sample. Flat increments are usually chosen since microcrack-
surfaces are machined on each specimen in at free linear stress-strain behaviour generally
least three orthogonal directions, by first sawing occurs in this range of pressures. A strain
them with a diamond saw and then by hand gauged and sleeved fused silica specimen of
lapping or surface grinding. After ovendrying at known physical properties is also tested adjacent
35 ~ strain gauge rosettes are attached to the to the rock test specimen to provide any
specimen following the arrangement shown in corrections necessary for environment-related
Fig. 1. The rock sample is finally encapsulated non-linearities in the specimen and strain gauge
with epoxy resin in an elastomer membrane. responses.
The strain gauged and encapsulated specimen
is placed in an oil-filled pressure vessel and left Procedure of analysis
for approximately 24 h in order to ensure
temperature stabilization. This procedure avoids Additional input data is required for analysing
any temperaturerelated fluctuations while strain DSA measurements, including: vertical in situ or
IN SITU STRESS PREDICTION 187

Pressure medium through which it propagates remains


(MPa) relatively novel despite its origins in studies of
100 earthquake seismology. Despite the abundance
of observations, it is still not clear exactly what
80

60
/ causes S-wave splitting in the Earth's crust
(Crampin & Lovell 1991), although it is gen-
erally taken to be caused by aligned disconti-
nuities. Crampin (1978) recognized that most
rocks in the crust are likely to contain disconti-
40
~G a u g, e _ _ _, _ , 6 nuities, and that S-wave splitting is probably the
most diagnostic feature of wave propagation in
2O such anisotropic rocks.
Attempts have been made to relate S-wave
0 splitting to the orientation of in situ stress-relief
0 1 2 3
Strain(millist rains) induced microcracks in cores taken from great
depths. Yale & Sprunt (1989) utilized ultrasonic
S-wave splitting on oriented core samples, and
Fig. 2. Typical compression curves from DSA test. concluded that this approach can be used to
predict the orientation of the major horizontal in
situ stress.
overburden stress, /n situ pore pressure, and Shear-wave splitting results from the division
Poisson's ratio for the tested rock. The orienta- of a polarized S-wave into two separate polar-
tion of the reference line with regard to the axes ized S-waves travelling at different speeds when
of the cubic specimen is also required if the true a source of anisotropy is encountered in its path.
orientation of the in situ stress field is to be When a plane polarized S-wave is propagated
determined. through a medium containing a set of aligned
The strains recorded by the data logger are microcracks it will be split into two orthogonally
fitted by a series of curve-fitting programs to polarized S-waves, with one polarized parallel to
produce the pressure-strain curves by way of a the plane of the microcracks and the other,
quadratic fit using five adjacent data points travelling at a lower velocity, polarized perpen-
(Dyke 1988). A microcrack closure strain tensor dicular to the plane of the microcracks. Garbin
is obtained from these quadratic compression & Knopoff (1975) proposed a theory to explain
curves (Fig. 2) for each of the specimens tested. the velocity variations caused by a single set of
To create a complete microcrack strain tensor, parallel cracks which is based on scattering of
six components only are required from the elastic waves by penny-shaped cracks. The
twelve strain gauge measurements. This permits theory explains the variation of S-wave velocity
a statistical analysis to be made of the redundant with changes in ray path angle and wave
data. From the microcrack strain tensor, the polarization relative to the crack plane. The
principal microcrack strains and their orienta- degree of splitting is related to the time lag
tions relative to the reference line are calculated between the arrival of the two waves at the
using the method proposed by Strickland & Ren receiver. The degree of S-wave splitting and
(1980). The ratios of principal strains are taken hence velocity anisotropy increases with an
as the ratios of the principal effective stresses increase in crack density.
and, by a series of tensorial transpositions to
vertical and horizontal planes, the principal Experimental procedure and analyses
strains can be converted to principal in situ
stresses using values of overburden stress, pore As part of this study, S-wave splitting tests were
pressure and Poisson's ratio. Furthermore, the conducted on the same specimens as used in the
true orientations of the stresses can be deter- earlier DSA tests, except in the case of one which
mined if the true orientation of the reference line exhibited such a poor degree of cementing that
is known. satisfactory acoustic coupling between specimen
and transducers could not be achieved. The
Ultrasonic shear wave splitting (USWS) elastomer sleeves were removed from the DSA
samples tested earlier and the latter were re-cut
Technique with flat surfaces perpendicular to the vertical
(Z-axis) and horizontal (X- and Y-axes).
The use of acoustic S-wave splitting (birefrin- They were then oven dried at 35~ so that the
gence) as a source of information regarding 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 Azimuth Dip 0"2/0"1 Azimuth Dip 0"3/0"1 0"3/0"2 Azimuth Dip
1 (0)2 (0)2 (0)2 (O) (0)2 (o)

SSA-1 1.005 204N 80 0.892 71N 7 0.849 0.955 340N 7


SSA-2 1.017 106N 73 0.824 3 08N 17 0.807 0.982 219N 1
SSA-3 1.010 116N 76 0.877 213N 2 0.832 0.948 304N 14
SSA-4 1.130 74N 59 0.939 351N 4 0.805 0.858 262N 26
SSA-5 1.000 38N 87 0.832 293N 3 0.817 0.982 203N 3
SSA-6 1.059 331N 66 0.760 52N 2 0.676 0.983 140N 23
SSA-7 1.001 229N 86 0.902 326N 0 0.824 0.914 57N 4
SSA-8 1.004 164N 81 0.866 20N 7 0.835 0.966 290N 6
SSB-1 1.004 240 83 0.888 132 2 0.763 0.858 42 7
SSB-2 1.008 179 74 0.928 88 1 0.879 0.948 358 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 180 ~, measured relative to the reference line, in
same as those used in the DSA tests, in order to increments of 15~ At each 15~ step, the transit
maintain compatibility between the DSA and S- time of first arrivals and waveforms in digital
wave splitting results. form are recorded. After each test, the S-wave
USWS measurements are performed by rotat- transit time is converted to velocity.
ing the rock specimen containing stress relief
microcracks while pulses of planar S-wave are Experimental results
transmitted parallel to the specimen axis under
nearatmospheric pressure conditions. In the D i f f e r e n t i a l s t r a & analysis
presence of aligned microcracks, S-wave first
arrivals observed by the receiving transducer Results of the in situ stress predictions for all
(polarized parallel to the polarization of the specimens tested are shown in Table 2, in which
transmitter) show changes in magnitude as the stress magnitudes are presented as ratios in
specimen is rotated. When the direction of order to provide a comparison of results. Figure
polarization of the transducers is parallel to 3 shows plots of the results for SSA sandstones
the aligned microcracks, the S-wave first arrival in the form of an equal-angle stereonet (lower-
time is a minimum. Conversely, when the hemisphere projection). Trends of the stresses
transducer polarization is perpendicular to the determined from the oriented cores of the SSA
aligned microcracks, the first arrival time is a specimens are given in degrees measured clock-
maximum. This direction is that of the greatest wise from North, whereas those obtained from
stress relaxation and hence is the major in situ the two unoriented SSB sandstones are simply
stress direction. measured clockwise from an arbitrary reference
Each specimen was first assembled between line. Consequently, the results for SSB samples
pairs of broadband transducers having S-wave are not plotted on stereonet projection, since no
frequencies in the range 300 to 800 kHz (as common reference line exists among the speci-
described by King et al. 1995) with a proprietary mens tested, and such plots could imply
visco elastic S-wave couplant and a stress of 2.5 misleading relations.
MPa applied to the transducers to provide good The results shown in Table 2 and Fig. 3 show
acoustic coupling. This level of axial stress has that the major principal stress (o.~) lies near
been shown experimentally to have a negligible vertical, to within 0 ~ to 23 ~ This result can be
effect on cracks oriented sub-parallel to this regarded as sufficiently precise for a technique
direction of propagation of S-waves. At the start based on stress relief microcracks, since it is
of a test, each specimen is arranged so that the common that the orientations of the micro-
transducer polarization is in the Y-axis direction cracks are modified by grain scale inhomogene-
and the propagation of acoustic energy is either ities. The similarity in magnitudes between
in the Z-axis direction (vertical) or X-axis major (o1) and vertical (i.e. overburden) stresses
direction (horizontal). During a test, the speci- (o'v), represented by the ratio o'1/o'v being near to
men is rotated through an angle between 0 ~ and unity, also indicates that o'1 lies in the vertical
IN SITU STRESS PREDICTION 189

North between approximately 2% and 7% with regard


f to the principal stress ratios from any single
DSA tests, and an error 'cone' of approximately
1 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
2 6
be seen that the combined results yield a scatter
of approximately 30 ~ in the azimuths of the
b
horizontal (i.e. intermediate and minor princi-
7%o8~3 2 pal) stresses.
1 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
3 have necessitated the microcrack densities and
0(~. 2 stress magnitudes being isotropic in the hori-
1,2,3 ... sample number A~ 3 zontal plane. Whilst this is possible, it is only so
for those relatively few cases where the magni-
Fig. 3. Lower hemispherical projection showing in sltu tudes of the minor and intermediate stresses are
stress orientations from DSA tests on sandstone SSA. exactly the same. For all other combinations
within the limits or error, the general directions
of the principal stress orientations are as seen in
direction. It can also be concluded from the Fig. 3 and only the magnitudes vary. Thus,
results for SSA sandstone that the depths from although the intermediate and minor stresses are
which the core samples were recovered are not very close in magnitude, the DSA method has
far above the depth at which o.H becomes equal still been able to identify the general orientations
to o-1, as indicated by the closeness of the ratios of the stresses.
o.2/o.1 and o.3/0-~ to unity. This interpretation is in From the hemispherical projection, it is clear
accordance with global data for the vertical that one of the two horizontal stresses lies within
horizontal stress ratio versus depth compiled by the range 290 ~ to 350 ~ from North, with the
Brown & Hoek (1978). scatter in azimuth due probably to variation in
In the case of the intermediate and the minor the orientation of the stress-relief microcracks
stresses, 0-2 and 0-3, it is evident that these lie in a caused by grain-scale heterogeneities. On the
horizontal plane, or at least sub-parallel to other hand, the other horizontal stress lies
horizontal. However, it is obvious from the within the range 20 ~ to 71 ~ from North such
scatter observed that the orientations of the two that, with the two horizontal stresses being
horizontal stresses are interchangeable. This is similar in magnitude, the intermediate principal
understandable, since the two stresses are very stress (o.H) can lie in either of the ranges 290 ~ to
close in magnitude, and each has caused a 350~ or 20 ~ to 71~ Arguably, therefore, o.H
similar degree of oriented microcracking in the can lie in either of the ranges 310+10 ~ N (or
samples. If these ratios (0"3/0" 2 column in Table 2) 130+10~ or 220+10~ (or 40+10~ These
are averaged, a minor-intermediate stress ratio results are in reasonably good agreement with
of 0.95 (with the exclusion of SSA-4, which is results of an earlier study reported by Desroches
significantly different from the others) is ob- et al. (1995), who conducted analyses of the in
tained. Thus the difference of stress relief situ state of stress in this area using DSA on
microcrack density in the two principal direc- samples from the very same core sections as
tions is minimum and, taking into account tested in this study, as well as using hydraulic
experimental and analytical error (e.g. determin- fracturing and sleeve fracturing in downhole
ing ranges for slopes of the pressure-strain tests in the same well and at the same depths.
curves), the implied stresses could be inter- The earlier combination of the results from these
changed in direction and magnitude. At this three techniques indicated o.h to lie in the range
point, it is worth noting that, with a mere 5% N65~ ~ Note also that this earlier study
error bar to represent the errors, both trends and showed that stress data from DSA analyses on
magnitudes of the two horizontal stresses are these core sections were not influenced by slight
interchangeable. In fact statistical analysis of anisotropy in the samples nor by variations in
each over-determined DSA dataset (which was the elastic properties or rock strengths over the
necessary to obtain the best fit tensor of crack cored intervals.
closure strains) yielded a maximum error of 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 60 63 66 69 72
Transit time (pSec) Transit time (pSec)

Fig. 4. S-wave waveforms for sample SSA-1 at various Fig. 5. S-wave waveforms for sample SSA-I at various
rotation angles. (Propagation in vertical Z-axis, and rotation angles. (Propagation in horizontal X-axis, and
polarized in Y-axis direction at X-axis reference line. polarized in Y-axis direction at X-axis reference line.
Arrows indicate detected first peak). Arrows indicate detected first peak).

sandstone specimens is similar to the results for in other directions. Although this behaviour is
SSA, even though the true orientations cannot exhibited by all specimens tested in this study,
be determined due to the unoriented nature of there is one case (SSA-8) where signal attenua-
the core. Since the core is from vertical bore- tion was so severe that it proved impossible to
holes, it can be inferred from Table 2 that the pick the first arrival time. The degree of
major principal stress lies near-vertical; conse- attenuation, nevertheless, varies from one speci-
quently the other two principal stresses lie near- men to another in a manner related to the state
horizontal. of cementation of the specimens (Table 1).
For the purpose of predicting in situ stress
S h e a r w a v e splitting orientations, transit time values were identified
and selected from the waveforms. Since the S-
During each test, a set of waveforms was wave signals observed during the experiment
recorded for each 15~ rotation relative to the vary in quality due to different degrees of
reference line (X-axis for the vertical, and Z-axis attenuation, it was found that the S-wave
for the horizontal wave propagations). Figures 4 velocity calculated from the first peak (or
and 5 illustrate examples of waveforms recorded trough) is more reliable for interpretation than
during measurements on SSA-1 in the vertical the group velocity calculated from the first
(Z-axis) and horizontal (X-axis) directions. arrival.
As expected, the plots show that the transit- Figure 6 shows examples of the variation in
time (At) varies with rotation angle. This first peak velocity with rotation angle as seen for
variation in At can be considered as an sample SSA-1. The velocity plots are shown with
indication of the existence of an oriented set of an error bar of +0.5 %, estimated as represent-
cracks, or at least (provided that a homogeneous ing the confidence in picking transit time values
background pore system exists) is more influen- (+0.125 #s) from waveforms due to variations in
tial in reducing S-wave velocity than any other signal quality and oscilloscope resolution. In
sets of microcracks with lower density oriented general, the velocities plotted against rotational
IN SITU STRESS PREDICTION 191

S-wave [first peak] minimum velocity direction. For horizontal


velocity (m/s) propagation (X-axis), the vertical (or sub-paral-
2000 lel) in situ stress (o-v) coincides with the direction
a) Z-direction propagation
of maximum velocity, as in the case of 0-H for
Y-axis polarization
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%

Sample crI inclination O"H azimuth ~h azimuth


18oo 1~o ' 18o ' 2~o ' 24o 270 300 (~ vertical) (o)l (o)l
o from North

S-wave [first peak]


SSA-1 0 7 6 ( 2 5 6 ) N 166(346)N
velocity (m/s) SSA-2 75 5 3 ( 2 3 3 ) N 143(323)N
SSA-3 2
2100
b) X-direction propagation SSA-4 0 83(263)N 173(353)N
Y-axis polarization SSA-5 0 197(17)N 107(287)N
SSA-6 0 120(300)N 210(30)N
SSA-7 0 343(163)N 253(73)N
2000 9 SSA-8 - 106(286)N 196(16)N
SSB-1 - 105(=285) 15(--195)
SSB-2 0 75( = 255) 345(= 165)
1• 9

1Azimuths measured clockwise with respect to North


19oo t
~b I
6;
i

120 9'o
90
I

150 180 (Z-axis) or, for unoriented cores (SSB), clockwise from
o from vertical reference line (X-axis).
2 = Poor acoustic coupling due to poor cementation of
Fig. 6. Variation of S-wave first peak velocity with the rock.
rotation angle for sample SSA-1. (a) Vertical Z-axis
propagation, polarized in Y-axis direction; (b) hor- Table 3 summarizes the results of the S-wave
izontal X-axis propagation, polarised in Y-axis direc- splitting measurements. Note that USWS mea-
tion. surements on SSA-3 and SSA-8 (Z-axis propa-
gation) were not carried out, due to poor
transducer-sample acoustic coupling. The results
angle result in a sinusoidal pattern, indicating show that one of the principal in situ stresses lies
the presence of azimuthal velocity anisotropy in the vertical direction, even though accuracy is
over 90 ~ As shown in Fig. 6, similar behaviour limited by the 15 rotational sampling. The
is also observed in both the vertical and vertical stress lies within 7.5 of the peak of the
horizontal directions, indicating the presence of velocity profiles shown in the plots (except for
aligned stress relief microcracks of different sample SSA-2). The same resolution limit
densities as the source of anisotropy. Generally, applies to 0-H. Furthermore, from comparisons
for SSA sandstone specimens, S-wave velocity between the horizontal and vertical velocity
splitting of 1 to 3.5% and 3 to 13.5% (relative to plots, it is apparent that 0-v is the major principal
the highest velocity) for vertical and horizontal stress, o1, as indicated by relatively larger
propagation, respectively, are observed. For velocity anisotropy for horizontal than for
SSB samples, SSB-1 shows 2.2% splitting, vertical propagation. To illustrate this, 1 to
whereas SSB-2 shows 1.1% and 1.6% velocity 3.5% velocity variation exists over 0~ ~
splitting, respectively in the vertical and hor- relative to the highest velocity for vertical
izontal propagation direction. propagation, in comparison to 3 to 13.5% for
Accepting the hypothesis of a relation be- horizontal propagation in the SSA sandstone.
tween S-wave splitting and in situ stress relaxa- Consequently, 0-2 and 0-3 lie in the horizontal
tion and orientation of stress relief microcracks, plane.
velocity plots such as in Fig. 6 indicate the As seen from DSA, the trends of the
orientations of the in situ stresses. From the horizontal in situ stresses are more easily
velocity profiles, for S-waves propagating in the identified if illustrated on stereonet projections.
Z-axis direction, the major horizontal stress (O-H) Figure 7 illustrates the orientations of the major
lies in the direction at which maximum velocity horizontal in situ stress axes for the SSA
occurs, whereas the minor horizontal stress (o-h) sandstones plotted. As with the DSA stereonet
lies at right angles to it, as indicated by the projections, the orientations are relative to
192 B. WIDARSONO ET AL.

Reference line agreement between trends for 0.I-1 (USWS) and


0.3 (DSA), which is incorrect by definition, since
~ / 7 7 f 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 (approxi-
mately 0.95), such that differences in micro-
1 cracking 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
e C~,H
1,2,3 ... sample number and uniform tensile forces across the rocks at the
granular scale.
Fig. 7. Lower hemispherical projection showing rh Grain scale heterogeneities certainly appear to
orientations from DSA tests on sandstone SSA. 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 ~
North. The projection in Fig. 7 and the from the vertical, whereas the USWS results
corresponding DSA results (Fig. 3), show that show almost all the 0.1 to be vertical (Table 3).
the major horizontal stress can lie either in the Undoubtedly, rotational sampling of 15~ re-
N W - S E or N E - S W directions. At this stage, duces the accuracy of stress orientation, and it is
therefore, no definitive conclusion can be drawn likely that grain scale heterogeneities (reflected
regarding the orientation of 0.iJ. in local preferential orientation of the micro-
cracks) contribute significantly to any disagree-
Comparison of the two techniques ments between the two techniques. Such
heterogeneities certainly have greater effects on
Orientation o f in situ stresses DSA measurements, since the portions of the
specimen measured in DSA are limited to the
In comparing the results from the DSA and the surface areas underneath the strain gauges. This
S-wave splitting, only the orientations of the is much less representative than the rock mass
stresses can be considered. No comparison can tested by the USWS technique. The results from
be made of stress magnitudes, since S-wave S-wave splitting are probably more reliable in
splitting tests do not provide stress magnitudes, this particular case.
even though a limited qualitative assessment can The results for SSB sandstone in Table 4 show
be made using velocity anisotropies. A summary a reasonable agreement between the two techni-
of results of the two techniques are listed in ques, even though the results for SSB-2 exhibits
Table 4. In the table, only the results for aH and about 30 ~ difference between o-2 and 0.H, and
ah from acoustic measurements and 0.2 and 0.3 between 0.3 and 0.h- Although there is no
from DSA are compared, since almost all evidence, this disagreement is probably caused
analyses of S-wave splitting tests have shown by other sources of acoustic velocity anisotropy
that the principal stress 0.1 is essentially vertical such as preferential alignment of sandstone
(i.e. they show a minimum velocity at 90 to the grains.
vertical). The differences in the fundamental concepts
In general, the results for SSA sandstone have of the two techniques provide, nevertheless,
shown that there is a reasonable agreement advantages and disadvantages for both techni-
between the two techniques, although some ques. In S-wave splitting, any acoustic propaga-
inconsistencies appear when individual results tion is influenced by the 'averaged' properties of
are studied. It is clear that only SSA-I and SSA- the whole specimen volume, and therefore
7 exhibit total agreement between trends for ei-i representative of the whole specimen. In con-
(USWS) and o2 (DSA), which is by definition trast, any microcrack strain measured in a DSA
correct for this case (i.e. intermediate principal test represents only the areas covered by the
stress is the major horizontal stress). The rest of strain gauges, which are generally small com-
the samples have, in contrast, tended to show pared to the overall specimen dimensions.
IN SITU STRESS PREDICTION 193

Table 4. Comparison of/n situ stress orientations from DSA and USWS

DSA S-Wave Splitting


0-1 ~ ~ O'H Oh
Sample Azimuth Dip Azimuth Dip Azimuth Dip Azimuth Azimuth
(o)l (o) (o)1 (o) (o)1 (o) (o)1 (o)1

SSA-1 204 80 71 7 340 7 76(= 256) 166(= 346)


SSA-2 106 73 308 17 219 1 53(-- 233) 143(= 323)
SSA-3 116 76 213 2 304 14 2 2
SSA-4 74 59 351 4 262 26 83(= 263) 173(= 353)
SSA-5 38 87 293 2 203 3 197(= 17) 107(= 287)
SSA-6 331 66 52 2 140 23 120 = 300) 210( = 30)
SSA-7 229 86 326 0 57 3 343 (= 163) 253(= 73)
SSA-8 164 81 20 7 290 6 106(=286) 196(= 16)
SSB-1 240 83 132 2 42 7 105(= 285) 15(= 195)
SSB-2 179 74 88 1 358 16 75(=255) 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.

However, in the presence of large discontin- dominant and mutual perpendicular micro-
uities, the reverse is true. Large discontinuities in cracks sets. Although this assumption is logical,
specimens influence acoustic wave propagation, as demonstrated by Charles et al. (1986), DSA
whereas any non stress relief behaviour in the does not provide a direct illustration of the
DSA tests can usually be avoided by strain microcrack system. Direct observation such as
gauge emplacement. Consequently, DSA can scanning electron microscope (SEM) of oriented
produce more reliable results in this case. samples used in conjunction with DSA can,
The DSA results in Table 2 show that the however, provide an insight into microcrack
principal stresses do not lie in exactly horizontal orientations and distributions.
or vertical directions. In other words, the The results of S-wave splitting measurements
induced cracks in most cases are not exactly can provide, to some extent, additional informa-
parallel or perpendicular to the vertical axis in tion on microcrack configurations. Velocity
situ. However, the reasonably good agreement plots, such as shown in Fig. 6, have demon-
between the two techniques has shown that strated that velocity anisotropy can occur
acoustic velocity anisotropy can still be observed between measurements in the vertical and
even though the microcracks dip from these horizontal directions. There are several factors
directions, which is more often than not likely to that can lead to such acoustic anisotropy, and to
be the case. This fact is very important in any S-wave splitting in particular, but it is generally
effort to establish USWS as an alternative accepted that aligned cracks are the major cause
technique for in situ stress determination, bear- of S-wave splitting (Crampin & Lovell 1991). At
ing in mind that it is most likely that grain scale the small scale, such as in USWS tests, there is
heterogeneities can cause local deviations in always a possibility that other sources of
microcrack orientations. anisotropy, such as lamination and crystal
alignment, can contribute to the overall aniso-
tropy. As shown in Table 1, SSA sandstone
Configuration o f stress relief microcrack samples exhibit signs of bedding (between 4 ~ and
system 22 ~ from the horizontal), even though not all
samples were found to exhibit dominant bed-
In DSA the principal strains are determined ding. However, in this study, there is evidence
from analysis of the microcrack closure strain that the presence of bedding planes has not
tensor obtained from a test. The principal strains contributed to the velocity anisotropy for the X-
obtained provide the orientations of the stress axis (horizontal) propagation direction. For
relief microcracks, since it is assumed that the example, the results for SSA-5 (Table 3) show
greatest strain occurs in the direction normal to that, despite the 18.5 ~ dip in bedding (Table 1),
the plane of the microcracks. In other words, it the USWS measurements indicate the presence
is assumed that it is represented by three 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 successfully. In general, individual results of the
with SSA-2, for which the bedding dipped at 22 ~ two techniques are found to be in agreement,
from the horizontal and where measurements in taking into consideration the facts regarding the
the horizontal direction indicated the minimum local in situ state of stress field. The DSA and
velocity to be reached at a rotation angle of 165~ USWS results are also found to be in reasonable
(i.e. 15~ from the vertical, implying Crl acts at agreement with results from hydraulic fracturing
75 ~ from the vertical. Clearly this is incorrect if and sleeve fracturing. Grain scale heterogeneities
one is to take the results of the DSA to be strongly influence the deviation of stress relief
reliable (Table 3); the real cause of the aniso- microcracks, as reflected by the scatter in
tropy is unclear, although it may be caused by horizontal in situ stress orientations shown by
strong crystal alignment. The evidence of results both techniques. Results for SSA sandstone
for SSA-2 and SSA-5 have shown that bedding samples have shown that, in areas with small
planes do not, in these cases, strongly influence differences in principal stresses, a greater num-
velocity anisotropy for the horizontal direction, ber of samples will be required to overcome the
and hence do not confuse the subsequent influence of grain scale heterogeneities.
interpretation of USWS measurements. Comparisons between individual results ob-
The results obtained from the S-wave splitting tained from the two techniques have shown that
tests do, however, tend to support the existence S-wave splitting is a reliable one for determining
of aligned horizontal microcracks. In the vertical orientations of in situ stresses by virtue of the
direction, results of the acoustic test have shown existence of stress relief microcracks. The study
that it is most likely that vertical aligned has also shown that S-wave splitting analysis can
microcracks cause S-wave velocity splitting, be used independently with reliable results. This
since no other apparent causes are observed. confidence, together with simplicity in sample
The question arises whether the suggested preparation and speed in conducting tests, can
presence of two sets of mutually perpendicular be considered as the major advantage of this
vertically aligned microcracks, as implied by technique compared to other techniques based
DSA, can be justified. For this, the results of S- on stress relief microcracks such as DSA,
wave splitting cannot be used, since they do not differential wave velocity analysis, anelastic
show the existence of the second (i.e. the less strain recovery or differential thermal analysis.
dense) vertical set of microcracks (if it does exist, Despite the indicated advantages, the study
its existence is probably 'overlooked' by the has also revealed some disadvantages in the
transmitted acoustic energy and treated as acoustic technique. Its major limitation is its
merely a background for the first and more inability to provide estimates of in situ stress
dense vertical microcrack set). Charles et al. magnitude. Another less important disadvan-
(1986) outline theories of brittle fracture me- tage is the necessity to perform the test only in
chanics which explain crack opening under vertical and horizontal directions without com-
tensile forces (as might occur in the case of promising the simplicity in sample preparation
stress relaxation). If the existence of two sets of and analysis of results. This requires that the
microcracks (one vertical and the other hor- principal in situ stresses always lie in vertical and
izontal) whose generation is related to principal horizontal planes, which is not necessarily true
in situ stress relaxation can be proven experi- in all cases. Theoretically, however, this dis-
mentally, it is likely that a third set of micro- advantage can be reduced by reducing the
cracks (i.e. the second and less dense set of sampling rotational interval, hence enabling
vertical microcracks) also exists, since the more careful examination of the alignments of
processes leading to it are essentially the same sets of microcracks. The fact that stress relief
as those causing microcracks in the other two microcracks are not the only source of S-wave
principal directions. Comparing results for the splitting is another disadvantage. However, well-
two techniques has demonstrated directly that prepared samples can minimize this significantly.
three sets of mutually perpendicular microcracks When the two techniques are employed
exist in a rock material experiencing a process of together, DSA provides information on in situ
relaxation from three in situ principal stresses, stress orientations and magnitudes, whereas S-
provided the relaxation forces