Vous êtes sur la page 1sur 9

A 3D stochastic model integrating depth, fault and property uncertainty

for planning robust wells, Njord Field, oshore Norway


Jan C. Rivens, Cecilie Otterlei, Eli Zachariassen, Chris Dart and Jorunn Sjholm
Norsk Hydro ASA, PO Box 7190, 5020 Bergen, Norway
ABSTRACT: The Njord Field, situated in the Haltenbanken area, is one of the most
challenging reservoirs on the Norwegian shelf. The structural complexity is high,
with numerous intersecting faults compartmentalizing the reservoir. The reservoir is
dominated by alternating shale and sand intervals of tidal, estuarine and deltaic
depositional origin.
One of the main segments on Njord the Central Area was the target for a
one-year improved oil recovery study. The Central Area reservoir is produced by
depletion, with a current recovery factor of only 6%, as most faults appear to seal
during production. The previous drilling experience revealed quite large uncertainty
with respect to both depth and missing faults due to poor seismic data quality.
In order to plan further wells in this area a model that comprises the full
uncertainty with respect to both structural components (depth uncertainty and
faults) and property components (facies, permeability, porosity) was made. The
model was built by combining commercial modelling software (IRAP RMS and
STORM:HORIZON) with an R&D tool (HAVANA) in a manner that has not been
attempted previously, and 200 stochastic realizations were run. Each realization has
a different structural 3D grid (depth and fault pattern) and different petrophysical
properties. A streamline simulator in IRAP RMS was applied to rank the realizations
and ten realizations were chosen for further well screening in the flow simulator
(ECLIPSE). The model was applied to propose two new well targets that were
robust economically (given all uncertainty). In addition, the model was applied in
testing various unconventional well types, for example so-called connector-wells,
which are open holes in the reservoir (but not connected to the surface), in order to
connect fault blocks bounded by sealing faults.
Drilling and subsequent seismic interpretation after the model had been built
revealed that the structural uncertainty was even greater than predicted. It is crucial
to capture the full interpretation uncertainty and, in particular, to address the
problem of jump-correlation across fault blocks. Despite this, however, the observed
cumulative production from the new oil producer compares well with the prognosis
from the model.
KEYWORDS: Njord, structural uncertainty, stochastic faulting, well screening, reservoir modelling

INTRODUCTION
The Njord Field (Lilleng & Gundes 1997) is c. 6 km in
diameter and located c. 130 km NW of the operations base in
Kristiansund (Fig. 1a). It is operated by Norsk Hydro, with
ExxonMobil, Petoro, Paladin, ConocoPhillips, OER and Gaz
de France as partners. The reservoir comprises a c. 80120 m
thick Lower Jurassic Tilje Formation interbedded sand and
shale sequence that was deposited in a tidal/estuarine setting.
The field has been involved in rifting phases at various times
throughout the Triassic, Jurassic and early Cretaceous (Blystad
et al. 1995). It is located in the hangingwall of the NW-dipping
master fault that defines the southeastern margin of the Halten
Terrace. This fault has a complex geometry adjacent to the field
and has been interpreted variously as either a rampflatramp
(Osmundsen et al. 2002) or, alternatively, as part of a breached
Petroleum Geoscience, Vol. 11 2005, pp. 5765

relay system. This geometrical complexity has resulted in the


complicated and dense pattern of NESW and NS-trending
faults observed at Njord (Fig. 1b).
The marine-deltaic Pliensbachian Tilje Formation forms
the main reservoir unit. This formation unconformably
overlies the fluvial Triassic/Sinemurian Aare Formation, and is
succeeded by deep-water marine deposits of the Toarcian
Ror Formation. The reservoir is divided into four main
sections: Tilje 14 (Fig. 2). Lithologically, the Tilje 1 and 2 units
are characterized by immature mouthbar sands displaying
varying grain size and sorting characteristics, alternating with
silty to shaly lower energy deposits. Tilje 3 is characterized
by a tide-dominated facies with high degree of heterogeneity
and generally better sorting of the sandstones (Dreyer et al.
1999).
1354-0793/05/$15.00  2005 EAGE/Geological Society of London

58

J. C. Rivens et al.

Fig. 1. (a) Njord is located in the


southern part of the Halten Terrace,
offshore Mid-Norway. (b) Overview of
Njord Field, showing major faults and
segment names (CA, Central Area; EF,
East Flank; NA, Northern Area). The
three wells producing from CA prior to
IOR study are shown.

The marine Tilje 1A forms the basal part of the Tilje


Formation. It is a 2535 m thick unit in the Central Area (Fig.
1), dominated mainly by aggradational shoreface and delta-front
deposits. It is separated from the fluvial Aare Formation by a
regressive unconformity and transgressive erosion surface. Tilje
1A is further subdivided into a lower and upper unit. Calcitecemented zones are often seen in the lower part of Tilje 1A.
Tilje 1B is a thin zone dominated by delta-plain fines and coals
that are continuous throughout Njord. It is bounded below by
an erosional sequence boundary and above by a lithological
boundary grading into Tilje 1C, which is a thin zone consisting
mainly of fluvial channel sandstones. Tilje 2A consists of
fine-grained distal marine, heavily burrowed deposits. The Tilje
2A maximum flooding surface is correlatable regionally and a
possible field-wide flow barrier. Tilje 2B is sandstone rich and
most probably deposited as the lower to middle part of an
estuarine infill sequence. The large thickness difference of this
zone (840 m TVT) throughout Njord and its basal incision
implies that it may have formed in response to a tectonic event.
The base of the Tilje 2B sequence corresponds to a field-wide
erosional sequence boundary separating the underlying distal

marine shale of Tilje 2A from the marginal-marine (estuarine)


deposits of Tilje 2B and 3A. In the Central Area the Tilje 2
sequence is 2045 m TVT thick.
Tilje 3A is a thinly bedded and very heterolithic zone
containing a mosaic of laterally discontinuous tidal facies with
few discernible vertical or lateral trends. It becomes increasingly
marine-influenced stratigraphically upwards and towards the
north and has a thickness of 1545 m TVT in the Central Area.
Tilje 4A is a thin field-wide zone containing heterolithic
fine-grained deposits formed in a relatively distal marine environment. Being thoroughly bioturbated, the open-marine
deposits act as a flow barrier. The uppermost Tilje 4B is poorly
understood in terms of depositional environments, partly because of lack of core coverage. It is 310 m thick and consists
of shallow-marine bioturbated sand- and siltstone with some
low-angle cross-bedding.
Production on Njord began in 1997 and, by 2001, 14
production and injection wells had been drilled. Some faults
with throws of less than 25 m are sealing on a production
time-scale as a result of the high percentage of shale/clay
content within the reservoir interval that has been dragged into

Fig. 2. Stratigraphy of the Tilje


Formation. Based on 29 wells of which
nine are cored (simplified from Dreyer
et al. 1999).

Stochastic model for well planning, Njord Field


the fault zones. This behaviour is especially observed in the
Central Area of the field. In order to combat compartmentalization by sealing faults, previous production wells were drilled
with U- and W-shaped geometries, designed to penetrate as
many isolated fault compartments and zones as possible and to
increase the chance of hitting the reservoir in locations where
the reservoir has been removed by displacement along faults.
At the end of 2000, relatively poor well performance
necessitated a pause in drilling and the establishment of a
one-year increased oil recovery (IOR) project to define new
well targets better. The goal of the IOR work was to provide a
decision basis for whether and when a second drilling phase
should be initiated. To accomplish that, new modelling
methods and new well design/strategies were needed.
The Central Area was of special focus for the IOR project
(Fig. 1b). This area is the most complex on Njord, being heavily
faulted, and existing wells A-10AH, A-11 and A-13H experienced numerous surprises. In particular, the A-10AH well
missed long sections of reservoir interval and is one of the
poorest producers on Njord.
The Central Area is produced by pressure depletion. The
highly compartmentalized nature of the reservoir makes it
virtually impossible to use pressure support by water or gas
injection, since the risk of placing an injector on the wrong
side of a sealing fault is large. Consequently, in the reserve
estimates prior to IOR work, the simulated recovery factor was
only 6%.
The purpose of this paper is to summarize the IOR work for
Njord Central Area. By applying unconventional technology for
structural modelling, along with advanced 3D modelling of tidal
deposits and their rock properties, the basis for the drilling of
well A10BH in 2003 was established.
MODELLING
Digital modelling is an important task in modern field
development. The complex nature of the reservoir, together
with the various scales, coverage and quality of input sources
(seismic, wells, analogue studies), makes the geological model
(geomodel) a common integration point.
As a model is always a simplification of true nature, some
conceptual choices have to be made prior to modelling,
depending on the intended purpose of the model. For example,
studies of fluvial reservoirs show that it is more important to
capture the large-scale heterogeneity (channel sand versus
overbank deposits) than small-scale variations of porosity
within a channel (Clemetsen et al. 1990).
Experience on Njord, however, shows that structural complexity and poor seismic data quality are more significant
factors to consider when screening for new targets than the
depositional complexity of tidal sediments. For this reason it
was decided early on not to follow the conventional approach
of most oil company asset teams of building a deterministic
(rigid) structural model, then populating it with stochastical
petrophysical parameters. Instead, it was deemed crucial to
capture the structural uncertainty, in addition to the petrophysical component, in a model that was suitable for investigating
the robustness of the new well targets. An alternative modelling
approach was, therefore, adopted.
Facies and petrophysical modelling
When building a 3D geological model, modelling facies and
facies architectures can be used as a tool for understanding the
distribution of petrophysical properties in the reservoir. Facies
on Njord were defined originally based on cores using a
traditional sedimentological approach.

59

If facies are to be used in the modelling process, a clear


connection between facies and petrophysical properties must
be established. If properties overlap too much, facies modelling
will serve no purpose in understanding property distribution. A
robust and consistent set of facies definitions is, therefore, of
primary importance when building a 3D model.
Consequently, a full QC and revision of all facies logs were
carried out, focusing on the need to identify all facies unambiguously on well logs by using the cored wells and their log
responses as templates for identification. As part of this work
palaeogeographical maps for different intervals within the Tilje
reservoir were generated and used as guides when modelling
the different facies in IRAP RMS (e.g. Fig. 3). Facies used in the
model are listed in Table 1. In the modelling each zone was
treated individually and poor quality zones, such as 2A and 4A,
were treated as non-reservoir. Facies modelling techniques
varied according to zone, with zones 4B, 1BC, 1AU modelled
using Facies Belts, and zones 3A, 2B, 1AL using the
Composite module in IRAP RMS.
For example, Tilje 3A was modelled with five different facies
(Fig. 3): tidal channel (TC), tidal delta (TD), marine estuary
mudstone (MEM), calcite (C) and tidal estuary heterolithics
(TEH).
The facies model provided the spatial classification of
porosity and permeability, which were modelled using
variogram-based Gaussian simulation techniques. As the estimation of vertical permeability is particularly difficult in tidal
deposits, a methodology using SBED software was applied
(Nordahl et al. 2005).
Structural modelling
Although most geoscientists acknowledge that structural uncertainty is an important factor to consider when building a
reservoir model, representing it digitally is difficult because
existing commercial software does not handle this task well.
To overcome this problem on Njord, a research and development tool (HAVANA), combined with the commercial
STORM:HORIZON was applied. IRAP RMS (Shanor &
Whelan 2004) was used to make the base case framework, to
model the facies and petrophysics and as a general project
manager and visualization tool. In addition, several in-house
programs were made in order to link the various applications
and data formats.
HAVANA is a tool that extends fault modelling into a
mathematical model which makes it possible to model a broad
range of fault scenarios. Features include: Elliptic and parametric (PFM) faults; modelling of fault planes and concurrent
reservoir deformation; advanced stochastic methods; fault truncation and interaction; use of fault density maps; conditioning
well observations to faults; and fault seal properties
(Hollund et al. 2002). It is developed and marketed by the
Norwegian Computing Centre, which developed the code in
co-operation with Statoil, Norsk Hydro and formerly Saga.
In the Njord study, HAVANA was used for modelling two
types of faults:
+ near-resolution faults (NRF) faults on the limit of seismic
resolution, which are marked as poly-lines from the seismic
interpretation. These faults were simplified to HAVANA
elliptical faults;
+ stochastic sub-seismic faults (SSF) faults with throw below
seismic resolution, known to be present from analogue
studies, general fault theory (Yielding et al. 1996) and field
production history, core data and borehole image data.
HAVANA can output faults in several ways. First, it is
possible to modify the 3D grid by allowing HAVANA to add

60

J. C. Rivens et al.

Fig. 3. Example of palaeogeographic


map applied as guide in modelling,
taken from the lower part of zone Tilje
3A.

fault splits along the grid pillars. This will change the grid
geometry. Another approach is to use transmissibility multipliers only, without offset or deformation. It was decided to
implement HAVANA faults using this second approach, allowing faults with 10 m throw or more to form a complete barrier.
The reason for making such a simplification was to restrict
computing time for flow simulation and to avoid wellconditioning problems. Fault splits increase the number of
non-neighbour connections in the flow simulation grid which
in turn will increase computing time.
HORIZON is a tool for advanced surface modelling. It is
developed by Roxar Software Solutions and the Norwegian
Computing Centre, and is a part of the commercial STORM
Table 1. Facies table
Facies code
CSH
PMS
DMS
D
MEM
TEH
TC
TDS
BDS
DPF
DPS
BHM

Description
Coarse-grained shoreline
Proximal bar sands
Distal bar sands
Distal marine
Marginal to estuarine mudstone
Tidal estuarine heterolithics
Tidal channel
Tidal delta sands
Bayhead delta sands
Delta plain fines
Delta plain sands
Bayhead massflow sands

software. HORIZON can be used in various ways including


advanced depth conversion, fitting surfaces to well points using
geostatistical kriging methods and taking all surfaces into
account (preserving isochore information), Bayesian kriging,
Gaussian simulation, etc. (Abrahamsen et al. 1991). In this
study, HORIZON has been applied to generate the uncertainty
in structural position of the top and bottom horizon.
Model set-up
Structural and property modelling were combined in a single
workflow loop to produce model realizations. The procedure
involved a number of steps.
1. Establish a base case 3D grid model. This was modelled
conventionally using IRAP RMS. This model included the
large dipping faults modelled deterministically (Fig. 4a).
2. Use top and bottom reservoir as expected (trend) surfaces
in HORIZON and generate numerous realizations. The
values of stochastic input parameters (standard deviations,
variogram range, etc.) were chosen to reflect the assumed
uncertainty and reasonable visual output. The Gaussian
field was conditioned to current well paths in order to avoid
depth changes there.
3. Subtraction of the expected from the simulated surface to
produce a smooth difference surface. The difference surface
was then added to the base case grid using in-house
software, to produce a 3D representation of the stochastic
realization (Figs. 4b, c).
4. Each grid realization (from HORIZON) was given a unique

Stochastic model for well planning, Njord Field

61

Fig. 4. Procedure for making different realizations of the structural


model: (a) generate a 3D deterministic model incorporating all major
faults; (b) make surface realizations using the Horizon software; (c)
add the difference map and adjust the grid (using in-house software);
and (d) incorporate Havana faults as zero transmissibility multipliers
(tight faults).

fault realization (from HAVANA) (Fig. 4d) and a unique


facies/petrophysical realization (from RMS).
Examples of output from the procedure are shown in
Figure 5.
History matching
As most models are not conditioned a priori to dynamic
information (wells tests and production data), the model needs
to be adjusted (matched) in order to reproduce the dynamic
history of the reservoir. The motivation for history matching is
(1) to achieve a model which is better suited for production
forecast and (2) to make better predictions for new well targets.
In this work the second point was the main focus.
Traditionally, history matching is done by taking the upscaled
geological model and modifying the values of porosity, permeability, fault properties and fluid/rock properties. In many
cases, the changes made in the simulation model are not
transferred back into the geological model, giving rise to subtle
differences between the geomodel and the simulation model.
The new approach is somewhat different. As numerous
realizations had been generated and each behaved quite differently in the simulator, it was virtually impossible to perform
traditional history matching on all realizations. Instead, changes

Fig. 5. Output from structural modelling: (a) deterministic depth


grid made in Irap RMS; (b) fault pattern (dark) for two Havana
realizations; (c) Horizon surface realizations. Orange represents the
expected (deterministic), while blue represents areas where the
realization is above.

were made directly in the geological model and a set of


realizations that fitted the history adequately was chosen.
+ For each model set-up that was tested, a set of 10 or 20
realizations were run and compared with current history.
The simulator (ECLIPSE) was forced to produce the same
oil rate as the observed production, while the simulated
bottom hole pressures (BHP) and gas: oil ratio (GOR) were
compared with the actual measured values. The BHP was
the most important matching criterion, since downhole
pressure gauges were installed in well A-10AH and A-13H
(Fig. 1). Well A-11 was not used in matching as it penetrates
both the East Flank and the Central Area, and the relative
contribution from each segment is uncertain.
+ In the first round it was found that all realizations were too
optimistic compared with historical BHP through time. This
suggested that a denser fault pattern was needed in the
model, so HAVANA was re-run. After some iterations, two

62

J. C. Rivens et al.

Fig. 6. (a) Response of five random


realizations for well A-13H. The
connected volume is given for each
realization. A close relation of
connected volumes and pressure drop is
seen. (b) The connected volume for
wells A-13H and A-10AH for all
realizations. The realizations that fell
within the two red windows for both
wells were chosen for well planning.

HAVANA models were identified that had a fault pattern


giving realizations at both sides of the historical profile. A
total of 200 realizations was run.
+ In order to understand why realizations came at the positive
or negative side of the historical profile, it was decided to try
fast streamline simulation (RMS: STREAMLINE) in the
geomodel to provide a connected volume parameter. A
systematic correlation was found between the connected
volume and how well the realization matched the BHP (Fig.
6). With lower volumes, a more rapid BHP decline was
observed, which is expected.
+ Based on a range of connected volumes for each well (Fig.
6), ten realizations that gave a satisfactory match were
applied in further well planning (Fig. 7). The match was not
perfect, but, for the purpose of screening new well targets, it
was found to be fully acceptable.

Fig. 7. Improvement of simulated BHP versus time, showing six


realizations from Havana (well A-13H)

WELL PLANNING, DRILLING AND PRODUCTION


The well planning process can be broken into two stages:
screening and detailed planning. The purpose of screening is to
find the approximate position of the well targets, run them in
the flow-simulator and evaluate discounted cost/income during
the life of the well (net present value). Uncertainty evaluations
are very important in screening and it is especially important to
look at possible downsides to ensure that the well economy is
robust.
Screening study
The objective of the new modelling process was to evaluate the
potential of new well targets so a decision could be made on
whether or not to start a second drilling phase on Njord. In
order to be a potential well candidate in the next drilling period,
the production potential of a new well target had to be at least
as good as the actual production from the old well A-10AH,
which had produced about 200 000 Sm3.
A more general experience from Njord production wells was
that wells located in areas with relatively good seismic data
quality produced better than wells in areas with noisy seismic
data. A possible explanation for this could be that poorer
seismic data lead to poor well targeting, but also that the noisy
areas tend to have high fault density and, consequently, smaller
drainage area for the wells. This understanding of the production experience was already included in the model realizations
by allowing a denser fault pattern in areas with bad seismic data
quality.
By analysing all realizations of the model, two target areas for
new wells were identified as promising. These targets were
based on average maps of probability to find undepleted areas
above oilwater contact (Fig. 8), which were created by
averaging the best realizations (in terms of matching) taken

Stochastic model for well planning, Njord Field

63

Fig. 8. Integrated pressure based on the best stochastic realizations


indicates the probability of finding undepleted areas in the Central
Area. Two possible well targets are drawn, one with a branch. Well
1 formed the basis for well A-10BH drilled 15 months after the IOR
study was completed.

from the ECLIPSE simulator. The simulated production


potential varied from about 0.25 to 0.65 MSm3 (see Fig. 9 for
one of the wells). This showed that the proposed wells were
expected to be better than the old well A-10AH, but not as
good as the best existing well in the Central Area.
Different well types, including branched wells, were simulated and the conclusion was that a well in the Central Area was
robust enough to be included in a new drilling phase. Final
optimization of the well was not done at this stage, as new 4D
and azimuth seismic datasets were still undergoing delivery.
The model was also used for testing various unconventional
well types, for instance the so-called connector wells. These are
open holes in the reservoir that are drilled to connect fault
blocks bounded by sealing faults, but the wellbores are not
connected to the surface after drilling is completed. The idea is
that such wells can be drilled from a floating rig, without
interrupting daily production from the Njord platform. Simulated recovery from a set of four connector wells combined

Fig. 9. Risked reserves for one of the wells from the screening study.

Fig. 10. Flow simulations show that the use of four connector wells
(in yellow) will double the recovery with existing wells.

with the existing producer wells in the Central Area suggests


that the estimated recovery factor could be doubled (Fig. 10),
but more work is needed to prove that connectors will be viable
both technically and economically. However, a simpler concept,
with connector wells drilled as branches from a mother-well
and into undrained areas, was identified as economically interesting. This connector well branch is not the same as an
ordinary dual-lateral branch, since the branch is not physically
connected to the main well tubing through a junction.

Detailed well planning and drilling


The screening process established an initial well-path proposal
and a decision to proceed was made. However, the uncertainty
of the digital model presented here made it unsuitable as a
geometric tool for more detailed well planning. Instead, the
newly arrived seismic surveys made it possible to plan the well
based on enhanced interpretations.
The drilling of the new well (called A-10BH) confirmed the
complexity of the reservoir and the detailed model made for
well-planning had to be revised several times during the drilling.
Various techniques combining 3D visualization, frequent biostratigraphic sampling and active geosteering were necessary to
make a good well path. The well planning and subsequent
geosteering is described in detail in Dart et al. (2004, 2005).

64

J. C. Rivens et al.

Fig. 11. The jump-correlation problem. Interpreting reflectors


crossing fault blocks.

Production
One and half years after the well A-10BH was drilled, the
cumulative oil production had reached 0.405 MSm3. This compares well with the prognosed value in Figure 9 (values by
mid-2004). This clearly suggests that the complex modelling
descibed here was able to predict the production of this
challenging reservoir successfully.
DISCUSSION
An attempt has been made to assess the depth uncertainty by
taking the best guess of the structural model and shaking it
using HORIZON and HAVANA. But was it possible to assess
the full uncertainty prior to drilling? In retrospect it is thought
not, as no consideration was given to large seismic interpretation uncertainty. One particular problem is the jumpcorrelation uncertainty. For a fault block with well tie the
uncertainty in interpretation is rather low. However, once the
reflector crosses a fault the interpreter has to find the correct
reflector on the other side, which may be a difficult task when
the seismic data quality is poor (Fig. 11).
The interpretation uncertainty was tested after the IOR
project had finished. Two interpreters worked independently of
each other on the Central Area, having a somewhat different
database but also a conceptually different structural understanding of the area. The difference in interpretation was alarmingly
large (Fig. 12), up to 60 m difference in depth and with a
significantly different fault pattern. In order to take account of
such discrepancies (and, hence, uncertainty) in digital modelling, several scenario-based models have to be made and each
of these will need to add the uncertainty as described earlier
in this paper. In practice, such an approach would be too
time-consuming with current software and human resources.
Still, it is thought better to evaluate some of the structural
uncertainty range rather than simply use a conventional rigid
model.
The assumption in this modelling that all faults with throws
greater than 10 m are sealing is an oversimplification. Based
on the high shale content in this tidally dominated reservoir
rock, it is possible that most faults represent a severe restriction
to flow. However, in most cases, fault sealing is not static; once
a pressure drop over a fault is large enough, the capillary entry
pressure within the fault may be exceeded and flow could be
established (Rivens & Dart 2002). Fault leaking after some
production time may explain why the BHP in well A-13H (Fig.
7) has stabilized around a constant level (130 bar). Most of the
model realizations have an increased pressure fall after some

Fig. 12. Two seismic interpreters give a quite different interpretation


of the same line. (a) The interpreter assumes rather steep faults; (b)
the other interpreter uses sloping faults. The depth difference of the
two interpretations was up to 60 m.

time, which gave a poorer match from 2001. Attempts have


been made to enhance the match through various means, e.g.
by investigating the effects of relative permeabilities, but the
dynamic leakage across faults remains the most probable cause.
The treatment of sub-seismic faulting in the Njord Central
Area model may be somewhat incomplete in the sense that
sub-seismic tips of seismically visible faults have not been
acknowledged fully. For some of the mapped faults this was
compensated for by the interpreter who extended the fault by
best-guess, usually to make a termination at a fault junction.
HAVANA has the capability to extend the sub-seismic tips of
seismically visible faults, but in this study, it was decided not to
implement this feature, since most of the faults on Njord are
truncated by other faults. However, it cannot be ruled out that
consistent fault tip extension could give a better model.
There is no doubt that modelling the full reservoir uncertainty (surfaces, faults and properties) as outlined in this study
was a cumbersome process and it is possibly too complicated
for ordinary (asset team) model work. An additional problem
is that proper treatment of interpretation uncertainty and
dynamic fault seal is not captured well enough in existing
modelling software as far as is known. Hence, the challenge to
the software vendors is to streamline all aspects of structural
uncertainty modelling in their tools.
CONCLUSIONS
1. The low recovery factor (6%) in the Njord Central Area
necessitated an IOR study in order to prove that further
drilling would be economically viable.
2. A technique was established using HAVANA, HORIZON,
RMS and in-house software to provide 3D grid realizations
that reflected the uncertainty in faults, horizons and
properties.

Stochastic model for well planning, Njord Field


3. In the history-matching process, changes in the rock properties (in this study mainly fault density) were carried out in
the geomodel. This was done to ensure that history matching was predictive for the full field while screening for new
well targets.
4. Combined with stochastic facies and petrophysical parameters, the model supported that at least one well target
would be economically feasible. A well based on this work
was drilled subsequently (well A-10BH).
5. Drilling and later interpretation work revealed that the
structural uncertainty was even higher than the model
presumed. Careful geosteering while drilling was necessary
to make an optimal well path.
6. The actual production from well A-10BH is quite close to
the prognosis made during modelling and flow simulation
prior to drilling.
7. Connector wells (open holes connecting reservoir segments
but not connected to surface) may increase the pressure
support and, hence, improve the recovery factor.
8. Modelling structural uncertainty is cumbersome with todays
tools, which presents a challenge to the software vendors.
The authors thank Norsk Hydro, Petoro, ExxonMobil, Paladin,
ConocoPhillips, OER and Gaz de France for permission to present
the material in this paper. Thanks also go to the scientists at the
Norwegian Computing Centre for valuable support and discussion
on software tools, and Jon Dexter (Norsk Hydro) and the reviewers
for useful suggestions.

REFERENCES
Abrahamsen, P., Omre, H. & Lia, O. 1991. Stochastic models for seismic
depth conversion of geological horizons. Oshore Europe 91, 36 September,
Aberdeen.
Blystad, P., Brekke, H., Frseth, R., Larsen, B., Skogseid, J. & Trudbakken,
B. 1995. Structural elements of the Norwegian Continental Shelf, Part II: The
Norwegian Sea region. Norwegian Petroleum Directorate Bulletin, 8.

65

Clemetsen, R., Hurst, A. & Omre, H. 1990. A computer program for


evaluation of fluvial reservoirs. In: Buller, A.T., Berg, E., Hjelmeland, O.,
Kleppe, J., Torster, O. & Aasen, J.O. (eds) North Sea Oil and Gas Reservoirs
II. Graham & Trotman, London, 373385.
Dart, C., Cloke, I., Herdlevr, . et al. 2004. Use of 3D visualization
techniques to unravel complex fault patterns for production planning:
Njord field, Halten Terrace, Norway. In: Davies, R., Cartwright, J., Stewart,
S., Lappin, M. & Underhill, J. (eds) 3D Seismic Technology: Application to the
Exploration of Sedimentary Basins. Geological Society, London, Memoirs, 29,
249261.
Dart, C., Denichou, J., Bridger, T. et al. 2005. Three-dimensional visualisation
for well placement and geosteering: Njord Field, offshore mid-Norway. In:
Dor, A.G. & Vining, B.A. (eds) Petroleum Geology: North-West Europe and
Global Perspectives Proceedings of the 6th Petroleum Geology Conference. Geological Society, London, 14871506.
Dreyer, T., Sharp, I. & Lien, T. 1999. Update of the Tilje Formation reservoir
geological model in the Greater Njord area. Internal Norsk Hydro Report.
Hollund, K., Mostad, P., Nielsen, B. et al. 2002. Havana a fault modelling
tool. In: Koestler, A.G. & Hunsdale, R. (eds) Hydrocarbon Seal Quantification.
NPF Special Publication, 11, 157171.
Lilleng, T. & Gundes, R. 1997. The Njord field: a dynamic hydrocarbon
trap. In: Mller-Pedersen, P. & Koestler, A.G. (eds) Hydrocarbon Seals:
Importance for Exploration and Production. NPF Special Publication, 7,
217229.
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, 1728.
Osmundsen, P., Sommaruga, A., Skilbrei, J. & Olesen, O. 2002. Deep
structure of the Mid-Norway rifted margin. Norwegian Journal of Geology, 82,
205224.
Rivens, J. & Dart, C. 2002. Reservoir compartmentalisation by watersaturated faults Is evalution possible with todays tools?. In: Koestler, A.G.
& Hunsdale, R. (eds) Hydrocarbon Seal Quantification. NPF Special
Publication, 11, 173186.
Shanor, G. & Whelan, M. 2004. A review of reservoir modelling methodologies. Paper presented at the 29th Annual Indonesian Petroleum Association (IPA) Convention and Exhibition, Jacarta, 1416 October, 2003.
Yielding, G., Needham, T. & Jones, H. 1996. Sampling of fault populations
using sub-surface data: a review. Journal of Structural Geology, 18, 135146.

Received 17 November 2003; revised typescript accepted 19 October 2004.

Vous aimerez peut-être aussi