Sunteți pe pagina 1din 52

SPE Paper Number SPE-133518-PP

UNCERTAINTY MANAGEMENT: A STRUCTURED APPROACH TOWARDS


RECOGNIZING, QUANTIFYING AND MANAGING SAMPLING BIASES IN
SUBSURFACE UNKNOWNS*
Laurent Alessio, Leap Energy Partners Sdn Bhd, Arnout Eerts, Leap Energy Partners Sdn Bhd, and !aee" #ahmat,
Leap Energy Partners Sdn Bhd
$opyright %&1&, So'iety o( Petroleum Engineers
)his paper *as prepared (or presentation at the SPE Asia Pa'i(i' +il , -as $on(eren'e and E.hibition held in Brisbane, /ueensland, Australia, 180%& +'tober %&1&1
)his paper *as sele'ted (or presentation by an SPE program 'ommittee (ollo*ing reie* o( in(ormation 'ontained in an abstra't submitted by the author2s31 $ontents o( the paper hae not been reie*ed by the So'iety
o( Petroleum Engineers and are sub4e't to 'orre'tion by the author2s31 )he material does not ne'essarily re(le't any position o( the So'iety o( Petroleum Engineers, its o((i'ers, or members1 Ele'troni' reprodu'tion,
distribution, or storage o( any part o( this paper *ithout the *ritten 'onsent o( the So'iety o( Petroleum Engineers is prohibited1 Permission to reprodu'e in print is restri'ted to an abstra't o( not more than 3&& *ords5
illustrations may not be 'opied1 )he abstra't must 'ontain 'onspi'uous a'6no*ledgment o( SPE 'opyright1
ABSTRACT
This paper illustrates, through field studies examples, why and
how a structured approach towards managing uncertainties, and
especially sampling biases, delivers valuable insights through the
successive early asset life stages -exploration, appraisal and field
development phases. In doing so, we respond to three
fundamental questions.
Firstly, hat are the !ey uncertainties " those that matter#$ Field
studies should begin with a comprehensive upfront assessment of
uncertainties$ impact on historical and future well and field
performance. %owever, often ma&or factors are overloo!ed,
leading to under-prediction of true outcome ranges and the
inability to reconcile historical production. 'ur illustration is a
large producing carbonate field, where after () years of
production, large scale *arstification was finally evidenced to be
the explanation for the field performance that couldn$t be history
matched with the measured matrix porosity and permeability
ranges.
+econdly hat are realistic ranges for these uncertainties#$
*nown Industry best practices include intensive expert-assist,
integration of drilling, mud-logging and other traditional sources
of data from the field, resorting to analogue benchmar!ing.
,espite these, we often fail to understand and correct for
sampling bias, which we show often leads to over-optimism. The
paper will highlight why such biases are present and propose
simple and practical methods to remove them. The case study is
the volumetric assessment of a gas discoveries portfolio, where
geophysical techniques were instrumental in exploration and
appraisal drilling.
Finally %ow these uncertainties will evolve with time#$ This is
an important question for assessing value of Information- the
impact that additional data may have on the uncertainty range of
uncertainties and the base case. .nconventional fractured plays,
often characteri/ed by data abundance but extreme variability,
provide surprising insights on how uncertainties ranges evolve.
This paper presents methods to develop confidence curves for
important parameters.
INTRODUCTION
This paper illustrates, through field studies examples, why and
how a structured approach towards managing sampling biases in
reservoir evaluation delivers valuable insights through the
successive early asset life stages exploration, appraisal and field
development phases.
hilst the purpose of the paper is not to provide a comprehensive
review of uncertainty management best practices, a wor!flow and
fundamental steps are discussed herein, to provide some
contextual framewor!. e then focus our illustration around
identification of !ey uncertainties, defining realistic ranges for
these, and finally assessing how ranges should evolve with time.
The first step through the .ncertainty 0ssessment wor!flow is
Identification. e$re essentially responding to the question hat
are the !ey uncertainties " those that matter$. 1ost subsurface
professionals would agree on the need for a comprehensive
assessment of uncertainties conducted upfront, and developing an
understanding of their impact on historical and future well and
field
23reviously presented at the I30 0nnual 4onvention, 5a!arta 67(7
performance. %owever, often ma&or factors are overloo!ed,
leading to under-prediction of true outcome ranges and when
applicable, the inability to reconcile historical production. 0 field
case of a very large producing carbonate field, where after over
() years of production history, the main uncertainty impacting
field performance was found to an intense *arstification8 a factor
that had been essentially overloo!ed until, as part of a ma&or field
review, a comprehensive reservoir modelling history match-ing
exercise highlighted the impossibility to obtain a match with the
measured carbonate matrix porosity and permeability ranges.
The second step in the wor!flow is 0ssess 9anges$- this equates
to as!ing the following question what are realistic ranges for
these uncertainties#$ This is a very typical issue for green field
developments. *nown and well documented Industry best
practices include intensive expert-assist, integration of drilling,
mud-logging, and other traditional sources of data from the field,
resorting to analogue benchmar!ing. hat remains generally an
issue is our ability to understand and correct for sampling bias "
unfortunately at the early stage of the asset life, bias often leads to
over-optimism. The paper will highlight why such biases are
present and propose simple and practical methods to remove
them. The illustrating case is a portfolio of clastics gas
discoveries. :eophysical techniques are often instrumental in
exploration and appraisal drilling, and as a consequence,
assessing sampling bias and correlations are a very important step
in the volumetric assessment.
The third and final question is how some of these uncertainties
will evolve with time. This is an important question to answer for
an adequate assessment of ;alue of Information <;oI= but also to
understand how much impact should additional data have onto
the range of uncertainties, and of course, the base case. Fractured
resource plays, with their abundance of data but extreme
variability, provide surprising insights on how uncertainties
ranges evolve. This paper will present a practical method to
develop confidence curves for important parameters.
UNCERTAINTY MANAGEMENT BEST PRACTICES
AND CONTEXT
>ecause of the inherent difficulty in understanding what cannot
be clearly seen and measured " the subsurface, it is recogni/ed
widely that managing uncertainties is critical to ?@3 ventures,
and therefore a very important responsibility of the +ubsurface
and Front ?nd teams. It is by nature an inter-disciplinary
capability that operators must develop and apply across their
assets.
0s a consequence, fairly well established practices do exist, and
have been published in the past <0lessio et. al., 677)=8 <4harles
et. al., 677(=8 <4orre et. al., 6777=, so it is not the ob&ective of this
paper to provide a thorough account of the best practices of
uncertainty management,
but rather to focus on the particular aspect of sampling bias and
its impact on the uncertainty management wor!flows.
.ncertainty management can be bro!en up into two main phases-
0ssessment and 1itigation. ?ach of these can be articulated
around a number of steps- an account of those is proposed in
Figure (. +ampling bias problems do occur mostly in the
0ssessment phase, and so this is the part that we will be focusing
in this paper.
IDENTIFICATION OF UNCERTAINTIES
Ensuring all critical infr!atin an" "ata is cnsi"#r#"
Aet$s introduce the first problem- why despite elaborate and
rigorous wor!flows, using probabilistic and geostatistical
techniques <4harles et. al., 677(=8 <4orre et. al., 6777=, it still
sometimes happens that field performance expectations fall
outside the 3B7C3(7 range. The first example provides an
interesting case whereby a producing field <which we will call
Field D through this paper= consistently out-performed
expectations, and couldn$t initially be history matched
conventionally, despite the availability of a fairly extensive core
dataset, and post-drilling transient testing data on the early
production wells. Field D is a large, multi-Tscf, carbonate gas
field, located in a prolific carbonate province, in 0sia.
Field D$s production started in (BEF from a total of (( deviated
wells placed in the central part of the field, drilled on a dataset of
multi-vintage, good quality 6, seismic. Aittle was !nown about
the detailed reservoir architecture away from well penetrations.
Following () years of production and half of the expected
reserves produced, a G, survey was acquired in mid 6776 to
support an infill drilling campaign and a potential field re-
development. 0t the same time, the field review initiative started
and whilst seismic processing and inversion was carried out and
iterated, a first pass material balance and static-dynamic
modelling exercise was initiated.
These early static-dynamic modelling iterations rapidly led to the
conclusions that a combination of enhanced permeabilities in the
lower producing intervals and possibly higher volumes were
required to match <and slow down= the combination of rise of
contact and pressure data in the field. This was in part consistent
with the fact that hori/ontal permeabilities derived from core data
were consistently lower than the well-test derived ones. In
addition, it was found that lower gas residual saturations in these
/ones also improved the quality of the history match, and so
could greater gas volumes in that /one. This indicated the
possible role of a non-matrix element to flow, either fractures or
*arsts. In the earlier dynamic models <pre-6776=, permeability
multipliers were applied to layers to match the well test results,
but neither the mechanism nor the spatial distribution of the
property enhancements were properly understood.
0s these observations were consistently reported from the
reservoir engineer bac! into the subsurface team, a new
hypothesis emerged- presence of an extensive *arsts networ! in
the lower part of the reservoir could explain the required
enhancement of properties of this /one <permeability, storability
and residual gas=. 0 close examination of the drilling history,
bac! in the late E7
$
s, confirmed that nearly E7H of the wells suffered some losses in
this interval, and about )7H experienced very severe losses.
Following this lead, seismic multi-attribute extractions revealed
an extensive dendritic *arsts networ! <>ourdon et. al., 677I=8 <A.
>ourdon, 677I= covering the vast ma&ority of the Aower reservoir.
>ecause of their shear si/e and density, this imaged *arsts system
could finally explain the matrix property enhancement, in terms
of volumes, effective permeability, and petrophysical property
alterations.
The process of introducing a new facies <*arsts= provided the
team with a defendable, plausible, explanation for the
performance of the field, without reverting to arbitrary transforms
of the reservoir properties.
It is here interesting to ac!nowledge the sampling bias issue with
this case study- despite very early significant observations that
non-matrix properties had to be responsible for the massive and
consistent drilling induced losses, this fact was not eventually
carried through, as !nowledge, into the later phases of reservoir
management.
.p until it was recognised that the field performance couldn$t be
reconciled using the sampled ranges of petrophysical parameters
<porosity, permeability, residual gas saturation=, the uncertainty
assessment wor! essentially ignored a fundamental parameter.
This is an extreme case of sampling bias- the omiting of a critical
parameterJ The lessons learned from this case study are-
?nsure all possible sources of information are
considered in the first step of the .ncertainty
0ssessment stage <Figure (=.
Field reviews must be a fully multi-disciplinary
exercise
?nsure a learning system is in place within the
organisation to ensure critical information is retained
DEFINING REA$ISTIC RANGES FOR UNCERTAINTIES
%&at ar# r#alistic rang#s fr unc#rtainti#s'
In this part, the technical contribution of this paper is to provide
some practical insights, through another case study, towards
sampling bias presence in volumetric assessment.
This case study depicts a fairly common situation where
exploration is conducted using seismic attribute high-grading
techniques, such as 0;', inversion etc. ,irect hydrocarbon
indicators <,%I= have become a very popular de-ris!ing
exploration application that has helped increase dramatically the
<3g= probability of geological success <9oden et. al.=.
0n indirect consequence of this improvement of geological
success is the introduction of a sampling bias, within the fields
that are being discovered. Kot dissimilarly to the basin creaming
curve effect well !nown to explorationists, guided exploration
and appraisal drilling has a tendency to produce a s!ew in the
petrophysical and geological sums and averages, for a
combination of the following reasons- ?arly exploration wells
are drilled on amplitude$
highs which are hoped to and often do hold higher net
hydrocarbon volumes
?arly exploration wells are drilled on structural highs,
to maximise the chance of encountering a charged hydrocarbon
column, hence encountering nonrepresentative elevated
saturations
,elineation wells may be drilled down-dip to test
minimum economic columns, and therefore may find
low saturations Kone of these remar!s should come at a surprise
to the seasoned subsurface professionals. 0ccounting for such
biases is however not always straight-forward, will involve
elaborate data analysis and integration and may actually
emphasise remaining uncertainty in the field.
Cas# stu"() *rtfli f gas "isc+#ri#s, "rill#" t&rug& D-I
.Dir#ct -("rcar/n In"icatrs0
0t an early appraisal stage, we are often confronted to a dataset
of wells that were predominantly targeted at amplitude sweet
spots$. 9emoval of sampling bias will typically start with
obtaining statistics of the area covered by sweet spots$ versus the
total field area. This would then be combined with either
geologically andCor geophysically driven estimates of the
expected reduction in net pay andCor reservoir properties in the
seismically dimmer$ areas compared to the sweet spots$.
To illustrate the issue of dealing with a biased well dataset
consider the case study of Field L, consisting of fluvial and
marginal marine sandstones in a three-way dip closure with a
bounding fault and possibly a stratigraphic trapping component.
0n amplitude map of the main reservoir level in Field L is shown
in Figure G. Two wells were drilled to appraise the structure and
from the amplitude map, it is obvious that both wells were
specifically targeted at reservoir Msweet spotsN. The problem
presented to the asset team was to arrive at a realistic range in net
pay for the field despite the obvious bias in the well data. The
wor!flow followed by the team on this particular example
is one of the many possible methodologies. First, geophysical
analysis and seismic modelling studies established that for the
main reservoir in Field L, the thic!ness cut-off for
M
seismically visible pay
N
is around
G.) to ).) m <about (CE
th
of the seismic wavelength= whilst the
maximum constructive interference <i.e., brightest amplitudes= are
expected for a pay thic!ness of around F to (( meter <around O of
the seismic wavelength=. This information was then used to
consistency chec! and complement the well data in establishing
the relationship between pay and seismic amplitude <Figure I=.
%istograms of amplitude distribution in the field were then used
to determine Marea weighedN average amplitude for the amplitude
and non-amplitude supported domains in the field, as well as
ranges of pay expected in these two domains. 0s expected, the
range in pay for the amplitude supported area of the field as
estimated with this method <Figure I= is substantially less than
the average of the two wells.
Sa!*ling /ias an" 1D g#statistical !"#lling 2r3fl2
The issue of sampling bias is again at play when it comes to more
sophisticated resource assessment techniques involving G,
reservoir modelling. 9eason being, commonly used techniques to
stochastically simulate reservoir andCor property distribution in
6, or G, domain such as :aussian simulation, are designed to
replicate the statistics of the input data. If the input data consists
of a biased set of wells li!e in the example of Field L, the ris! is
this bias will be extrapolated to field scale. hilst it is true that
modern mapping and G, modelling tools do provide ample
mitigation options against this ris! such as use of areal trends,
such sophistication is not always applied especially not in Mfast
trac!N assessments <which, not surprisingly, tend to lead to over
optimism=.
1oreover, without considering the possibility of the actual
statistics of pay andCor property distribution in the field being
different from what is seen in the wells, repeated stochastic
simulation in nested wor!flows -a popular technique in modern
field assessment -will simply reproduce the well data -albeit with
different areal distribution -resulting in unrealistically narrow
uncertainty ranges for the field. In other words, to cover the full
range of outcomes for a field, it is important to consider the
possibility of a bias in the well data especially if the data is
scarce.
E4O$UTION OF UNCERTAINTY RANGES AS A
FUNCTION OF A4AI$AB$E DATA
A 5uantitati+# a**rac& using Cnfi"#nc# Cur+#s
O/6#cti+# an" *r/l#!
In this part, we are focusing specifically on highly heterogeneous
systems, such as highly fractured carbonates, chal!s, or
unconventional reservoirs such as tight sands, shales and coalbed
methane. e$re trying to
establish a systematic method to quantify, as a function of the
amount of data available, the uncertainty in a field average
metric, such as an average permeability, or pea! rate per well.
4aria/ilit( an" unc#rtainti#s
ith highly heterogeneous systems, where large differences are
found from one observation <ex- a given well= to another
observation <a neighbouring well=, it is important to recognise the
distinction between variability and uncertainty, as these are two
often confused for one another, and thereby leading to significant
misrepresentations of field uncertainty ranges.
;ariability is defined as a short to medium scale <up to
inter-well scale= variations of a given parameter, such as
permeability, porosity, gas content <for 4>1 reservoirs=,
hydrocarbon saturation etc. These variations can be often
extreme, with several orders of magnitude differences in
permeability commonly observed in fractured reservoirs.
;ariability is intrinsic and non-temporal, which means it that does
neither change with time nor with the number of data points, and
it is a characteristic of the reservoir <for a given sampling scale2=.
.ltimate understanding of variability often remains spatially
poorly predictive, so the authors recommend a statistical
approach is always conducted in parallel.
.ncertainty is defined at the field scale, or at least, a
sector or segment of the field <field unit=, where multiple wells
will be ultimately drilled. It represents, at a given time, how well
a field unit is understood. :enerally, uncertainty reduces with
time and information becoming available, provided the right
framing and uncertainty assessment was conducted <ref previous
section=. 0rguably, the uncertainty in subsurface givens, such as
field porosity average, or in place volumes is strictly a
consequence of our lac! of !nowledge. ,evelopment related
metrics, such as field recoverable volumes or production
performance, at a given time, are a consequence of our level <or
lac! of= of understanding of the subsurface and the concept
development choices we have made and will be ma!ing.
2Kote- variability is indeed intimately lin!ed to sampling scale8
for the purpose of this paper, we are assuming that all measures
are ta!en at the same scale <ex- vertical wells through a reservoir
section=, and we will not discuss this point further.
A statistical r#*r#s#ntatin f +aria/ilit(
;ariability of a particular parameter can always be represented by
a 3,F <3robability ,ensity Function=. In our approach, we are
using lognormal curves as those describe well a number of
naturally occurring phenomena and parameters in the subsurface,
such as permeability of a
fractured system. %owever, the approach can be used for any
statistical distribution, as we
$
re using a discretisation of the curves to conduct our analysis. The
Figure F shows an example distribution <this is an illustration
only and not associated with a specific field or asset=
D#+#l*ing t&# cnc#*t f cnfi"#nc# cur+#s fr! a *ractical
/s#r+atin
The idea of 4onfidence 4urves emerged from a practical
observation of a resource play, where the variability of production
metrics such as pea! production rates <gas, water= was assessed.
>ecause of the very large variability of !ey subsurface parameters
such as permeability usually present in fractured resource plays,
estimating field or area performance with the !nowledge of only
a few wells$ actual performance can be highly misguided, and
lead to erroneous conclusions on the attractiveness of an acreage.
In essence, this large variability, where pea! rates or permeability
can vary well over a log cycle or more, creates an intrinsic
sampling bias when only a few samples <wells= are available.
This sampling bias must be recognised, understood, and
accounted for in the assessment of expected field averages.
The problem at hand is as follows- having drilled a number p$ of
wells in an area where it is planned to drill a total number n$ of
wells, and having established !ey subsurface parameters for each
of these wells, what level of confidence can we have in the
averages computed from these wells to be representative of the
final n$ wells average.
In trying to answer this question, the subsurface team selected as
an observation set a large enough area, that could be considered
fully drilled and developed by some (77 wells. .sing some
assumptions, productivity of each well was calculated, and
converted to an average well permeability. It was therefore
possible to create a distribution of average well permeability over
this area. The wor!flow presented in this section was devised on a
real field case, but for the purpose of this paper, an illustrative
case with generic and non-case specific numbers are used. The
methodology is re-constructed around this illustrative case. The
parameter input 3,F <well permeability= that are used in the
following sections is shown on Figure F " note this is illustrative
variability distribution curve was selected to have a variance that
is not uncommon of permeability in fractured reservoir plays
<3(7C3B7 P (7=
Kow, assuming the si/e of the area is significant enough,
i.e. (77 wells are a sufficiently large dataset, this distribution can
be ta!en as representative of the variability of the field. This
assumption only holds provided that this area is itself
representative of the rest of the field, and we are not again
suffering from a sampling bias, at a larger scaleJ Kext, the
following mathematical operations were applied
For a range of p values <p number of wells=, within the
universe of nQ(77 values-
4ompute a statistically significant number of average values
resulting from combinations of p samples within nQ(77. Kote that
it is not easy to compute all the possible ones- 4(77
(7
R (.Fe(G, so
a practical approach is called for. In our case, to !eep within
spreadsheet limitations, a maximum of G7,777 simulations was
used.
3lot the averages of p$ wells combinations against the number
of wells drilled p$. The D-axis can also be normalised to a
normalised area by using pCn$. This plot will be called
$3arameter .ncertainty 4urves$
0n example plot of the resulting 3arameter .ncertainty curve is
shown on Figure ) +uch curves can be read as follows8 note that
as a parameter we are going to use productivity$- In a given
area, where eventually a number n$
21&& (or our e.ample3 o( *ells may be drilled, *e
are loo6ing at the possible alues o( aerages o(
7p8 *ells, 'hosen randomly1
9e are essentially simulating the possible out'omes o(
sampling randomly this resour'e play and obsering the
range o( hypotheti'al tra4e'tories o( a 'omputed aerage
2:ean31
)hese 'ures proide a isual and ;uantitatie assessment
o( the possible tra4e'tories o( ho* aerages o( populations
o( *ells 'ould eole *ith the number o( *ells drilled1 )he
e.ample sho*n on !igure <, sho*s ho* a reseroir
engineer reie*ing the results o( early *ells and 'omputing
a aerage *ell 'ure 2or type *ell 'ure3, 'ould initially post
a ery high aerage, his assessment (alling i'tim o( haing
sampled a (e* 7high8 outliers1 )hen, *ith more drilling, the
ariability P=! 'ure is sampled, and (or the sa6e o( this
e.ample, lets assume no* that the 7lo*8 end o( the P=! is
sampled predominantly1 >n this 'ase, the red tra'e *ould
sho* the assessment o( a representatie 7:ean8 based on
the population o( *ells aailable at gie times1
)his e.ample is eidently sele'ted to illustrate the point that
under large ariability, as it is 'ommon *ith (ra'tured plays,
there is a real possibility (or e.treme sampling biases to
o''ur and 'loud a 7type8 'ure assessment1 )his has a
parti'ularly important impli'ation (or appraisal o( highly
ariable resour'e play? >( the play is e.pe'ted to be ery
ariable, then the
un'ertainty at an early drilling phase is ery
signi(i'ant1 >n the e.ample presented, the true
un'ertainty *hen less than 5@ o( drilling has
o''urred is 'a1 3 (old1 )hat means the (inal (ield
aerage may be *ithin a (a'tor 3 o( the 'urrent
'omputed aerage 2$$A31


)he un'ertainty de'line rapidly *ith a (e* early *ells, sin'e
the pro'ess o( drilling these *ells redu'es s*i(tly the
'han'es o( sampling 'onsistently outliers in the same part
o( the 'ure 2either Ai or Lo31 !or the 'ures presented in
this e.ample, this o''urs in the (irst 5@ o( the drilling1 >t
may be possible to more rigorously generalise this result,
(or arious parameter ariability input P=!s1
)he important (indings at this stage are that, (or highly
ariable resour'e plays, *here intensie drilling does o''ur?
1 Assuming 6no*n the ariability o( an important
parameter 2e.? *ell permeability3, it is possible to predi't
'on'eptually the un'ertainty bands o( a (ield aerage s1 the
amount o( data and 6no*ledge aailable1
2 9ith highly ariable plays, early in(ormation (rom a (e*
*ells 2B5@ drilling o( the play3, 'an be ery misleading1 #ealisti'
un'ertainty bands may be up to *ithin a 3-5 (old range o( the
mean o( *ells drilled to date1
3 )he un'ertainty band de'reased rapidly *ith the (irst
1&@ o( the *ells and hit a pleateau *ith %&@ o( the *ells drilled
2!igure <31 )his suggests the possible optimum early drilling *ell
'ount o( up to %&@ o( the total *ells drilled in a target area as
in'reased *ell 'ount *il not signi(i'antly redu'e the un'ertainty1
Intr"ucing t&# cnc#*t f Cnfi"#nc# Cur+#s
e are introducing here a particularly effective manner to
construct quantitatively, versus available data or time, a
representative range for field-wide uncertainty metrics
such as 3B7, 3(7. e$re also able to answer more
generally the following question-
For a given level of confidence <for instance E)H=, what is the
level of uncertainty in the final field average parameter <for
instance permeability=, measured as a fold or multiplier of the
current measured average from the available data#
0s a function of the available data collected to date. In our
approach, we are using as a measure the percentage <H= of data
available. For resource plays, a percentage <H= of wells drilled
can be used, if one wants to assess the uncertainty in field overall
well rates or average field permeability.
:iven our understanding of the variability- importantly, a
model of the statistics of the variability needs to be !nown, or
assumed. This can be achieved either through a direct measure, or
by analogy.
The resulting assessment can be plotted as shown on Figure F
where the confidence curves, expressed as a fold
of the current 1ean can be represented vs. the percentage of data
available.
A**licatins f t&# cnfi"#nc# cur+#s !#t&"lg(
4onfidence curves were initially devised for resource plays,
where large variability is present and uncertainty is !nown to
remain high after appraisal and only progressively reduce through
the development phase. They provide a structured way to conduct
three main assessments- .ncertainty levels as a function of
available data, and
there(ore determining a realisti' (ield un'ertainty range
(rom an understanding o( ariability1
=etermine the alue o( in(ormation (or ne* data5 Co> 'an
be e.pressed in terms o( (ield un'ertainty redu'tion
potential1 Co>Ds 'an a'tually be statisti'ally assessed 2this
*ill be deeloped in a (uture publi'ation31
=etermine i( areas *ithin the resour'e play 'an be
4usti(iably high-graded or do*n-graded 0 statisti'al
representatieness 'an again be determined (rom the use
o( 'on(iden'e 'ures1
Futur# ar#as f stu"(
The technical contribution of this paper$s section on confidence
curves is limited to understanding their basic construction and
simply highlighting their possible applications. Future wor! will
be focus on developing further the !ey applications areas of these
curves.
CONC$USIONS
e have presented, through examples, how critical are the
.ncertainty Identification and 9ange 0ssessment phases in the
.ncertainty 1anagement or!flow <.1=, and provided a
number of tools and approaches to improve the recognition,
assessment and management of subsurface un!nowns.
Firstly, we have shown how !ey subsurface features may be
omitted if not all data and information sources are considered8 the
necessity of a multi-disciplinary approach is highly
recommended, in order to reduce this ris!. +econdly, we are
providing clear examples of how sampling bias creeps into the
subsurface assessment wor!, and provided practical illustration
on how this phenomenon can be accounted for and removed.
Thirdly and finally, we have proposed a method to quantify
uncertainty based on an understanding of variability vs. available
date, using 4onfidence 4urves.
The methodologies and practical solutions to the problem of
sampling bias in quantifying uncertainty ranges for subsurface
assets have been illustrated through case studies inspired from
actual field reviews, field development planning pro&ects and
other subsurface assessments8 note that for the purpose of this
paper, all confidential information has been duly removed by the
authors, data, maps, well results have all been suitably altered so
that no sensitive information is made public.
AC7NO%$EDGMENTS
+pecial than!s to 3eter Friedinger and 0rtur 9yba for providing
valuable insights and support for the production of this paper.
+pecial than!s to Indonesian 3etroleum 0ssociation <I30= for
granting permission to publish this paper.
REFERENCES
A.0lessio, +. 4oca, A.>ourdon, 677), M?xperimental ,esign as a
Framewor! for 1ultiple 9ealisation %istory 1atching- FS
Further ,evelopment +tudiesN. +3? BG(SI presented at the +3?
0sia 3acific 'il and :as 4onference and ?xhibition held in
5a!arta, Indonesia, ) " F 0pril 677).
A.0lessio, +. 4oca, A.>ourdon, 677I, TG,-0ll-The-ayT In FS
Field Further ,evelopment. +3? EF7(I presented at the +3? 0sia
3acific 4onference on Integrated 1odelling for 0sset
1anagement held in *uala Aumpur, 1alaysia, 6B-G7 1arch
677I.
A. >ourdon, A. 0lessio, +. 4oca, 677I, *arsts Identification and
Impact on ,evelopment 3lan. +3? EE)67, presented at the +3?
0sia 3acific 'il and :as 4onference and ?xhibition held in
3erth, 0ustralia, (E"67 'ctober 677I.
A. >ourdon, 677I, .nravelling the reservoir architecture from G,
seismic using multi-attribute ;olume interpretation- 4entral
Auconia *arsts 1odelling. 9:4677I, internal +hell event.
>ratvold, 9., >egg, +.%., 4ampbell, 5.4., 6776, ould you !now
a good decision if you saw one#T, +3? FF)7B.
4heong, U.3 and 9. :upta, 677I, ?xperimental ,esign and
0nalysis 1ethods in 1ultiple ,eterministic 1odelling for
Vuantifying %ydrocarbon In-3lace 3robability ,istribution
4urve, +3? EF776, presented at the +3? 0sia 3acific 4onference
on Integrated 1odelling for 0sset 1anagement held in *uala
Aumpur, 1alaysia, 6B-G7 1arch 677I.
4heong U.3, 9. :upta, 677G, ?xperimental ,esign in
,eterministic 1odelling- 0ssessing +ignificant .ncertainties.
+3? E7)GF.
9obert :. Aouc!s, (BBB, 3aleocave 4arbonate 9eservoirs-
'rigins, >urial-,epth 1odifications, +patial 4omplexity, and
9eservoir Implications. 003: >ulletin, ;. EG, Ko. ((, 3. (FB)-
(EGI.
4heong U.3, 9. :upta, +mith :. et al, ?xperimental ,esign
1ethodology for Vuantifying .9 ,istribution 4urve
-Aessons learnt and still to be learnt, +3? EE)E)
hite 4, 9oyer +., 677G, ?xperimental ,esign as a Framewor!
for 9eservoir +tudies. +3? FBSFS.
4harles T., :uemene 5.1., 4orre >., ;incent :., ,ubrule '.,
677(, ?xperience with the Vuantification of +ubsurface
.ncertainties paper, +3? SEF7G.
4orre >., Thore 3, ;.de Feraudy, ;incent :., 6777, Integrated
.ncertainty assessment for pro&ect evaluation and ris! analysis.
+3? S)67).
;an ?l! 5.F, :uerrera A., ;i&ayan * and :upta 9., 6777,
Improved .ncertainty management in field development studies
through the application of the experimental design method to the
multi-realisation approach. +3? SIIS6.
4harles T, :uemene 5.1, 4orre >, ;incent :, ,ubrule '., 677(,
?xperience with the quantification of subsurface uncertainties.
+3? SEF7G.
1. Feraille, F. 9oggero,?. 1anceau, A.U. %u, I. Labal/a-
1e/ghani, A. 4osta 9eis, 0pplication of 0dvanced %istory
1atching Techniques to an Integrated Field 4ase +tudy. +3?
EIISG
?ide, A. %olden, ?. 9eiso, +. 0anonsen, (BBI, 0utomatic %istory
1atching by use of 9esponse +urfaces and ?xperimental design,
?41'9 F-(7 5une (BBI
T. ?geland, ?. %atleba!!, A. %olden, ?.0 Aarsen, ,esigning
>etter ,ecisions, (BB6, +3? 6I6F).
+axena .mesh, ;&e!oslav 3anelic, Factorial ,esigns as an
?ffective Tool in 1ining and 3etroleum ?ngineering,(BF(, +3?
GGGG.
+awyer et al, Factorial ,esign of et combustion drive, (BFI,
+3? I(I7
9. 9oden, 1. Forrest, 9oger %oleywell. ,'I-(BBEG, ,%I
Threshold ?ffect in 3rospect 9is!ing
Figur# 8) Unc#rtaint( Manag#!#nt %rfl2 .UM%0 ) Unc#rtaint( Ass#ss!#nt an" Mitigatin 2r3fl2s
Figur# 9 ) A""itin f 7arstificatin as a significant *ara!#t#r all2s a !atc& t /# /tain#"
Ampli!"# m$p Fi#l" Z
M$p %$& '##( $l#)#" *)+m $,!$l ,$&#
Figur# 1 ) A!*litu"# !a* f fi#l" :; T&# t2 a**raisal 2#lls "rill#" t "at# /t& us#" t&# s#is!ic t targ#t r#s#r+ir <s2##t s*ts=; T&#
r#sult is a cl#ar #>a!*l# f a /ias#" 2#ll "atas#t; Ma* an" 2#ll lcatins &a+# /##n alt#r#" fr! actual cas# stu"( t #nsur
Area *eighed aerage o(
E
amplitude supported
F
domain Amplitude alue G %11
H1I 0 J1J m pay 2PI&-P1&3
9ell-% 9ell-1 2< m pay3
21%18 m pay3
Amp #ange
Amp #ange
Lo*er amplitude area Lo*er
pay 1-5 m range 2PI&-P1&3
:a.imum
'onstru'tie
inter(eren'e G
1KH seismi'
*aelength G J
0 11 m pay
Limit o( isible pay
G 1K8 seismi'
*aelengthG 315 0
515 m pay
Bright Very Bright (Brightest 5%)
R#l$i-#l. "im R#l$i-#l. ')i/%
0 N+(1Ampli!"# &!pp+)#" 0 Ampli!"# &!pp+)#"
Figur# ? ) -istgra! f a!*litu"# +alu#s in fi#l" : "#ri+#" fr! t&# !a* s&2n in Figur# 1 In"icat#" ar# t&# *a( s##n in t&# 2#lls, t&#
#sti!at#" *a( cutff fr <s#is!icall( +isi/l#= *a( an" t&# *ti!u! tuning t&ic3n#ss rang#; All f t&#s# 2#r# us#" /( t&# ass#t
CALCULATION OF AVERAGES
10.0
P$)$m##) U(,#)$i(. C!)-#&
R#l$i-# p$)$m##) -$l!# -&2 S$mpl# &i3# 45 + 5667
Enelop o( ma.imum deiation to the
(inal mean 2(or 1&& *ells3
No Samples 50
No Samples 1
No Samples 2
1.0
No
Samples 4 0 10
20 30 40 50 60 70 80 90100
No Samples 6
No Samples 8
No Samples 10
No Samples 15
No Samples 20 :ean (or the total population No Samples 25
21&&*ells3, Normalisedto 1
No Samples 30
No Samples
35
0.
1 No Samples 40
No Samples 45
No samples 100
0.0
Figur# @ ) Para!#t#r Unc#rtaint( Cur+#s) *ltt#" /#l2 is t&# r#lati+# *ara!#t#r +alu# +s; t&# sa!*l# siA# .ranging fr! 8 t 8BB0
$
u
m
ul
ati
e
@
1&
&
&
CALCULATION OF AVERAGES
10.0
P$)$m##) U(,#)$i(. C!)-#&
2!inal3 :ean (or the total population
21&&*ells3, Normalised to 1
1.
0
0 10 20 30 40 50 60 70 80 90 100
0.
1
A possible tra4e'tory o(
'omputed aerages (rom
drilled *ells )his e.ample
assumes that 2Ai3 outliers are
drilled early, and later *ells are
sampled in the lo* end o( the
ariability 'ure
R#l$i-# p$)$m##) -$l!# -&2 S$mpl# &i3# 45 + 5667
No Samples 50
No Samples 1 No
Samples 2 No
Samples 4 No
Samples 6 No
Samples 8 No
Samples 10 No
Samples 15 No
Samples 20 No
Samples 25 No
Samples 30 No
Samples 35 No
Samples 40 No
Samples 45 No
samples 100
0.0
Figur# C ) Pssi/l# tra6#ctr( f a+#rag#s +s; Para!#t#r Unc#rtaint( Cur+#s) s&2ing in r#" a *ssi/l#, alt&ug& nt +#r( li3#l(,
tra6#ctr( f **ulatin 2#ll a+#rag#s
illustrative well parameter PDF well permeability
istributi!" #urve $%e"eri# a" "!"&spe#i'# #ase(
100%
90% E.ample P=! 0 *ell L 2m=3
)his e.ample assumes a P1&KPI& range o( 1&
(old, *hi'h is not
80%
un'ommon in (ra'tured resour'e plays )his is the
8ARIABILITY 'ure
70%
C
u
m
Pr
!
b
a
bi
lit
y
60%
50%
40%
30%
10%
)ell parameter &permeability $mD(
0%
1.00 10.00 100.00 1000.00
20%
Figur# D ) 4aria/ilit( PDF cur+# fr 2#ll *ara!#t#r .*#r!#a/ilit(, !D0) t&is is a *ssi/l# illustratin nl( f a 2#ll +aria/ilit(, 2&ic&
2as us#" fr t&# c!*utatin f t&# Unc#rtaint( an" Cnfi"#nc# cur+#s
Figur# E ) Cnfi"#nc# Cur+#s) c&arting cnfi"#nc# fl"s .as a F c#rtaint(0 +s; F "ata a+aila/l#

S-ar putea să vă placă și