Professional Documents
Culture Documents
Calculation Method
C.R. Chia, SPE, W.J. Phillips, SPE, and D.L. Aklestad, SPE, Schlumberger
Summary sis by, for example, reducing the size of the geological target by
A method for combining multiple wellbore surveys to obtain a the size of the EOU to define a drilling target. In this fashion, the
single, composite, more accurate well position is described. Es- geological target will be achieved if the wellbore penetrates the
tablished methods for defining the wellbore position and its asso- drilling target. Likewise, EOUs are used to assess collision risk by
ciated uncertainty rely on accepting the position obtained from the considering to assess collision risk by considering their proximity
most accurate survey instrument used in each section of the well- to adjacent wells.
bore. This position is then assigned an uncertainty based on the
information from this single survey-instrument run. When a mod- Survey Program
ern wellbore is constructed today, each section may be surveyed A detailed survey program may be prepared to verify that a well’s
for position many times with one or more magnetic, gyroscopic, or position requirements will be achieved. The survey program is a
inertial survey instruments. By statistically combining the wellbore planned sequence of survey instruments to be used at different
positions obtained from all the survey instruments run in a given phases of the well construction. It will normally be presented as a
section of the wellbore, a new position, designated the “most ac- listing indicating the survey depths for each survey tool to be used,
curate position” (MAP), is calculated. The main advantage of the required survey frequency, running conditions (run in cased or
MAP is that its uncertainty is smaller than that of any of the open hole or in drillpipe), and any special corrections or contin-
constituent surveys. The major benefits of this technique is facil- gencies to validate the tool error model to be used for each sur-
itating drilling smaller targets at greater distances, allowing new veyed interval. The survey program takes into consideration the
wellbores to be drilled in closer proximity to existing wellbores available drilling room based on the proximity of existing nearby
while maintaining accepted safety clearance rules, and improving (object) wells and their respective EOUs, the expected EOU of the
reservoir delineation. (subject) well to be drilled based on the performance of the speci-
fied surveying tools, any spacing requirements for adjacent future
Describing Well Position wells, a collision-avoidance safety clearance rule, the size and
The wellbore trajectory is defined as a series of surveyed points in location of the geological target, and any relief-well-planning po-
3D space, typically described in a north, east, and down reference sitional-accuracy criteria.
system. These points are joined together to form a continuous
trajectory with a geometric calculation method.1 Most magnetic or Well Construction
gyroscopic survey instruments in use today provide a survey point Well construction is conducted in a number of drilling stages or
that is referenced to measured (or along hole) depth obtained from hole sections by drilling with decreasing drill-bit sizes, subse-
the driller’s pipe tally or a wireline spooling measurement. The quently cementing a steel casing or liner into place in each hole
survey instrument provides inclination (hole angle) and azimuth section. During this process, various survey instruments will be
(direction) measurements. When these parameters are used to cal- run in different hole sections (through drillpipe) and casings (on
culate trajectory with an assigned survey depth, the horizontal wireline) in accordance with the survey program to achieve well-
displacement (or north and east coordinates) and the vertical depth positioning objectives. It is not uncommon to have multiple sur-
(or down coordinate) can be derived from the origin and the el- veys in one or more hole sections. For the top section, for example,
evation reference, respectively. Alternatively, some inertial survey we may have a measurement-while-drilling (MWD) survey ob-
instruments measure displacement in 3D space from a known ini- tained during the drilling phase, a gyro survey obtained after reach-
tialization point, from which all the previous parameters, including ing the first casing point, and, perhaps, additional gyro surveys as
depth, can be obtained to achieve the same purpose. deeper hole sections are completed. The current practice is to
establish the final or “definitive” well position by using the most
Wellbore Position Uncertainty accurate survey in each hole section and disregarding the rest of
Wellbore survey requirements are typically driven by the need to the survey data.
guide the well to a geological target, to avoid other wells, to ensure
that property boundaries are respected, and to record the position Application of MAP
of the wellbore for future reference. To visualize and quantify our As the ability to drill directional wells has improved, making it
ability to hit a target or avoid colliding with another well, position possible to reach targets many kilometers away from the drillsite,
uncertainty is assigned to wellbore trajectories. This position un- and as the size of those targets has decreased, a need has developed
certainty represents our modeled knowledge of the collective er- to survey the position of the wellbore more accurately. Likewise,
rors arising from both the intrinsic performance limitations of the as fields mature and the density of wellbores in a given area
survey sensors and those induced by the operating environment.2 increases, the need to avoid these wells while drilling new ones has
This uncertainty is defined as a statistical confidence region with also created a desire to lower wellbore positional uncertainty. The
an associated confidence level. In 3D, the confidence region is MAP technique helps meet this need by using all available survey
most often depicted as an ellipsoid3 because ellipsoids are the data to calculate a composite survey with an EOU that is smaller
constant value contours of the 3D Guassian probability density than the EOUs of any of the individual surveys.
function. Such a confidence region is commonly referred to as an Berger et al.4 previously described a method of averaging mul-
“ellipsoid of uncertainty” (EOU). The EOU is used in target analy- tiple inclination measurements within a single bottomhole assem-
bly (BHA) to reduce the uncertainty in true vertical depth (TVD).
The MAP technique is a more general means of combining survey
Copyright © 2003 Society of Petroleum Engineers
data that reduces positional uncertainty in all directions and can be
This paper (SPE 85111) was revised for publication from paper SPE 77221, first presented used to combine surveys from instruments that use completely
at the 2002 IADC/SPE Asia Pacific Drilling Technology, Jakarta, 9−11 September. Original
manuscript received for review 7 October 2002. Revised manuscript received 14 April 2003.
different measurement methods. For example, it could be used to
Paper peer approved 22 April 2003. combine position coordinates calculated from a system that mea-
冤冥冤冥 冤 冥
Mathematical Basis of the MAP I1
The MAP is based on the generalized linear regression model: rជ2 I2 ␦rជ2
= rជt + , . . . . . . . . . . . . . . . . . . . . . . . . . . (9)
⯗ ⯗ ⯗
yជ = Xជ + ជ , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (1) rជk Ik ␦rជk
in which yជ⳱an m by one vector of observations; ជ ⳱a p by one in which each Ij is a 3 × 3 identity matrix and 1 ⱕ j ⱕ k.
vector of model parameters; X⳱an m by p matrix of regression The expression has the same form as Eq. 1 with:
variables, which establishes a linear relationship between the ob-
rជ1 ␦rជ1
冤冥 冤冥 冤冥
servations and the model parameters; and ជ ⳱an m by one vector I1
of random errors that characterizes the uncertainty in the observa- rជ2 I2 ␦rជ2
tions. Assuming ជ is zero mean and has a Gaussian probability yជ = ,X= , ជ = rជt, and ជ = . . . . . . . . . (10)
⯗ ⯗ ⯗
distribution, the probability density function for the random vari- rជk Ik ␦rជk
able yជ – Xជ is
冋 册
The covariance matrix, Cជ , may be written as:
1
exp − 共yជ − Xជ 兲T Cជ−1共yជ − Xជ 兲
冢 冣
2 具␦rជ1␦rជ1T 典 具␦rជ1␦rជ2T 典 … 具␦rជ1␦rជkT 典
p共yជ;ជ 兲 = , . . . . . . . . . . . . . . . (2) 具␦rជ2␦rជ1T 典 具␦rជ2␦rជ2T 典 … 具␦rជ2␦rជkT 典
共2兲m Ⲑ 2公det共Cជ兲 Cជ = 具ជ ជ T典 = . . . . . . (11)
⯗ ⯗ ··· ⯗
in which Cជ⳱the covariance matrix for the random vector ជ .
For a given set of observations, yជ, p(yជ;ជ ) is a function of the 具 k 1 典 具 k 2 典
␦rជ ␦rជ T
␦rជ ␦ជ
r T …
具 k␦rជkT 典
␦rជ
parameter ជ and is known as the likelihood function. An estimate Each term of the form 〈␦rជi␦rជTj 〉 is a 3 × 3 covariance matrix. The
of the unknown parameters can be obtained by finding the set diagonal terms (i.e., i⳱j) are simply the position covariance ma-
of parameters that maximizes the likelihood function. This tech- trices for each of the individual surveys and are, therefore, non-
nique is called the “method of maximum likelihood,” and the zero. The off-diagonal terms Cij (i.e., i ⫽ j), are zero if the position
resulting estimate, ជ̂ , of the unknown parameters is known as the measurements are independent.
“maximum likelihood estimate” (MLE).4 Maximization of Eq. 2 An estimate of the true position, rជ̂t, and its covariance matrix,
with respect to ជ yields the following MLE estimate, ជ̂ , and its Cˆrជt, can be computed directly from Eqs. 3 and 4. The MAP is the
covariance,Cˆជ . estimated true position, rជ̂t.
ˆ The error in each position measurement, rជ̂t , is the sum of many
ជ = 共XTCជ−1X兲−1XTCជ−1yជ, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (3) independent root sources of error, including sensor effects, like
bias and scale-factor drift, as well as environmental effects, like
and Cˆជ = 共XTCជ−1X兲−1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (4) misalignment of the surveying instruments with respect to the
borehole and drillstring magnetism. Consequently, the central limit
MLE estimates are optimal in the sense that they are unbiased theorem7 ensures that the statistical distribution of each ␦ rជ̂t will be
and have a lower variance than all other unbiased estimators.5 In approximately Gaussian and independent of the distributions of the
the event that ជ is not Gaussian distributed but remains zero mean, individual error sources, provided no single error source dominates
Eq. 3 can still be used to estimate ជ . The resulting estimate is no the error budget. In this case, as discussed previously, Eq. 3 yields
longer the MLE estimate, but it is guaranteed by the Gauss- the optimum (minimum variance and unbiased) estimate for the
Markov theorem6 to be unbiased and to have a lower variance than true position rជ̂t. Also keep in mind that even if the distributions for
all other unbiased estimators, which are linear in the observations, the ␦ជrs are not Guassian, Eq. 3 still yields an unbiased estimator
yជ. However, it may not have minimum variance when compared to with minimum variance within the class of estimators that are
all other unbiased estimators, including ones that are not linear in linear in the observations (i.e., r1, r2, ⭈ ⭈ ⭈ rជk).
the observations. While the MAP technique does not require knowledge of the
In the wellbore surveying application in which multiple mea- exact nature of the statistical distributions of the sources of posi-
surements of the same point in a wellbore have been recorded, tional error, it is necessary that all significant error sources are
the position vectors are combined to get a better estimate of enumerated and quantified (i.e., the covariance matrix Cជ must be
the true wellbore placement. The position of any point in the known). Specifically, the method breaks down if one or more of
wellbore is typically specified by its north, east, and down coor- the position measurements used to calculate the MAP estimate
dinates. Assume we have k measurements of the position of a contains a significant error that is either not accounted for in Cជ or
specific point in the wellbore. Each measurement can be written in is included in Cជ but the magnitude of which is much larger than
the following form. the typical value (three standard deviations or more, for example).
rជi = rជt + ␦rជi, . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (5) Examples might include measurements taken with survey instru-
ments that are out of calibration or data-entry errors, such as drop-
in which the position vector of the ith measurement is ping a minus sign when entering the grid convergence angle. Good
冋册
Furthermore, we must remember that the error sources were
nt assumed to be zero mean. Any error source that is nonzero mean
rជt = et , . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (7) will introduce a bias in the position measurement. To compensate
dt for this effect, we must subtract the bias from the affected position
measurement. A good example might be the vertical misalignment
and the error in the ith measurement is of the surveying tool with the borehole because of weight-induced
冋册
BHA bending (BHA sag). If a particular BHA consistently points
␦ni the survey tool below the centerline of the borehole, it will result
␦rជi = ␦ei . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (8) in a positive bias in the vertical component of the calculated po-
␦di sition (i.e., the calculated position will be too deep). Consequently,
rជ1 = 冋 册 冋 册 冋 册
5000
4950
5050
rជ2 =
5050
5000
4950
rជ3 =
4950
5050 . . . . . . . . . . . . . (12)
5000 Fig. 1—Theoretical example showing how three 10 × 10 × 1
Furthermore, suppose these measured positions have the fol- ellipsoids are combined with the MAP technique to produce a
lowing covariance matrices. substantially more accurate position determination.
C11 = 具␦rជ1␦rជ1 典 = T
冉 1 0 0
0 100 0
冊 rជ
On the other hand, if we use Eqs. 3 and 4 to calculate rជt and C
, we get the following results.
冋 册
t
0 0 100
冉 冊
5000
100 0 0 r̂ជ = 5000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (17)
C22 = 具␦rជ2␦rជ2T 典 = 0 1 0 , . . . . . . . . . . . . . . . . . . . . (13) 5000
冋 册
0 0 100
冉 冊
0.98 0 0
100 0 0 C rជ̂ = 0 0.98 0 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . (18)
C33 = 具␦rជ3␦rជ3T 典 = 0 100 0 0 0 0.98
0 0 1
Plotting the 20% confidence EOU corresponding to Crជt in Fig. 2,
and that the terms of the form Cij are zero if i ⫽j. we see that rជt provides good estimates of all three coordinates
Each matrix in Eq. 13 defines a 3D Gaussian distribution with (north, east, and down). When we compare the EOU of rជt to the
a probability density function in the following form. EOUs of r1, r2, and r3, the advantage of statistically combining
冋 册
multiple surveys is clear.
1 Another special case for which the result can be easily com-
exp − 共 rជi − rជt兲T Cii −1共 rជi − rជt兲
2 puted is that of multiple independent surveys, each with a covari-
p共 rជi兲 = . . . . . . . . . . . . . . . . . (14)
共2兲3 Ⲑ 2 公det共Cii兲
ance matrix Crជ. In this case, the all the 3 × 3 diagonal blocks of
Cឈ are equal (i.e., C11, C22, . . . Ckk⳱Crជ). The resulting MAP
Ⲑ
k
Because the covariance matrices, Cii, are diagonal, the prob- estimate and its covariance are 兺 rជi Ⲑ k and Cˆ =Crជ k,respectively.
ability density function reduces to: i=1
The MAP estimate in this case is the simple average of the individual
p共 rជi兲 =
再 冋
exp −
1 共 xi − xt兲2 共yi − yt兲2 共zi − zt兲2
2 Ciixx
+
Ciiyy
+
Ciizz
册冎 . . . . . . (15)
position measurements, and the size of MAP EOU is smaller than
those of individual position measurements by a factor of (1/√k).
Combining just two surveys in this way reduces the EOU by a factor
共2兲3 Ⲑ 2 公det共Cii兲 of (1/√2), approximately a 30% reduction.
The constant value contours of Eq. 15 are a family of ellipsoids Practical Example of Results
defined by equating the quadratic expression in the exponent to
In practice, we normally get neither wellbore surveys with EOUs
a constant.
that differ so dramatically in shape, as in the first example, nor
共xi − xt兲2 共yi − yt兲2 共zi − zt兲2 ones that have equal size, as in the second. More commonly, we
s2 = + + . . . . . . . . . . . . . . . . . . . . . . (16) have a set of EOUs of similar shape but perhaps with differing
Ciixx Ciiyy Ciizz sizes, each of which represents the precision of the respective
For each ellipsoid, the length of the north, east, and down semi- surveying instrument and its operating environment.
major axes are: s⭈公Ciixx, s⭈公Ciiyy, and s⭈公Ciizz, respectively. For example, a typical survey program for a planned directional
well is given in Table 1.
As discussed previously, these ellipsoids, or EOUs, also rep- Fig. 3 shows a schematic of the well position and the EOU for
resent the confidence regions of the distributions and are an ex- each part of the survey program plotted in plan view, which figu-
cellent means of visualizing the distribution. The ellipsoid defined ratively provides an indication of the scale of the EOU at the
by setting s equal to one, for example, is an approximately 20%
confidence region.3 This means that there is a 20% probability that
the true position, rជt, is somewhere within the volume enclosed by
this ellipsoid. The s⳱1, or 20% confidence level, EOUs corre-
sponding to C11, C22, and C33 are shown in Fig. 1.
Fig. 1 makes the difference between the three measurements of
rជt quite obvious. Each measurement gives a good estimate of one
of the coordinates and poor estimates of the other two. The first
measurement, for example, gives a good estimate of the north
coordinate but poor estimates of the east and down coordinates,
while the second measurement gives a good estimate of the east
coordinate but poor estimates of north and down; similarly, the
third measurement provides a good estimate of the down coordi-
nate. It is also clear that if we simply choose one of the three Fig. 2—Theoretical example centroid expanded, showing the
measurements as the definitive estimate of rជt, we will not make approximate resultant position uncertainty as a 1 × 1 × 1 sphere
optimal use of all the available data. and the MAP location.
Fig. 5—The MAP and a reduced EOU are once again obtained
Fig. 3—Schematic plan view of an example survey program for from the combination of the intermediate (9.625-in.) casing gyro
a planned directional well depicting figurative ellipses of uncer- survey and the 12.25-in. hole section MWD surveys, both of
tainty. Multiple surveyed sections of the well make use of the which had been tied onto and had continued from the position
most accurate survey instrument run for each overlapped section. of MAP Part 1 to obtain Part 2.