XXX10.1144/petgeo2015-056G. G. Correia & D. J.

SchiozerReservoir characterization of the Norne Field

2016

2015-056 research-articleResearch article10.1144/petgeo2015-056Reservoir characterization using electrofacies analysis in the sandstone reservoir of the Norne Field (offshore Norway)Gil G. Correia &, Denis J. Schiozer

Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016

Research article Petroleum Geoscience

Published Online First doi:10.1144/petgeo2015-056

Reservoir characterization using electrofacies analysis in the

sandstone reservoir of the Norne Field (offshore Norway)
Gil G. Correia & Denis J. Schiozer
Department of Petroleum Engineering, State University of Campinas, PO Box 6052, 13083-970,
Campinas, SP, Brazil
* Correspondence: gil@dep.fem.unicamp.br

Abstract:  The Norne Field reservoir sandstone comprises Early–Middle Jurassic interbedded sandstones and shales to mas-
sive sandstones with some thin continuous cemented interlayers. A detailed characterization of the geological heterogeneities
through electrofacies analysis, together with the simulation grid refinement, has been used to derive representative facies and
petrophysical models (porosity, net-to-gross (NtG) and permeability).
An electrofacies database was created comprising six rock types, ranging from cemented carbonates through shales and
into clean sandstones. In the absence of available cored sections, the electrofacies scheme was validated by the geological
and petrophysical reports of 26 wells using gamma-ray, neutron and density logs. An artificial neural network algorithm
enabled the probabilistic discrimination of the different types of electrofacies, with a sampling rate of 0.125 m. This high-
resolution electrofacies database, together with a high-resolution geomodel grid, enabled us to map the fine-scale hetero-
geneities mainly materialized by decimetre shales and cemented layers that could represent stratigraphic barriers to vertical
fluid displacement.
The high-resolution datasets created in this study will form the working basis on which to perform a probabilistic and
multi-objective history matching guided by production and 4D seismic data, and assisted by geostatistical parameterization
techniques.

Received 10 August 2015; revised 7 December 2015; accepted 18 December 2015

Introduction these facies were also modelled to a high-resolution geogrid along

their petrophysical properties (porosity, permeability and NtG). In
The Norne pilot project is a benchmark case based on real field this paper, it is shown how the definition of a high-resolution elec-
data established between the Center for Integrated Operations in trofacies scheme and a high-resolution geogrid can help to identify
the Petroleum Industry (the IO Center), the Norwegian University and characterize small variations in the reservoir quality, and to
of Science and Technology (NTNU) and Norne Field Operations identify thin shale and carbonate cemented layers (thickness,
(Statoil, ENI and Petoro). The main goal is to provide a real dataset extent and frequency) that can act as vertical barriers to fluid flow,
to different research groups in order to evaluate and compare dif- having a significant impact on the estimation of the effective verti-
ferent methods for history matching and, ultimately, closed-loop cal permeability and on reservoir behaviour. The 3D porosity and
reservoir management (Rwechungura et al. 2010). The Norne permeability distributions become directly related to the geostatis-
Field is located on a horst block (9 × 3 km) in the southern part of tical facies models through specific ranges of uncertainty. The
the Nordland II area in the Norwegian Sea (Fig. 1), and was dis- definition of the reservoir through the NtG also becomes directly
covered in 1991. It is a siliciclastic reservoir that comprises the related to the electrofacies by defining specific ranges of uncer-
shallow-marine and deltaic sands of the Middle Jurassic Fangst tainty to each electrofacies instead of using permeability cut-offs
Group at the top, and the heterolithic tidal units of the Early and Vshale curves, which could be problematic owing to measure-
Jurassic Tilje Formation of the Båt Group at the base (Dalland
ments limitations, thin-bed effects and complex porosity–permea-
et al. 1988; Hammer et al. 2010).
bility correlations (Worthington & Cosentino 2005). Rather than a
The heterolithic tidal deltaic reservoirs comprise an important
deterministic solution provided in the Norne Field benchmark case
petroleum reservoir play. However, their highly heterogeneous
(Rwechungura et al. 2010) and in the absence of a geological
lithology may lead to interpretational problems, where often it is
model, the high-resolution datasets created in this study will form
difficult to differentiate one lithofacies from another (Gupta &
the working basis on which to perform a probabilistic and multi-
Jonhson 2001). This heterogeneity introduces additional chal-
objective history matching guided by production and 4D seismic
lenges when modelling this type of reservoir, including facies,
data (Maschio & Schiozer 2013), preserving the geological con-
three-dimensional (3D) porosity and permeability distributions
sistency. Hence, it is an important contribution to future integra-
(Storvoll & Bjørlykke 2004; Martinius et al. 2005; Nordahl et al.
tion of history matching procedures and closed-loop reservoir
2005), effective net-to-gross (NtG) ratios (Henriquez & Jourdan
management of the Norne Field or in similar sandstone reservoirs
1995; Martinius et al. 2005), the estimation of the sealing capacity
worldwide.
of faults (Martinius et al. 2005), the estimation of the effective per-
meability, and the identification of the producing intervals (Begg
Geological setting of the Norne Field
& King 1985; Martinius et al. 2005; Ringrose et al. 2005).
This work goes towards solving some of these issues in terms of The petroleum system that comprises the reservoir sandstones of
facies occurrence and recognition in the Early and Middle Jurassic the Norne Field are post-rift sediments that were deposited during
sandstones of the Norne Field. Through geostatistical techniques, the continued northwards drift of mid-Norway after the pronounced

© 2016 The Author(s). Published by The Geological Society of London for GSL and EAGE. All rights reserved. For permissions: http://www.geolsoc.org.uk/
permissions. Publishing disclaimer: www.geolsoc.org.uk/pub_ethics
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
2 G. G. Correia & D. J. Schiozer
Fig. 1. Location of the Norne Field in the Norwegian Sea (modified
from Verlo & Hetland 2008).
rifting of the northern North Sea during the Triassic (Swiecicki
et al. 1998). The Norne Field is located in one of the horst blocks
(9 × 3 km) that resulted from the tectonism of the rifting phases.
In the Norne Field, the hydrocarbons are located in the Early–
Middle Jurassic sandstones, comprising the Garn (called the Not 2
Formation after 2006 zonation: Fawke 2008 in Rwechungura et al.
2010), Not and Ile formations of the Fangst Group, and the Tofte
and Tilje formations of the Båt Group (Fig. 2). Nevertheless, 80%
of the oil is located in the Ile and Tofte formations, and the gas in
the Not Formation (Rwechungura et al. 2010). The entire reservoir
thickness ranges from 260 m in the southern parts to 120 m in the
northern region due to increasing erosion to the north, namely in
the Ile and Tilje formations. The Norne Field is a shallow reservoir
(2500–2700 m true vertical depth below mean sea level (TVD
msl)) located in the specific diagenetic interval, representing the
end of the mechanical compaction and the beginning of the chemi-
cal compaction, which reduces the reservoir quality (Storvoll &
Bjørlykke 2004; Martinius et al. 2005). However, most of the
sandstones have good reservoir properties, with porosity in the
range of 25–30%, and permeability between 20 and 2500 mD
(Rwechungura et al. 2010).
The Tilje Formation, which is of Sinemurian–Pliensbachian
age, is composed of interbedded sandstones, shales and siltstones,
often heterolithic, deposited in a nearshore marine to intertidal
environment. The formation thins towards the NE and is absent on
the Nordland Ridge due to erosion (Dalland et al. 1988). An
unconformity separates the Tilje Formation from the overlying
Tofte Formation, probably as a result of tectonic uplift and subse-
quent erosion (Swiecicki et al. 1998). The Tofte Formation
(Toarcian) consists of an expressive coarsening-upwards sequence
of moderately to poorly sorted, fine–coarse sandstones with fre-
quent large-scale cross-bedding (Fig. 2). These sandstones, which
interfinger with the Ror Formation mudstones to the east, were
deposited by eastwards-prograding fan deltas probably as a result
of tectonic uplift to the west (Dalland et al. 1988). The overlying
Fangst Group (Aalenian–Early Bajocian) is marked by an intensely
cemented zone at the base, followed by the Ile Formation,
which is dominated by a generally coarsening-upwards sequence
with fine–medium sandstones interbedded with thinly laminated
siltstones, shales and some carbonate-cemented stringers. It was
deposited in a tidal-influenced delta or shallow-marine setting
(Dalland et al. 1988; Harris 1989). The Not Formation (here
equivalent to the Not 1 zone) is a thin unit of heavily bioturbated
claystones coarsening upwards into shaly sandstones, which
behaves as a sealing layer between the upper Not and Ile formations
(Fig. 2). The deposition pattern reflects a regional transgression that
led to the development of lagoons and sheltered bays at the base
changing progressively to prograding deltaic patterns towards the
top (Dalland et al. 1988). The coarsening-upwards trend, which
continues into the overlying Garn Formation, indicates deposition
during a regression (Fig. 2). The Garn Formation (here called the
Not 2 zone) consists of fine- to coarse-grained, moderately to well-
sorted sandstones with a few carbonate-cemented layers deposited
in progradations of brained delta lobes (Dalland et al. 1988).
The purpose of the electrofacies analysis is for the identification
and high-resolution characterization of the different facies in the
wells. The electrofacies dataset is then used to generate high-reso-
lution 3D facies models through geostatistical techniques.
Methods
The methods used in this study were applied in order to address the
lack of a geological model, core descriptions and facies models for
the Norne Field benchmark case (Rwechungura et al. 2010). These
methods are part of a geological modelling workflow that resulted
in new information on the Norne Field benchmark case, highlight-
ing the electrofacies estimation and the refined simulation grid that
will allow a better control of the fine-scale heterogeneities and
their effects on the reservoir behaviour. The electrofacies database
forms the basis for generating high-resolution 3D facies and petro-
physical models constrained by the facies distribution for the
Norne Field and, eventually, to other sandstone reservoirs.
The major steps used in this study are as follows (Fig. 3): (1)
data preparation; (2) electrofacies classification; (3) simulation
grid refinement; (4) upscaling of well logs and the created electro-
facies to the high-resolution geomodel grid; (5) facies modelling;
and (6) petrophysical modelling (porosity, permeability and NtG)
controlled by the facies models.
The data preparation of environmentally corrected logs involves
gathering all the available geological information regarding the
reservoir or analogues, a common suite of logs for all the wells
used as input data, depth matching of various logs and the creation
of a DIFF curve (Gupta & Johnson 2001) for each well.
The electrofacies classification used in this study is based on an
artificial neural network (ANN) algorithm, inspired by the way in
which brain neural networks process information. A characteristic
feature of the ANN is the capability of assimilating highly com-
plex relationships between several variables, learning the nature of
the dependency between the input and output variables of the train-
ing set provided to the network (Maschio & Schiozer 2014). In
terms of learning features, the ANN can be categorized as super-
vised or unsupervised. The unsupervised method, also known as a
self-organizing network, is used to subdivide the input data into a
number of specified classes. The method looks for regularities or
trends in the input data, making adaptions without being shown the
correct output. In the supervised method, both input and output
data are provided. The error in the network is assessed by passing
the input data through the ANN and comparing the result with the
output data. Mohaghegh et al. (1996) presented one of the first
studies using the ANN concepts in the petroleum industry. Since
then, several studies using the ANNs can be found in the petroleum
literature. In the context of electrofacies analysis, some applica-
tions can be found in published studies (e.g. Chang et al. 2002;
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
Reservoir characterization of the Norne Field
Fig. 3. Workflow scheme of the electrofacies analysis integrated with
the 3D modelling phase.
Qi & Carr 2006; Ma & Gomez 2015). In this study, the unsuper-
vised ANN was used in the electrofacies classification because no
core data were available to train the neural network. In addition,
the difficulties in tuning the training and the artefacts created due
to over-training are common problems when using the supervised
3
Fig. 2. Early–Middle Jurassic
stratigraphy of the mid-Norwegian shelf
(modified from Morton et al. 2009).
The main formations are correlated with
the gamma-ray log of well 6608/10-
C-1 H. The transgressive–regressive
(T–R) cycles and the schematic facies
successions were based on Dalland et al.
1988. mst, mudstone; sst, sandstone.
method. The ANN should be used with caution when constructing
3D models due to over- and under-training problems, being more
useful when integrated with statistical methods (Ma & Gomez
2015).
After obtaining the electrofacies, an essential step is to upscale
the logs and the electrofacies to a high-resolution grid in order to
be able to generate 3D facies and petrophysical models through
geostatistical methods. This high-resolution grid should maintain,
as much as possible, the fine-scale heterogeneities seen in the well
logs.
For the facies and petrophysical modelling, stochastic methods
were used that are able to generate multiple realizations under cer-
tain uncertainty ranges. The 3D models of each petrophysical
property were restricted by the 3D facies distribution and by cor-
relation factors between each property that were obtained from the
data analysis process.
Application and results
The methods were applied to the real dataset of the Norne Field
benchmark case, which includes a reservoir simulation model with
grid cell sizes of 60 × 60 m, 44927 active cells, subdivided into 22
reservoir zones, well logs, production data and 4D seismic data.
Data preparation
The set of logs available in the majority of the wells in the Norne
Field included gamma-ray (GR), bulk density (RHOB) and neutron
porosity (NPHI) logs with a sampling rate of 0.125 m. In addition,
most of the wells included density porosity, permeability and Vshale
calculations (these and other calculations are compiled in Verlo &
Hetland 2008). Porosity calculations (∅density or PHIF) in the Norne
Field are based on the density log using the following equations:
ρMatrix − ρBulk
∅ density = (1)
ρMatrix − ρFluid
with ρMatrix being the matrix density generally set to 2.67 g cm−3
(Tofte Formation with values of between 2.65 and 2.71 g cm−3),
ρBulk is the bulk density from the density log, and ρfluid is the
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
4 G. G. Correia & D. J. Schiozer
apparent fluid density generally set to 0.19 
­ g cm−3 for gas,
0.72 g cm−3 for oil and 1.02 g cm−3 for water. Slightly different val-
ues were applied depending on the reservoir zones and wells.
Equation (1) was used to calculate the porosity in some wells,
namely those that used oil-based mud. For the rest of the wells,
empirical linear correlations were used based on core data from the
exploration wells:
∅ density = a + b × ρ bulk . (2)
Cross-plots of overburden-corrected core porosity v. density log
were used to obtain the constants a and b (Verlo & Hetland 2008).
Permeability was calculated from a log-linear correlation using the
constants obtained from the relationship between overburden-cor-
rected core porosity and overburden-corrected core permeability:
( a ∅density + b )
K log = 10 . (3)
When analysing the simplified lithological column of the geologi-
cal well reports with the well log profiles, it was observed that the
facies could best be discriminated using the GR log, and the separa-
tion between the density and neutron logs. When density and neu-
tron logs are displayed in the same track on compatible scales
(porosity units), the separation between the two can be attributed to
lithology or to the presence of gas or any other light hydrocarbon.
The presence of light hydrocarbons is indicated when the neutron
porosity is less than the density porosity in a porous and permeable
zone (Krygowski 2003). The progressively higher sand/shale ratio
(particularly evident in the Not Formation interval in the Vshale log)
is clearly reflected in the increasingly negative separation between
these two logs (Fig. 4). Based on the results obtained by Gupta &
Johnson (2001) for similar reservoir sandstones in the Gullfaks
Field, this separation has been quantified and termed as a new
curve, DIFF. The DIFF curve was calculated and used as input data
in the electrofacies classification, where:
DIFF=NPHI − PHIF (4)
with NPHI being the neutron porosity log and PHIF is the esti-
mated density porosity log. The shapes of the GR and DIFF curves
in the same track show an overall funnel-shape profile characteris-
tic of coarsening-upwards trends (Fig. 4). Before starting with the
electrofacies classification, the data preparation of environmen-
tally corrected logs also involved the quality control (QC) and the
depth matching of various logs from different wells. From the 48
wells included in the Norne Field database, only 26 were used in
the modelling processes. The reasons for excluding some wells
were: the absence of important well logs (especially the neutron
and the density logs); wells that did not bring any additional infor-
mation because they were plugged and abandoned before reaching
the reservoir, sidetracks in similar positions, and cutting the same
reservoir zones; and wells with gaps or poor-quality profiles
(erratic profiles, casing shoe effect).
Electrofacies classification
The unsupervised ANN was used in the electrofacies classification
because the core geological descriptions were not available to train
the neural network. The GR vs. DIFF and the RHOB vs. NPHI
curves were used to generate the cross-plots that could best distin-
guish the different facies and were used as data input for the ANN
method (Fig. 5). The cross-plots were generated for each of the 26
wells previously selected, resulting in the electrofacies profiles
(Fig. 6). As an example, in well 6608/10-C-4H, the neural net was
trained to classify the input data into six classes. Six well-individ-
ualized clusters were obtained, representing six facies (Fig. 5 and
the well on the right-hand side of Fig. 6). In well 6608/10-C-3H,
and using the same number of classes as for the initial guess, we
observed that electrofacies 5 or the gas sands (the red coloured
cluster in the pdf version) were practically absent. This was prob-
ably due to the absence of the gas cap close to the reservoir’s SW
boundary (the well on the right-hand side of Fig. 6). On the next
attempt only five classes were used. In addition, and for quality
control of the electrofacies database and in order to correct some
artefacts arising from the ANN method, a detailed visualization of
the depth plots was made for each of the wells. The electrofacies
predictions were compared with the input logs (GR, NPHI, RHOB)
together with some calculated logs (e.g. Vshale) and with the geo-
logical descriptions (simplified lithological column and cuttings
descriptions) that were available in some wells through the scanned
composite logs (Fig. 4). In most of wells, five or six well-individ-
ualized clusters were obtained that seemed to represent the best fit
to the geological background and the sedimentary interpretations
of the mid-Norway region. Only for a few wells (e.g. well 6608/10-
D-2 HT2) were more than six classes initially used to classify the
input data, but these were then grouped into six classes of electro-
facies (0–5).
Electrofacies 0 corresponds to carbonate-cemented zones char-
acterized by high density (RHOB > 2.3) and low GR values, and
high sand/shale ratio. Most of these zones correspond to thin layers
(<1 m), with low porosity and very low permeability (mean 1 mD),
occurring in all the reservoir formations. Owing to their petrophys-
ical characteristics, this electrofacies can be considered from a pro-
duction point of view as a non-reservoir or a reservoir of low
quality. As the effective vertical permeability will be very low,
these layers can also act as vertical stratigraphic barriers to fluid
flow, also increasing the tortuosity of the flow path (Begg & King
1985). It is very important to record the location and the lateral
extent of these electrofacies when upscaling to the high-resolution
3D geomodel in order to preserve their impact on future flow-sim-
ulation phases. Electrofacies 1 corresponds to shales, with a very
low sand/shale ratio, high GR values (GR > 100), high density
(RHOB > 2.3), neutron and DIFF values, and low density porosity
(PHIF) values (Fig. 5 and Table 1). They are more frequent in the
Not Formation and in the heterolithic facies of the Tilje Formation.
The permeability is low (mean 0.2 mD) and the effective vertical
permeability is dependent on the dimensions of the shale layers.
Normally, they occur in thin horizons (up to 1 m) except for the
Not Formation. This facies can be considered as non-reservoir or
as having poor reservoir properties, and, thus, their frequency and
lateral extent is very important in the vertical movement of the
fluids. Electrofacies 2 corresponds to shaly sandstones, with a
shale volume of less than 50%, and when compared with electrofa-
cies 1 has slightly lower GR, RHOB, NPHI and DIFF values, and
higher PHIF values (Fig. 5 and Table 1). This facies occurs as thin
to medium size horizons (up to few metres) mainly in the Not and
Tilje formations. From a production standpoint, this facies has
moderate reservoir properties, with low permeability (mean
10 mD). Electrofacies 3 corresponds to fine sandstones, with a
shale volume usually below 30%. Generally, this electrofacies
shows medium values for most of the properties (GR, RHOB,
NPHI, PHIF). The permeability reaches up to 2500 mD with mean
70 mD (Fig. 5 and Table 1). This facies occurs over the entire res-
ervoir, totalling approximately 25% of the reservoirs in the profile
of analysed wells, and is less common in the Tofte Formation.
From a reservoir production point of view, this facies has a moder-
ate to good reservoir quality. Electrofacies 4 corresponds to clean
sandstones characterized by low GR, RHOB and DIFF values, and
high PHIF values. This is the most abundant facies throughout the
reservoir (44.2%: Table 1), frequently materialized by thick sand-
stone beds, namely in the Tofte and Ile formations (see Figs 4 and 6).
It shows high permeability (mean 567 mD) and porosity values,
and, therefore, a very good reservoir quality. Electrofacies 5
­corresponds to sandstones of the Fangst Group, especially from
the Not 2 reservoir zone. Together with electrofacies 4, these
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
Reservoir characterization of the Norne Field 5

Fig. 4. Electrofacies profile obtained for the Not, Ile and Tofte formations with a 0.125 m sampling rate (example from well 6608/10-D-4 H).
Electrofacies were computed using GR, DIFF (NPHI–PHIF), RHOB and NPHI logs as input data to the unsupervised ANN. The Vshale log was also use
to validate the electrofacies database.

s­ andstones have the best reservoir quality, with the highest perme-
ability and porosity values (Table 1). Electrofacies 5 is distin-
guished from electrofacies 4 on the basis of its lower density,
lower NPHI and DIFF values, and generally higher GR values
(Fig. 5). This is probably due to the coarser grain of this facies
(Dalland et al. 1988) and to the presence of light hydrocarbons
(e.g. gas) (Gupta & Johnson 2001). The density and neutron logs
are both sensitive to the presence of light hydrocarbons, namely
gas. In the Norne Field, the gas is mainly confined to the Not
Formation, as well to electrofacies 5. For most of the reservoir, the
electrofacies are a function of their lithology. An exception was
made for the top of the reservoir (Not Formation) due to the pres-
ence of the gas cap. In this reservoir zone, electrofacies 5 refers to
gas sands and electrofacies 4 refers to oil sands. Suman & Mukerji
(2013) also referred to the presence of some well log data corre-
sponding to gas sands in the Not Formation. Distinguishing elec-
trofacies 4 and 5, and modelling their distribution over the
reservoir, would be useful in the flow-simulation stages during
integration with the probabilistic and multi-objective history
matching approach using both production and 4D seismic data.
In the ANN predictions of electrofacies the uncertainty could be
significant and incorrect predictions may arise if the ANN method
is not used with caution. A suitable selection of input variables and
an adequate number of classes will increase the predictive accu-
racy of the network. A meticulous data preparation integrated with
statistical analysis is also a key factor to improve the predictions.
A consistent application of geological background and available
sedimentary descriptions are critical to the successful application
of the neural network electrofacies prediction.
Simulation grid refinement
The Norne Field database provides only a reservoir-simulation
model with a low-resolution grid (c. 60 × 60 × 8.4 m) and 44.927
active cells, compatible with the multiple flow-simulation models
that are evaluated during a history matching process. The simula-
tion model has 22 zones that correspond as much as possible to the
actual change in lithology of the reservoir layers (Table 2). The
boundaries between zones were mainly chosen at sequence bound-
aries and maximum flooding surfaces (Verlo & Hetland 2008).
However, and mainly during the electrofacies classification, we
realized that for most facies the thickness was below 1 m (Fig. 7).
In order to maintain, as much as possible, the fine-scale heteroge-
neities seen in the well logs, namely the carbonate and the shale
layers that could act as vertical flow barriers due to their low per-
meability, the reservoir simulation-grid resolution was increased
in all directions to obtain the high-resolution grid (or geomodel).
The refinement was lower in the horizontal direction (1:2 propor-
tion) and more significant in the vertical direction (1:4–1:40 pro-
portion), especially in the heterolithic Tilje Formation. Thus, a
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
6 G. G. Correia & D. J. Schiozer

Fig. 5. Electrofacies display on 2D cross-plots (example from well 6608/10-C-4 H). The carbonates and shales have the worst reservoir properties, and
[AQ: 7] the
sandstones have the best reservoir quality.

Fig. 6. Position of the 26 wells with the electrofacies profiles. At the bottom some of the wells that cut the entire reservoir are displayed with the
obtained electrofacies and the GR log.

high-resolution grid (30 × 30 × 0.5–1 m) was obtained with
2.066.642 active cells capable of preserving the main electrofacies
features after upscaling the well logs to the geomodel grid (Fig. 8),
compatible with the best resolution of the well log tools (12 inch)
and maintaining the efficiency in terms of the computational
time spent during the 3D modelling processes. As an example, by
analysing both profiles in well 6608/10-D-4 H (A and B in Fig. 8)
it is possible to observe that the majority of the information was
preserved after the upscaling procedure. This would be an essential
step when generating the 3D facies and the petrophysical models.
The simulation-grid refinement also has the advantage of preserv-
ing the reservoir structure (horizons, zoning, faults network), and
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
Reservoir characterization of the Norne Field 7

Table 1. Some statistical ranges of each electrofacies for the entire reservoir. To generate the 3D porosity and permeability models, the distributions were
determined for each of the 22 reservoir zones

Facies Neutron (NPHI) Density porosity (PHIF) (NPHI+PHIF/2) Permeability (KLOGH) Net gross Well logs

  Mean Max SD Mean Max SD Mean Max SD Mean Max SD Mean Max SD %

0 – Carbonate 0.22 0.46 0.06 0.11 0.22 0.05 0.16 0.28 0.04 1.3 45.4 4.2 0.1 0.2 0.05 8.0
1 – Shale 0.32 0.68 0.07 0.10 0.20 0.04 0.21 0.43 0.05 0.2 8.4 0.9 0.1 0.2 0.05 4.4
2 – Shaly sand 0.28 0.50 0.05 0.16 0.30 0.05 0.22 0.33 0.04 9.9 983.5 63.7 0.5 1 0.20 11.1
3 – Fine sand 0.26 0.47 0.04 0.22 0.33 0.04 0.24 0.38 0.03 70.6 2510.5 157.9 1.0 1 – 24.6
4 – Sandstone 0.25 0.36 0.03 0.27 0.38 0.03 0.26 0.37 0.02 567.0 9053.4 809.2 1.0 1 – 44.2
5 – Not 2 sands 0.18 0.44 0.05 0.28 0.41 0.04 0.23 0.42 0.03 494.0 4352.2 615.0 1.0 1 – 7.7

Table 2. Variograms detail estimated by Suman (2013) used in this study to model each of the 22 reservoir zones individually. Also included is the number
of the cumulative layers of the high-resolution model

Variogram values (Suman 2013)

Geomodel Ranges Azimuth parallel 

Zones Layers Parallel Normal Vertical

Not 2.3 1 14 1231 912 0 45

Not 2.2 2 28 1584 549 0 22
Not 2.1 3 42 1081 500 0 42
Not 1 4 43  
Ile 2.2 5 49 2506 629 0 88
Ile 2.1.3 6 55 2574 500 0 45
Ile 2.1.2 7 61 839 500 0 0
Ile 2.1.1 8 67 1905 668 0 44
Ile 1.3 9 73 800 674 0 135
Ile 1.2 10 87 1527 500 0 45
Ile 1.1 11 95 830 500 0 125
Tofte 2.2 12 99 2486 971 0 45
Tofte 2.1.3 13 109 975 500 0 135
Tofte 2.1.2 14 119 2041 775 0 45
Tofte 2.1.1 15 129 1256 500 0 45
Tofte 1.2.2 16 139 1644 500 0 45
Tofte 1.2.1 17 143 1809 423 0 133
Tofte 1.1 18 147 1004 500 0 0
Tilje 4 19 163 2136 500 0 133
Tilje 3 20 183 1937 755 0 45
Tilje 2 21 223 2175 500 0 48
Tilje 1 22 253 1571 583 0 26

so allowing future comparisons with other studies that use the
database from the Norne Field benchmark case (i.e. studies related
to history matching procedures).
The neutron, density porosity, permeability and the electrofacies
logs were upscaled to the high-resolution grid using common aver-
aging methods available in commercial modelling softwares. For the
electrofacies, the most represented facies was selected and assigned
for each cell. For porosity and permeability, the arithmetic and har-
monic averages were used, respectively, biasing these continuous
log properties with the upscaled electrofacies. The quality control
undertaken between the upscaled cells and the well log values
revealed that the main trends were maintained for all the properties,
apart from some smoothing effects that had no significant impact on
the behaviour of the reservoir (Figs 8 and 9). The statistic values
taken from Table 1 and from the cross-plots and histograms (Figs 5
and 9) help in determining the uncertainty ranges that will drive the
petrophysical modelling constrained by the facies distribution.
Facies and petrophysical modelling
One of the objectives of this study was to show the advantages of
introducing a detailed electrofacies classification that will be used
in the generation of high-resolution 3D facies models which, in
turn, will be used to constrain the distribution of petrophysical
properties. In this way, this study ends with the application of the
generated electrofacies scheme as input data for the construction
of 3D facies and petrophysical models. Note that this is a first
approach. Future adjustments in the facies and petrophysical prop-
erties distribution could then be introduced during the history
matching procedure.
Prior to the facies modelling, 3D discrete trend properties were
created which means the generation of vertical proportion volumes
created by kriging interpolation and block averaging of the proba-
bilities of each of the electrofacies. The trend modelling algorithm
used the estimated probability curves obtained from the data anal-
ysis (Fig. 10) and the variograms estimated by Suman (2013) for
each of the 22 zones of the Norne Field reservoir simulation model
(Table 2). Figure 11 shows that the coarser sands (electrofacies 5)
are more abundant in the Not Formation, occurring also in the top
Ile Formation. The sandstones (electrofacies 3 and 4) occupy most
of the Ile and Tofte formations. The shales and shaly sandstones
(electrofacies 1 and 2) occur mainly at the base of the Not
Formation and in the Tilje Formation. The fine-scale carbonates/
cemented layers occur along the entire reservoir, being most sig-
nificant in the Tofte Formation. The 3D probability volumes
obtained were then used in the facies-modelling algorithm as a
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
8 G. G. Correia & D. J. Schiozer

Fig. 8. Electrofacies scheme obtained before upscaling to the high-

Fig. 7. Electrofacies thickness histogram showing that the majority of
the facies are less than 1 m. This highlights the importance of better
representing these fine-scale heterogeneities when generating the high-
resolution 3D models.
resolution grid (A) and after the upscaling (B) (example from well
6608/10-D-4 H). The thin intercalations that characterize the heterolithic
facies of the Tilje Formation highlight the need to generate a high-
resolution grid for the Norne Field. Comparing both profiles (A and
B), it is possible to observe that the majority of the information was
preserved after the upscaling procedure.

Fig. 9. Display of the distribution of properties using histograms. The porosity values were taken from the density porosity log and the permeability
values from the calculated permeability log. The main trends are preserved after the upscaling procedures, apart some smoothing effects.

s­ econdary control. For the facies modelling, a cell-based technique
(sequential indicator simulation (SIS)) was applied. It is a widely
used method applicable to heterogeneous facies that have no clear
geometrical shapes, and prior to porosity and permeability model-
ling (Deutsch 2002; Cabello et al. 2011). The required variograms
were the ones estimated by Suman (2013) (Table 2). The facies
model obtained clearly shows the influence of the 3D probability
volumes on the facies distribution (Figs 11 and 12). An example
can be seen in the top of the reservoir with the Not 2 sandstone
distribution when compared with the 3D probability volume for
the same facies (Fig. 12a, b). When these global proportions are
not included, a more randomly distributed facies volume is
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
Reservoir characterization of the Norne Field
Fig. 10. Electrofacies probability curves used to build the 3D
probability volumes.
obtained. In the detailed cross-section, the fine and continuous car-
bonate/cemented layers between the Ile and Tofte formations and
in the Tilje Formation are clearly visible . The frequent shale hori-
zons are also seen in the Tilje Formation, both of these act as pos-
sible barriers to vertical flow (Fig. 12c). Mapping these layers into
high-resolution facies models could be an important contribution
to better represent the production behaviour. After the required
upscaling procedures to a low-resolution grid prior the reservoir
flow simulation, most of these details will be lost. However, some
actions could preserve some of these features, such as the manipu-
lation of the transmissibilities in specific surfaces.
The facies must have a significant control on the petrophysical
properties distribution such as the porosity and permeability. In
fact, the previously created facies models are essential in ensuring
the geologically and statistically reliable distribution of the petro-
physical properties (Deutsch 2002; Zakrevsky 2011). In this study,
the Gaussian random function simulation algorithm was used. This
algorithm is usually faster than the widely used sequential Gaussian
simulation algorithm (SGS), and enables fast modification of the
correlation between primary and secondary variables through the
collocated co-simulation function. The porosity can be derived
individually from a sonic, neutron or density log, or by combining
two or three of the logs using average or cross-plot methods
(Moore et al. 2011). In this study, after testing the density and neu-
tron porosity logs individually to model the porosity distribution,
the most suitable pore volumes, when compared to the reservoir
9
pore volume available in the Norne Field benchmark case, were
obtained when using the average of the neutron and density poros-
ities (Table 1). The 3D permeability models were generated using
the geostatistical collocated co-simulation in relation to the 3D
porosity model through a range of correlation factors (mean 75%,
minimum 60% and maximum 90%) (Fig. 13a, b). As in the facies
modelling, each of the 22 reservoir zones was individually mod-
elled using the variograms previously estimated by Suman (2013)
(Table 2) and using specific distributions (truncated normal distri-
bution for porosity and truncated log-normal distribution for per-
meability) that were derived from the histograms and statistical
analysis of each facies in each reservoir zone. A NtG 3D volume
was also created in the course of the generation of the facies vol-
ume, giving an estimation for the reservoir volume and reservoir
connectivity that will be of great importance in hydrodynamic
simulation (Zakrevsky 2011). As a first guess and based on the
geological properties and production behaviour, specific ranges of
uncertainty were defined for each electrofacies instead of using
permeability cut-offs and Vshale curves. To the sandstones (facies 3,
4 and 5) was assigned a NtG = 1, to the shaly sandstones (facies 2)
an intermediate NtG (mean 0.5), and for the carbonate and shale
horizons it was assumed that the grid cells were practically discon-
nected (NTG < 0.2). When analysing Figure 13, is easily noticeable
that the lower Not Formation and the Tilje Formation are the sec-
tions with the worst reservoir properties owing to their low poros-
ity and permeability, which is mainly associated with the carbonate/
cemented and shale layers (Fig. 13a, b). In the NtG 3D volume, it
is clearly observed that these same sections along with some hori-
zons in the Ile and Tofte formations will represent the main chal-
lenges concerning the reservoir connectivity (Fig. 13c). To
highlight the importance and the choice for the construction of
high-resolution 3D models, the porosity, permeability and NtG
volumes available in the database of the Norne Field benchmark
case are shown (Fig. 13d–f, respectively). As can be observed and
in addition to the absence of 3D facies volumes, these models are
characterized by their low resolution and the high degree of
smoothing of their properties which make it much more difficult to
establish a cause–effect relationship between the geological prop-
erties and the fluid flow behaviour, and, thereby, maintain the
geological consistency during history matching procedures. In
­
addition, and in order to validate the new datasets, 200 facies and
petrophysical models were generated using the petrophysical
ranges of each electrofacies (in each of the 22 reservoir zones), the
above NtG values and the variograms (Table 2). The pore volumes
of each realization are in agreement with the deterministic pore
volume given in the Norne Field benchmark case (673 × 106 m3).
For the 200 realizations, the 10th, 50th and 90th percentiles (P10,
P50 and P90) were 661 × 106, 674 × 106 and 688 × 106 m3, respec-
tively (Fig. 14).
All these considerations and procedures concerning the 3D
modelling stage comprise only a first analysis and approach to
validate and highlight the importance of generating a high-resolu-
tion electrofacies scheme for the Norne Field. This stage will be
integrated in a wider loop of fluid-flow simulation and probabilis-
tic history matching using production and 4D seismic data, and
being subject to future revisions and improvements in order to
always ensure geological consistency.
Conclusions
An electrofacies approach based on unsupervised artificial neural
networks (ANN) analysis has been successfully applied to the dis-
crimination of different facies using conventional logs (gamma-
ray, neutron, density logs) of 26 different wells in the Norne Field
siliciclastic reservoir. These facies where deposited in a tidal,
shallow-marine and prograding deltaic depositional systems in a
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
10 G. G. Correia & D. J. Schiozer

Fig. 11. Three-dimensional probability volumes for each facies. The coarser sands (electrofacies 5) are more abundant in the Not Formation, also
occurring in the top Ile Formation. The sandstones (electrofacies 3 and 4) are more frequent in the Ile and Tofte formations. The shales and shaly
sandstones (electrofacies 1 and 2) occur mainly at the base of the Not Formation and in the Tilje Formation. The thin carbonates/cemented layers occur
along the entire reservoir, being most significant between the Ile and Tofte formations.

Fig. 12. (a) Three-dimensional facies model obtained using the SIS algorithm. The detached region (black square) shows the location, thickness, extent
and frequency of decimetre shale/cemented layers that could act as vertical barriers to fluid displacement. (b) The 3D probability volume was used
as a secondary parameter in the modelling process. The black circle highlights a particular region where this influence is most evident. (c) Particular
attention to the carbonate layer between the Ile and Tofte formations, and the intercalations in the Tilje Formation.
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
Reservoir characterization of the Norne Field 11

Fig. 13. Three-dimensional petrophysical models constrained by the facies distribution. Histograms and statistical ranges obtained for each facies and
each reservoir zone were used to model (a) the porosity, (b) permeability and (c) NtG. For permeability, the collocated co-simulation function was
used to condition the permeability distribution to the porosity distribution using a range of correlation factor (mean 75%). For comparison, the (d) 3D
porosity, (e) permeability and (f) NtG volumes are shown: these are available in the Norne Field database.

predominant coarsening-upwards sequence, comprising massive

Fig. 14. Pore volumes of 200 realizations (P10 = 661 × 106 m3;
P50 = 674 × 106 m3; P90 = 688 × 106 m3). The vertical line in the centre
represents the pore volume of the deterministic model included in the
Norne Field benchmark case (673 × 106 m3).
sandstone to thinly interbedded shale and carbonate cemented lay-
ers. Different electrofacies were recognized by grouping distinct
well log clusters on multiple log cross-plots (GR v. DIFF and
RHOB v. NPHI). The final electrofacies classification comprising
a total of six classes has been resolved down to a sampling rate of
0.125 m thick in 26 wells, allowing better interpretation and com-
prehension of the geological framework of the reservoir, and add-
ing new information to the Norne Field benchmark case database.
The electrofacies classification based on ANN becomes more
useful when integrated with statistical modelling methods, and with
geological and reservoir knowledge. The electrofacies were
upscaled to a high-resolution grid (cells 0.5–1 m thick) that forms
the basis for building high-resolution 3D reservoir facies and petro-
physical models for the Norne Field. The high-resolution grid was
obtained through refinement of the reservoir simulation grid avail-
able in the Norne Field database, keeping the same reservoir geom-
etry. In this way, and when integrating this study into a wider history
matching workflow, it will be possible to compare the results with
previous and future studies coming from the Norne Field database.
The final stage of this work, which comprises the generation of
high-resolution 3D facies and petrophysical models (porosity, per-
meability and NtG), highlighted the importance of generating an
electrofacies scheme. The histograms of porosity (permeability)
within the different facies add a significant control to the distribution
of these properties, reducing uncertainties and resulting in models
 Downloaded from http://pg.lyellcollection.org/ at University College London on March 4, 2016
12 G. G. Correia & D. J. Schiozer
with more predictive power. The definition of the reservoir and the
connectivity estimation through the NtG becomes directly related to
the electrofacies by defining specific ranges of uncertainty to each
electrofacies instead of using the frequently problematic permeabil-
ity cut-offs and Vshale curves. The high-resolution 3D models helped
to identify and characterize small variations in the reservoir quality,
and to identify the location, thickness, extent and frequency of deci-
metre-scale shale and cemented layers that could act as vertical bar-
riers to fluid flow, which has a significant impact on the estimation
of the effective vertical permeability and reservoir behaviour.
The high-resolution datasets that resulted from this study will
form the working basis in the integration with probabilistic and
multi-objective history matching approaches using both produc-
tion and 4D seismic data, and is assisted by geostatistical parame-
terization techniques. Future work will show the outcome of this
integration and, ultimately, closed-loop reservoir management of
the Norne Field.
Acknowledgements and Funding
The authors would like to thank the IO Center for providing the data, and
Statoil (operator of the Norne Field) and its licence partners ENI and Petoro
for the release of the Norne data. The authors also wish to thank the BG Group,
UNISIM, CEPETRO, DEP-FEM-UNICAMP for their support in this work. We
also thank Schlumberger and Emerson for the software licences.
References
Begg, S.H. & King, P.R. 1985. Modelling the Effects of Shales on Reservoir
Performance: Calculation of Effective Vertical Permeability. Paper SPE
13529 presented at the SPE Symposium on Reservoir Simulation, Dallas,
TX, 10–13 February 1985.
Cabello, P., Falivene, O., López-Blanco, M., Howell, J.A., Arbués, P. & Ramos,
E. 2011. An outcrop-based comparison of facies modelling strategies in fan-
delta reservoir analogues from the Eocene Sant Llorenç del Munt fan-delta
(NE Spain). Petroleum Geoscience, 17, 65–90, http://doi.org/10.1144/1354-
079309-021
Chang, H.C., Kopaska-Merkel, D.C. & Chen, H.C. 2002. Identification of litho-
facies using Kohonen self-organizing maps. Computers & Geosciences, 28,
223–229.
Dalland, A., Worsley, D. & Ofstad, K. 1988. A lithostratigraphic scheme for
the Mesozoic and Cenozoic succession offshore mid- and northern Norway.
Norwegian Petroleum Directorate Bulletin, 4, 1–65.
Deutsch, C.V. 2002. Geostatistical Reservoir Modelling: Applied Geostatistics
Series. Oxford University Press, New York.
Gupta, R. & Johnson, H.D. 2001. Characterization of heterolithic deposits using
electrofacies analysis in the tide-dominated Lower Jurassic Cook Formation
(Gullfaks Field, offshore Norway). Petroleum Geoscience, 7, 321–330,
http://doi.org/10.1144/petgeo.7.3.321
Hammer, E., Mørk, M.B.E. & Næss, A. 2010. Facies controls on the distribu-
tion of diagenesis and compaction in fluvial-deltaic deposits. Marine and
Petroleum Geology, 27, 1737–1751.
Harris, N.B. 1989. Reservoir geology of Fangst Group (Middle Jurassic),
Heidrun Field, offshore Mid-Norway. American Association of Petroleum
Geologists Bulletin, 73, 1415–1435.
Henriquez, A. & Jourdan, C. 1995. Challenges in modelling reservoirs in the
North Sea and on the Norwegian shelf. Petroleum Geoscience, 1, 327–336.
Krygowski, D.A. 2003. Guide to Petrophysical Interpretation. American
Association of Petroleum Geologists, Austin, TX.
Ma, Y.Z. & Gomez, E. 2015. Uses and abuses in applying neural networks
for predictions in hydrocarbon resource evaluation. Journal of Petroleum
Science and Engineering, 133, 66–75.
Martinius, A.W., Ringrose, P.S., Brostrøm, C., Elfenbein, C., Næss, A. &
Ringås, J.E. 2005. Reservoir challenges of heterolithic tidal sandstone res-
ervoirs in the Halten Tarrce, mid-Norway. Petroleum Geoscience, 11, 3–16,
http://doi.org/10.1144/1354-079304-629
Maschio, C. & Schiozer, D.J. 2013. A new procedure to reduce uncertainties
in reservoir models using statistical inference and observed data. Journal of
Petroleum Science and Engineering, 110, 7–21.
Maschio, C. & Schiozer, D.J. 2014. Bayesian history matching using artifi-
cial neural network and Markov Chain Monte Carlo. Journal of Petroleum
Science and Engineering, 123, 62–71.
Mohaghegh, S., Arefi, R., Ameri, S., Aminiand, K. & Nutter, R. 1996. Petroleum
reservoir characterization with the aid of artificial neural networks. Journal
of Petroleum Science and Engineering, 16, 263–274.
Moore, W.R., Ma, Y.Z., Urdea, J. & Bratton, T. 2011. Uncertainty analysis
in well log and petrophysical interpretations. In: Ma, Y.Z. & La Pointe, P.
(eds) Uncertainty Analysis and Reservoir Characterization and Modeling.
American Association of Petroleum Geologists, Memoirs, 96, 17–28.
Morton, A., Hallsworth, C., Strogen, D., Whitham, A. & Fanning, M. 2009.
Evolution of provenance in the NE Atlantic rift: The Early–Middle Jurassic
succession in the Heidrun Field, Halten Terrace, offshore Mid-Norway.
Marine and Petroleum Geology, 26, 1100–1117.
Nordahl, K., Ringrose, P.S. & Wen, R. 2005. Petrophysical characterization of
a heterolithic tidal reservoir interval using a process-based modelling tool.
Petroleum Geoscience, 11, 17–28, http://doi.org/10.1144/1354-079303-613
Qi, L. & Carr, T.R. 2006. Neural network prediction of carbonate lithofacies
from well logs, Big Bow and Sand Arroyo Creek fields, Southwest Kansas.
Computers & Geosciences, 32, 947–964.
Ringrose, P., Nordahl, K. & Wen, R. 2005. Vertical permeability estimation
in heterolithic tidal deltaic sandstones. Petroleum Geoscience, 11, 29–36,
http://doi.org/10.1144/1354-079303-614
Rwechungura, R., Suwartadi, E., Dadashpour, M., Kleppe, J. & Foss, B. 2010.
The Norne Field Case – A Unique Comparative Case Study. Paper SPE
127538, presented at SPE Intelligent Energy Conference and Exhibition,
Utrecht, Netherlands, 23–25 March 2010.
Storvoll, V. & Bjørlykke, K. 2004. Sonic velocity and grain contact proper-
ties in reservoir sandstones. Petroleum Geoscience, 10, 215–226, http://doi.
org/10.1144/1354-079303-598
Suman, A. 2013. Joint Inversion of production and time-lapse seismic data:
Application to Norne Field. PhD thesis, Stanford University, USA.
Suman, A. & Mukerji, T. 2013. Sensitivity study of rock-physics parameters
for modeling time-lapse seismic response of Norne field. Geophysics, 78,
D511–D523.
Swiecicki, T., Gibbs, P., Farrow, G. & Coward, M.P. 1998. A tectonostrati-
graphic framework for the Mid-Norway region. Marine and Petroleum
Geology, 15, 245–276.
Verlo, S.B. & Hetland, M. 2008. Development of a field case with real pro-
duction and 4D data from the Norne Field as a benchmark case for future
reservoir simulation model testing. Master’s thesis, Norwegian University of
Science and Technology, Trondheim.
Worthington, P.F. & Cosentino, L. 2005. The role of cutoffs in integrated res-
ervoir studies. Paper SPE 84387. SPE Reservoir Evaluation & Engineering,
8, 276–290.
Zakrevsky, K.E. 2011. Geological 3D Modelling. European Association of
Geoscientists and Engineers (EAGE), Houten, The Netherlands.