Accuracy Prediction for Directional

Measurement While Drilling

H.S. Williamson, SPE, BP

Summary areas of ambiguity or confusion. In such cases, reproduction of the

In this paper a new method for predicting wellbore position un- numerical results at the end of the paper will act as a powerful
certainty which responds to the current needs of the industry is criterion for validation.
described. An error model applicable to a basic directional mea-
surement while drilling MWD service is presented and used for Genesis of the Work. The content of this paper is the fruit of two
illustration. As far as possible within the limitations of space, the collaborative groups.
paper is a self-contained reference work, including all the neces- ISCWSA. The Industry Steering Committee on Wellbore Sur-
sary information to develop and test a software implementation of vey Accuracy is an informally constituted group of companies and
the method. The paper is the product of a collaboration between individuals established following the SPWLA Topical Conference
the many companies and individuals cited in the text. on MWD held in Kerrville, Texas in late 1995. The groups broad
objective is to produce and maintain standards for the industry
relating to wellbore survey accuracy. Much of the content of this
Introduction paper, and specifically the details of the basic MWD error model,
As the industry continues to drill in mature oil provinces, the dual had its genesis in the groups meetings, which were distinguished
challenges of small geological targets and severe well congestion by their open and cooperative discussions.
increase the importance of quantifying typical wellbore positional Four Company Working Group. The ISCWSA being too large
errors. The pioneering work of the 1970s culminated in the paper a forum to undertake the detailed mathematical development of an
by Wolff and de Wardt.1 Their approach, albeit extensively modi- error propagation model, this was completed by a small working
fied and added to, has remained the de facto industry standard to group from Sysdrill Ltd., Statoil, Baker Hughes INTEQ, and BP
this day. At the same time, various shortcomings of the method Exploration. The mathematical model created by the group and
have been identified,2-4 but are not discussed further here. described below has been made freely available for use by the
In recent years, a number of factors have created the opportu- industry.
nity for the industry to develop an alternative method:
risk-based approaches to collision avoidance and target hit-

ting require position uncertainties with associated confidence lev-

Assumptions and Definitions
els, something which Wolff and de Wardt specifically avoided;
changing relationships brought about by integrated service The following assumptions are implicit in the error models and
contracts have forced directional drilling and survey companies to mathematics presented in this paper.
Errors in calculated well position are caused exclusively by
share information on tool performance;
the development of several new directional software products the presence of measurement errors at wellbore survey stations.
Wellbore survey stations are, or can be modeled as, three-
and their integration with subsurface applications has provided the
necessity and the opportunity to develop new means of commu- element measurement vectors, the elements being along-hole
nicating and visualizing positional uncertainty. depth, D, inclination, I, and azimuth, A. The propagation math-
ematics also requires a toolface angle, , at each station.
This paper provides a three-part response to this need. Errors from different error sources are statistically indepen-

dent.
1. Error Model for Basic MWD. This is based on the current There is a linear relationship between the size of each mea-
state of knowledge of a group of industry experts. There are sev-
surement error and the corresponding change in calculated well
eral reasons why directional MWD is the most suitable survey
position.
service to illustrate a new method of error modeling. The error The combined effect on calculated well position of any num-
budget is dominated by environmental effects, so that accuracy
ber of measurement errors at any number of survey stations is
differences attributable to tools alone are minimal. It is the survey
equal to the vector sum of their individual effects.
tool of choice for most directional wells, where position uncer-
tainty is of greatest concern. The physical principles of its opera- No restrictive assumptions are made about the statistical distri-
tion, including the navigation equations, are in the public domain. bution of measurement errors.

2. Mathematical Basis. This is a rigorous description of the Error Sources, Terms and Models. An error source is a physical
propagation of errors in stationary tools. Fit-for-purpose error phenomenon which contributes to the error in a survey tool mea-
models using the same basis are in development for inertial and surement. An error term describes the effect of an error source on
continuous gyroscopic tools, although some simplification and a particular survey tool measurement. It is uniquely specified by
compromise are inevitable. A rigorous treatment of continuous the following data:
a name;
survey tools would probably have too restricted a cognoscenti to
a weighting function, which describes the effect of the error
be practical.
on the survey tool measurement vector p. Each function is re-
3. Standard Examples and Results. Despite the apparent sim- ferred to by a mnemonic of up to four letters.
A mean value, .
plicity of the Wolff and de Wardt method, different software
A magnitude, , always quoted as a 1 standard deviation
implementations generally give subtly different results. While an
effort has been made in this paper to provide a comprehensive value.
A correlation coefficient 1 between error values at survey
description of the new method, there will surely remain some
stations in the same survey leg. In a survey listing made up of
several concatenated surveys, a survey leg is a set of contiguous
Copyright 2000 Society of Petroleum Engineers survey stations acquired with a single tool or, if appropriate, a
This paper (SPE 67616) was revised for publication from SPE 56702 first presented at the single tool type.
A correlation coefficient 2 between error values at survey
1999 SPE Annual Technical Conference and Exhibition held in Houston, Texas, 36 Oc-
tober. Original manuscript received for review 31 January 2000. Revised manuscript re-
ceived 26 July 2000. Paper peer approved 14 August 2000. stations in different survey legs in the same well.

SPE Drill. & Completion 15 4, December 2000 1064-6671/2000/154/221/13/$5.000.50 221

 A correlation coefficient 3 between error values at survey that the shifts between successive calibrations are representative
stations in different wells in the same field. of the shifts between calibration and field performance. On this
To ensure that the correlation coefficients are well defined, only basis, two major MWD suppliers compared the results of succes-
four combinations are allowed. sive scheduled calibrations of their tools. Paul Rodney examined
Propagation Mode 1 2 3 288 pairs of calibrations, and noted the change in bias i.e., offset
Random R 0 0 0 error, scale factor, and misalignment for each sensor. Wayne
Systematic S 1 0 0 Phillips did the same for 10 pairs of calibrations, except that sen-
Well by well W 1 1 0 sor misalignments were not recorded.
Global G 1 1 1 Andy Brooks has demonstrated that if a sensor is subject to a
1, 2, and 3 are to be considered properties of the error source, scale error and two orthogonal misalignments, all independent and
and should be the same for all survey legs. of similar magnitude, the combination of the three error terms is
An error model is a set of error terms chosen with the aim of equivalent to a single bias term. This term need not appear explic-
properly accounting for all the significant error sources which itly in the error model, but may be added to the existing bias term
affect a survey tool or service. to create a lumped error. This eliminates the need for 20 extra
weighting functions corresponding to sensor misalignments.
An Error Model for Basic MWD The data from the MWD suppliers suggest that in-service sen-
For the survey specialist in search of a best estimate of posi- sor misalignments are typically smaller than scale errors. As a
tion uncertainty it is tempting to differentiate minutely among result, only a part of the observed scale error was lumped with
tools types and models, running configurations, bottomhole as- the misalignments into the bias term, leaving a residual scale error
sembly BHA design, geographical location and several other which is modeled separately. In this way, four physical errors for
variables. While justifiable on technical grounds, such an ap- each sensor were transformed into two modeled terms. The results
proach is impractical for the daily work of the well planner. The were as follows.
time needed to find out these data for historical wells, and for Weighting Propagation
many planned wells, is simply not available. Error Source Function Magnitude Mode
The error model presented in this section is intended to be Accelerometer biases ABX,Y,Z 0.004 ms2 S
representative of MWD surveys run according to fairly standard Accelerometer scale
quality procedures. Such procedures would include rigorous and factors ASX,Y,Z 0.005 S
regular tool calibration, survey interval no greater than 100 ft,
Magnetometer biases MBX,Y,Z 70 nT S
nonmagnetic spacing according to standard charts where no axial
interference correction is applied, not surveying in close proxim- Magnetometer scale
ity to existing casing strings or other steel bodies, and passing factors MSX,Y,Z 0.0016 S
standard field checks on the G total, B total, and dip. These figures include errors which are correlated between sen-
The requirement to differentiate between different services may sors, and which therefore have no effect on calculated inclination
be met by defining a small suite of alternative error models. Ex- and azimuth the exception being the effect of correlated magne-
amples covered in this paper are application or not of an axial
tometer errors on interference corrected azimuths. It could be
interference correction and application or not of a BHA sag cor-
argued that the magnetometer scale factor errors in particular
rection.
Alternative models would also be justified for in-field refer- which may be influenced by crustal anomalies at the calibration
enced surveys, in-hole gyro referenced surveys, and depth- sites should be reduced to account for this.
corrected surveys.
The model presented here is based on the current state of BHA Magnetic Interference. Magnetic interference due to steel
knowledge and experience of a number of experts. It is a starting in the BHA may be split into components acting parallel axial
point for further research and debate, not an endpoint. and perpendicular cross axial to the borehole axis.
Axial Interference. Several independent sets of surface mea-
Sensor Errors. MWD sensors will typically show small shifts in surements of magnetic pole strengths have now been made. Ob-
performance between calibrations. We may make the assumption served root-mean-square RMS values are the following.

Pin Box
Item RMS Pole Strength Sample Size Source
Drill collar 505 Wb 8 Grindrod and Wolff Ref. 5
605 Wb 11 435 WB 11 Lotsberg Ref. 5
511 Wb 4 McElhinney*
Stabilizer 177 Wb 6 Grindrod and Wolff
396 Wb 10 189 Wb 10 Lotsberg
369 Wb 5 408 Wb 10 McElhinney
Motor 340 Wb 12 419 Wb 10 Lotsberg

Oddvar Lotsberg also computed pole strengths for 41 BHAs value for the nonmagnetic spacing distance. Unfortunately, there
from the results of an azimuth correction algorithm. The RMS is no typical spacing used in the industry, and we must find
pole strength was 369 Wb micro-Webers. another way to estimate the magnitude of this error source.
These results suggest that 400 Wb is a reasonable estimate for A well-established industry practice is to require nonmagnetic
the 1 s.d. pole strength of a steel drillstring component when spacing sufficient to keep the azimuth error below a fixed toler-
further information is lacking. This is useful information for BHA ance typically 0.5 at 1 s.d. for assumed pole strengths and a
design, but cannot be used for uncertainty prediction without a given hole direction. This tolerance may need to be compromised
in the least favorable hole directions. For a fixed axial interference
field, and neglecting induced magnetism, the azimuth error is
*Minutes of the 7th Meeting of the ISCWSA, Houston, 9 October 1997. strongly dependent on hole direction, being proportional to

222 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
sin I sin Am . Thus to model the azimuth error in uncorrected sur- gauge. Data comparisons by the author suggest a typical effi-
veys, we require a combination of error terms which predicts zero ciency of 60% for these corrections, leaving a post-correction re-
error if the well is vertical or magnetic north/south, predicts errors sidual sag error of 0.08.
somewhat greater than the usual tolerance if the well is near hori- Assuming similar BHAs throughout a hole section, all BHA
zontal and magnetic east/west or predicts errors near the usual sag errors may be classified as systematic.
tolerance for other hole directions. Radially Symmetric Misalignment. This is modeled as equally
These requirements could be met by constructing some artifi- likely to be oriented at any toolface angle. John Turvill made an
cial weighting function, but this would violate our restriction to estimate of its magnitude based on the tolerances on several con-
physically meaningful error terms. A constant error of 0.25 and a centric cylinders.
direction-dependent error of 0.6 sin I sin Am is perhaps the best Sensor package in the housing. Tolerances on three compo-

we can achieve by way of a compromise. It is legitimate to con- nents are clearance, 0.023, concentricity, 0.003, and straightness
sider these values representative of 1 standard deviation, since the of sensor package, 0.031.
pole strength values which underlie the nonmagnetic spacing cal- Sensor housing in the drill collar. For a probe mounted in a

culations are themselves quoted at 1 s.d. centralized, retrievable case, 0.063.

Both error terms may be propagated as systematic, although Collar bore in the collar body. Typical MWD vendors toler-

there is theoretical and observational evidence4 that this error is ance is 0.05.
asymmetric, acting in the majority of cases to swing magnetic Collar body in the borehole. The API tolerance on collar

surveys to the north in the northern hemisphere. Giving the straightness equates to 0.03. MWD vendors specifications are
direction-dependent term a mean value of 0.33 and a magnitude typically somewhat more stringent.
of 0.5 reproduces this asymmetry with about 75% of surveys The root sum square of these figures is 0.094. Being based

being deflected to the north, while leaving the root-mean-square on maximum tolerances, it is probably an overestimate for stabi-
error unchanged. lized rotary assemblies.
Axial interference errors are not modeled for surveys which An analysis by the author of the variation in measured incli-

have been corrected for magnetic interference. nation over 46 rotation shots produced a root-mean-square mis-
Cross-Axial Interference. Cross-axial interference from the alignment of 0.046. Simulations show that within this figure,
BHA is indistinguishable from magnetometer bias, and propagates about 0.007 is attributable to the effect of sensor errors.
in the same way. Anne Holmes** analyzed the magnetometer An additional source of misalignment, collar distortion out-

biases for 78 MWD surveys determined as a by-product of a mul- side the vertical plane due to bending forces, may be estimated
tistation correction algorithm. Once a few outliers, probably due using three-dimensional BHA models and 0.04 seems to be a
to magnetic hot spots and hence classified as gross errors, had typical value. This error differs from those above by not rotating
been eliminated, the remaining observations gave a RMS value of with the tool. It should therefore strictly have its own weighting
57 nT. This figure is somewhat smaller than the 70 nT attributable function. Being so small, it seems justifiable on practical if not
to magnetometer bias alone. The conclusion must be that cross- theoretical grounds, to include it with the other sources of radi-
axial interference does not, on average, make a significant contri- ally symmetric misalignment. This leaves us with an estimate for
bution to the overall MWD error budget, and may be safely left the error magnitude of 0.06. This figure may be a significant
out of the model. underestimate where there is an aggressive bend in the BHA or a
probe-type MWD tool is in use. This error term may be consid-
Tool Misalignment. Misalignment is the error caused by the
ered systematic.
along-hole axis of the directional sensor assembly being out of
parallel with the center line of the borehole. The error may be Magnetic Field Uncertainty. For basic MWD surveys, only the
modeled as a combination of two independent phenomena: value assumed for magnetic declination affects the computed azi-
BHA Sag. This is due to the distortion of the MWD drill collar muth. However, conventional corrections for axial interference
under gravity. It is modeled as confined to the vertical plane and require estimates of the magnetic dip and field strength. Any error
proportional to the component of gravity acting perpendicular to in these estimates will cause an error in the computed azimuth.
the wellbore i.e., sin I. The magnitude of the error depends on
A study by the British Geological Survey and commissioned by
BHA type and geometry, sensor spacing, hole size and several
Baker Hughes INTEQ6 investigated the likely error in using a
other factors. Two-dimensional BHA models typically calculate
global geomagnetic model to estimate the instantaneous ambient
inclination corrections of 0.2 or 0.3 for poorly stabilized BHAs
magnetic field downhole. Five sources of error were identified:
in horizontal hole.6 For well stabilized assemblies the value is
modeled main field vs. actual main field at the base epoch, mod-
usually less than 0.15. In the absence of better information, 0.2
eled secular variation vs. actual secular variation, regular diurnal
at 1 s.d. may be considered a realistic input into the basic error
model. variation due to electrical currents in the ionosphere, irregular
Sag corrections, if they are applied, are calculated on the often temporal variation due to electrical currents in the magnetosphere
unjustified assumptions of both the hole and stabilizers being in and crustal anomalies.
By making a number of gross assumptions, and by considering
typical drilling rates, this author has distilled the results of the
**Minutes of the 8th Meeting of the ISCWSA, Trondheim, 19 February 1998. study into a single table.

Error Magnitude
Declination Dip Total Field Propagation
Error Source deg deg nT Mode
Main field model 0.012* 0.005 3 G
Secular variation 0.017* 0.013 10 G
Daily variation 0.045** 0.011** 11** R/S
Irregular variation 0.110** 0.043** 45** R/S
Crustal anomaly 0.476 0.195 120 G

*Below 60 latitude N or S.
**At 60 latitude N or S.

Daily and irregular variation are partially randomized between surveys. Correlations between consecutive stations are approximately 0.95 and 0.5 for the two error sources.

H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000 223
 The dominant error source is crustal anomalies caused by vary- errors may be considered globally systematic and summarized as
ing magnetization of rocks in the Earths crust. The figures shown the following.
are representative of those of the North Sea. Some areas, particu- Weighting Propagation
larly those at higher latitudes and where volcanic rocks are closer Error Source Function Magnitude Mode
to the surface, will show greater variation. Other areas, where Declination constant AZ 0.36 G
sedimentary rocks dominate, will show less. Declination
In the absence of any other information, the uncertainty in an B H dependent DBH 5000 nT G
estimate of the magnetic field at a given time and place provided Dip angle MFD 0.20 G
by a global geomagnetic model may be obtained by summing the Total field MFI 130 nT G
above terms statistically. There is one complication: some account
must be taken of the increasing difficulty of determining declina-
Along-Hole Depth Errors. Ekseth7 identified 14 physical sources
tion as the horizontal component of the magnetic field decreases.
This can be achieved by splitting this error into two components: of drill-pipe depth measurement error, wrote down expressions to
one constant and one inversely proportional to the horizontal pro- predict their magnitude and, by substituting typical parameter val-
jection of the field, B H . For the purposes of the model, the split ues into the expressions, predicted the total error for a number of
has been defined somewhat arbitrarily, while ensuring that the different well shapes. He then proposed a simplified model of just
total declination uncertainty at Lerwick, Shetland (B H four terms, and chose the magnitudes of each to match the pre-
15 000 nT) is as predicted by the BGS study 0.49. Being dictions of the full model as closely as possible. The results were
dominated by the crustal anomaly component, all magnetic field as follows.

Error Error Magnitude 1 s.d

Proportional Propagation
Error Source to Land Rig Floating Rig Mode
Random reference 1 0.35 m 2.2 m R
Systematic reference 1 0m 1m S
Scale D 2.4104 2.1104 S
Stretch type D.D v 2.2107 m1 1.5107 m1 G

For the purposes of the basic model, the values for the land rig Gross Errors. Any attempt at a comprehensive discussion of
or, equivalently a jackup or platform rig may be chosen. The MWD error sources must at least acknowledge the possibility of
stretch-type error, which dominates the other terms in deep wells, gross errors, sometimes called human errors. These errors lack the
models two physical effects, stretch and thermal expansion of the predictability and uniformity of the physical terms discussed
drill pipe. Both of these effects generally cause the drill string to above. They are therefore excluded from the error model, with the
elongate, so it may be appropriate to apply this term as a bias see assumption that they are adequately managed through process and
below. If this is done, a mean value of 4.4107 m1 should be procedure.
used, since Ekseth effectively treated his estimates of these errors
as 2 s.d. values.
Propagation Mathematics
Errors Omitted From the Basic MWD Model. Some errors The mathematical algorithm by which wellbore positional uncer-
known to affect MWD surveys have nonetheless not been in- tainty is generated from survey error model inputs is based on the
cluded in the basic error model. approach outlined by Brooks and Wilson.3 The development of
Tool Electronics and Resolution. The overall effect on accu- this work described here was carried out by the working group
racy caused by the limitations of the tool electronics and the reso- referred to in the Introduction.
lution of the tool-to-surface telemetry system is not considered A physical error occurring at a survey station will result in an
significant. Such errors will tend to be randomized over long sur- error, in the form of a vector, in the calculated well position. From
vey intervals. Ref. 3:
External Magnetic Interference. Ekseth7 discussed the influ-
dr p
ence of remanent magnetism in casing strings on magnetic sur- ei i , 1
veys, and gave expressions for azimuth error when drilling out of dp i
a casing shoe and parallel to an existing string. Although certainly where ei is a vector-valued random variable a vector error, i is
not negligible, both error sources are difficult to quantify, and the magnitude of the ith error source, p/ i is its weighting
equally difficult to incorporate within error modeling software. It function and dr/dp describes how changes in the measurement
seems preferable to manage these errors by applying quality pro- vector affect the calculated well position. It is sufficient to assume
cedures designed to limit their effect. that the calculated displacement between consecutive survey sta-
Effect of Survey Interval and Calculation Method. The tions depends only on the survey measurement vectors at these
method presented in this paper relies on the assumption that an two stations. Writing rk for the displacement between survey
error-free measurement vector p will lead to an error-free well- stations k1 and k, we may thus express the 1 s.d. error due to
bore position vector r. If minimum curvature formulas are used the presence of the ith error source at the kth survey station in the
for survey calculation, this assumption will only be true when the lth survey leg as the sum of the effects on the preceding and
well path between stations is an exact circular arc. The resulting following calculated displacements:
error may be significant for sparse data, but may probably be
neglected so long as the station interval does not exceed 100 ft.
Gravity Field Uncertainty. Differences between nominal and
ei,l,k i,l dpk dpk
drk drk1 pk

i
, 2

actual gravity field strengths will typically have no effect on where i,l is the magnitude of the ith error source over the lth
MWD accuracy since only the ratio of accelerometer measure- survey leg, and pk is the instrument measurement vector at the kth
ments is used in the calculation of inclination and azimuth. survey station.

224 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
 The total position error at a particular survey station K in sur-
vey leg L will be the sum of the vector errors ei,l,k taken over all
error sources i and all survey stations up to and including K. The
uncertainty in this position error is expressed in the form of a
covariance matrix:

CK
errors k 1 K k 2 K
i,l 1 ,k 1 , i,l 2 ,k 2 ei,l 1 ,k 1 ei,l
T
2 ,k 2
, 3
i

where ( i,l 1 ,k 1 , i,l 2 ,k 2 ) is the correlation coefficient between the

value of the ith error source at the k 1 th station in the l 1 th leg and
the k 2 th station in the l 2 th leg. In practice, it is more convenient
to sum separately the contributions of errors with different propa-
gation characteristics. Details are in Appendix A.

Weighting Functions. The weighting function for a particular er-

ror source is a 31 vector, the elements of which describe the Fig. 1Definition of tool sensor axes and toolface angle.
effect of a unit error on the measured along-hole depth, inclination
and azimuth. For example, the weighting functions for constant
and B H -dependent magnetic declination errors are

B x B cos cos I cos A m sin B sin sin I sin
0
p B cos sin A m cos , 16
0 , 4
AZ
1 B y B cos cos I cos A m cos B sin sin I cos

p
DBH
0
0
1/ B cos
. 5
B cos sin A m sin ,
B z B cos sin I cos A m B sin cos I.
Taking the X-accelerometer bias (ABX) as an example,
17
18

For BHA sag and direction-dependent axial magnetic interfer-

ence they are I 1 cos I 1 G xG z

cos I sin

, 19
0 G x sin I G x sin I G3 G
p
sin I , 6 and, similarly,
SAG
0
A m cos I sin A m sin cos A m cos tan cot I cos

.
p
0 Gx G
0 , 7 20
AM ID
sin I sin A m The appropriate weighting function is therefore
and for reference, scale and stretch-type depth errors they are p
p
DREF
1

0 ,
0
8
ABX

1
0
cos I sin

.
G
D cos I sin A m sin cos A m cos tan cot I cos
p
0 , 9 21
DSF
0 Effect of Axial Interference Correction. When a simple axial

magnetic interference correction is applied, Eq. 12 is no longer
D.D V
p used, and different weighting functions are required for sensor
0 . 10 errors. The following analysis is by Andy Brooks.
DST
0 Details of the interference corrections differ from method to
Weighting Functions For Sensor Errors. Tool axes and tool- method, but since all such methods suffer from similar limitations,
face angle, , are defined in Fig. 1. There are 12 basic sensor error it is reasonable to characterize them all with a single example.
sources a bias and scale factor for each of three accelerometers Methods which ignore the B Z measurement and find the solution
and three magnetometers and each requires its own weighting which minimizes the vector distance between the computed and
function. These are obtained by differentiating the standard navi- expected values of the magnetic field vector will satisfy Eqs. 16
gation equations for inclination and azimuth: and 17 and

B cos B cos
2 B sin B sin
2 minimum, 22
Gz
1
Icos , 11 where B and are the estimated values of total field strength and
G 2x G 2y G z2 dip angle, respectively. Solving these three equations for azimuth

A m tan1 G x B y G y B x G 2x G 2y G z2
B z G 2x G 2y G z G x B x G y B y
, 12
leads to
P sin A m Q cos A m R sin A m cos A m 0, 23

and making use of the inverse relations where

G x G sin I sin , 13 P B x sin B y cos cos IB sin

sin I cos I, 24

G y G sin I cos , 14 Q B x cos B y sin , 25

G z G cos I, 15 RB cos
sin2 I. 26

H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000 225
 TABLE 1 ERROR SOURCE WEIGHTING FUNCTIONS NOT GIVEN IN THE TEXT

Sensor Errors (without axial interference correction)

ABX
1
G 0
cos I sin
cos I sin Am sin cos Am cos tan cot I cos
ASX 0
sin I cos I sin2
tan sin Icos I sin Am sin cos Am cos cos I cos sin

ABY
1
G 0
cos I cos
cos I sin Am cos cos Am sin tan cot I sin
ASY 0
sin I cos I cos2
tan sin Icos I sin Am cos cos Am sin cos I sin cos

MBX 0
0
cos Am cos cos I sin Am sin /B cos

MSX 0
0
cos I cos Am sin tan sin I sin sin Am cos cos Am cos cos I sin Am sin

MBY 0
0
cos Am sin cos I sin Am cos /B cos

MSY 0
0
cos I cos Am cos tan sin I cos sin Am sin cos Am sin cos I sin Am cos

ABZ
1
G 0
sin I
tan sin Isin Am
ASZ 0
sin I cos I
tan sin I cos I sin Am
MBZ 0
0
sin I sin Am /B cos

MSZ 0
0
sin I cos Amtan cos Isin I sin Am

Sensor errors (with axial interference correction)

ABIX
1
G 0
cos I sin
cos2 I sin Am sin tan cos Isin I cos Amcos tan cos Amcot I/1sin2 I sin2 Am

ABIY
1
G 0
cos I cos
cos2 I sin Am cos tan cos Isin I cos Amsin tan cos Amcot I/1sin2 I sin2 Am

ASIX 0
sin I cos I sin2
sin sin I cos2 I sin Am sin tan cos Isin I cos Amcos tan sin I cos Amcos I/1sin2 I sin2 Am

ASIY 0
sin I cos I cos2
cos sin I cos2 I sin Am cos tan cos Isin I cos Amsin tan sin I cos Amcos I/1sin2 I sin2 Am

MSIX 0
0
cos I cos Am sin tan sin I sin sin Am cos cos I sin Am sin cos Am cos /1sin2 I sin2 Am

MSIY 0
0
cos I cos Am cos tan sin I cos sin Am sin cos I sin Am cos cos Am sin /1sin2 I sin2 Am

MBIX 0
0
cos I sin Am sin cos Am cos /B cos 1sin2 I sin2 Am

ABIZ
1
G 0
sin I
sin I cos I sin Amtan cos Isin I cos Am/1sin2 I sin2 Am

MBIY 0
0
cos I sin Am cos cos Am sin /B cos 1sin2 I sin2 Am

ASIZ 0
sin I cos I
sin I cos2 I sin Amtan cos Isin I cos Am/1sin2 I sin2 Am

Magnetic field errors (with axial interference correction)

MFI 0
0
sin I sin Amtan cos Isin I cos Am/BIsin2 I sin2 Am
MDI
0
0
sin I sin Amcos Itan sin I cos Am/1sin2 I sin2 Am

226 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
 Fig. 3Vertical section plot of standard well profiles. Note the
different section azimuths.
Fig. 2Plan view of standard well profiles.

The sensitivities of the computed azimuth to errors in the sensor

Survey Bias. Not to be confused with sensor biases which might
measurements are found by differentiating Eq. 23.
better be termed offset errors, survey bias is the tendency for the
Magnetic Field Uncertainty. The weighting function for mag-
most likely position of a well to differ from its surveyed position.
netic declination error is given above. Those for magnetic field
The only bias term defined by Wolff and de Wardt was for mag-
strength and dip angle, which are required when an axial magnetic
netic interference in poor magnetic surveys. The claims for
interference correction is in use, are derived by differentiating Eq.
stretch and thermal expansion of the drill pipe to be treated as bias
23 with respect to B and .
errors are at least as strong.
Misalignment Errors. Brooks and Wilson3 model tool axial Some vendors of directional software have neglected to model
misalignment as two uncorrelated errors corresponding to the x survey bias on the grounds that such errors should be corrected
and y axes of the tool. Their expressions for the associated incli- for, and that engineers do not like/understand them. The first ob-
nation and azimuth errors lead directly to the following weighting jection can be countered by the observation, yes, but they are
functions:

not, the second by careful software design.
0 The sign convention for position bias is from survey to most
p likely position i.e., opposite to the direction of the error. Since
sin , 27
MX the drill pipe generally elongates downhole, most likely depths are
cos /sin I

greater than survey depths and bias values are positive. For axial
0 drillstring interference, most likely azimuths are greater than sur-
p vey azimuths when the weighting function, sin I sin Am , is posi-
cos . 28
MY tive, so bias values are again positive at least in the northern
sin /sin I hemisphere. The additional mathematics required to model sur-
Table 1 contains expressions for all the weighting functions not vey bias is included in Appendix A.
cited in this section which are required to implement the error
models described in this paper. Calculation Conventions. The calculation of position uncertainty
requires a wellbore survey consisting of discrete stations, each of
Calculation Options which has an associated along-hole depth, inclination, azimuth,
The method of position uncertainty calculation described here ad- and toolface angle. Clearly, these data will not be available in
mits a number of variations. It can still claim to be a standard, in many cases, and certain conventions are required whereby as-
that selection of the same set of conventions should always yield sumed values may be calculated. The following are suggested.
the same results. Along-Hole Depth. For drilled wells, actual survey stations
should be used. For planned wells, the intended survey interval
Along-Hole Depth Uncertainty. The propagation model de- should be determined, and stations should be interpolated at all
scribed above is appropriate for determining the position uncer- whole multiples of this depth within the survey interval. Typi-
tainty of the points in space at which the survey tool came or will cally, an interval of 100 ft or 30 m should be used. For well plans,
come to rest. These may be called uncertainties at survey sta- the way points should be included as additional stations.
tions. Inclination and Azimuth. For drilled wells, measured values
Thorogood2 argued that it is more meaningful to compute the should be used. For planned wells, the profile should be interpo-
position uncertainties of the points in the wellbore at the along- lated at the planned survey station depths using minimum curva-
hole depths assigned to the survey stations. These may be called ture.
the uncertainties at assigned depths. This approach allows Toolface. If actual toolface angles are available, they should be
computation of the position uncertainty of points such as picks used. If not, several means of generating them are possible.
from a wireline log whose depths have been determined indepen- Random number generation. Possibly close to reality, but re-

dently of the survey. Thorogood made this calculation by defining sults are not repeatable and will tend to be optimistic.
a weighting function incorporating the local build and turn rates of Worst case. Several variations on this idea are possible, but

the well. The approach described in Appendix A achieves the each will require some additional calculation. The principle is
same result without the need for a new weighting function. questionable, and the computational overhead is probably not jus-
The results of the two approaches differ only in the along-hole tified.
component of uncertainty. The along-hole uncertainty at a survey Borehole toolface i.e., the up-down left-right change in bore-

station includes the uncertainty in the stations measured depth, hole direction. This angle bears little relation to survey tool ori-
while the uncertainty at an assigned depth does not. entation, but is at least well defined, and may be computed di-
The correct choice of approach depends on the engineering rectly from inclination and azimuth data. This approach will tend
problem being tackled, in many cases it is immaterial. The user of to limit the randomization of toolface dependent errors, giving a
well-designed directional software need not be aware of the issue. conservative uncertainty prediction. This is the convention used in

H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000 227
 TABLE 2 STANDARD WELL PROFILES

ISCWSA No. 1North Sea extended reach well

Lat.60N, Long.2E, G9.80665 ms2, B50,000 nT, 72,
4W, Station interval30 m, Vertical Section Azimuth75
MD Inc Azi North East TVD VS DLS
(m) (deg) (deg) (m) (m) (m) (m) /30 m

0.00 0.000 0.000 0.00 0.00 0.00 0.00 0.00

1200.00 0.000 0.000 0.00 0.00 1200.00 0.00 0.00
2100.00 60.000 75.000 111.22 415.08 1944.29 429.79 2.00
5100.00 60.000 75.000 783.65 2924.62 3444.29 3027.79 0.00
5400.00 90.000 75.000 857.80 3201.34 3521.06 3314.27 3.00
8000.00 90.000 75.000 1530.73 5712.75 3521.06 5914.27 0.00

ISCWSA No. 2Gulf of Mexico fish-hook well

Lat.28N, Long.90W, G9.80665 ms2, B48,000 nT,
58, 2E, Station interval100 ft, Vert. Sect. Azim.21
MD Inc Azi North East TVD VS DLS
(ft) (deg) (deg) (ft) (ft) (ft) (ft) /100ft

0.00 0.000 0.000 0.00 0.00 0.00 0.00 0.00

2,000.00 0.000 0.000 0.00 0.00 2,000.00 0.00 0.00
3,600.00 32.000 2.000 435.04 15.19 3,518.11 411.59 2.00
5,000.00 32.000 2.000 1,176.48 41.08 4,705.37 1,113.06 0.00
5,525.54 32.000 32.000 1,435.37 120.23 5,153.89 1,383.12 3.00
6,051.08 32.000 62.000 1,619.99 318.22 5,602.41 1,626.43 3.00
6,576.62 32.000 92.000 1,680.89 582.00 6,050.92 1,777.82 3.00
7,102.16 32.000 122.000 1,601.74 840.88 6,499.44 1,796.70 3.00
9,398.50 60.000 220.000 364.88 700.36 8,265.27 591.63 3.00
12,500.00 60.000 220.000 1,692.70 1,026.15 9,816.02 1,948.01 0.00

ISCWSA No. 3Bass Strait designer well

Lat.40S, Long.147E, G9.80665 ms2, B61,000 nT, 70,
13E, Station interval30 m, Vertical Section Azimuth310
MD Inc Azi North East TVD VS DLS
(m) (deg) (deg) (m) (m) (m) (m) /30 m

0.00 0.000 0.000 0.00 0.00 0.00 0.00 0.00

500.00 0.000 0.000 0.00 0.00 500.00 0.00 0.00
1100.00 50.000 0.000 245.60 0.00 1026.69 198.70 2.50
1700.00 50.000 0.000 705.23 0.00 1412.37 570.54 0.00
2450.00 0.000 0.000 1012.23 0.00 2070.73 818.91 2.00
2850.00 0.000 0.000 1012.23 0.00 2470.73 818.91 0.00
3030.00 90.000 283.000 1038.01 111.65 2585.32 905.39 15.00
3430.00 90.000 283.000 1127.99 501.40 2585.32 1207.28 0.00
3730.00 110.000 193.000 996.08 727.87 2520.00 1197.85 9.00
4030.00 110.000 193.000 721.40 791.28 2417.40 1069.86 0.00

the examples which follow. Formulas for borehole toolface are Example Results
given in Appendix B.
The error models for basic and interference-corrected MWD have
been applied to the standard well profiles to generate position
Standard Profiles uncertainties in each well. The results of several combinations are
At the Eighth Meeting of the ISCWSA participants were set the tabulated in Table 3.
task of designing a number of well profiles suitable for testing Examples 1 and 2 compare the basic and interference-corrected
software implementations of the error models and propagation models in well ISCWSA No. 1. Being a high inclination well
mathematics, studying and highlighting the behavior of different running approximately ENE, the interference correction actually
error models magnetic and gyroscopic and individual error
degrades the accuracy. The results are plotted in Fig. 4. Examples
sources, demonstrating to a nonspecialist audience the uncertain-
3 to 6 all represent the basic MWD error model applied to well
ties to be expected from typical survey programs.
The ideas generated at the meeting were used to devise a set of ISCWSA No. 2. They differ in that each uses a different permu-
three profiles: tation of the survey station/assigned depth and symmetric error/
ISCWSA No. 1: an extended reach well in the North Sea, survey bias calculation options. The variation of lateral uncer-
ISCWSA No. 2: a fish-hook well in the Gulf of Mexico, with tainty and ellipsoid semimajor axis, characteristic of a fish-hook
a long turn at low inclination, and well, is shown in Fig. 5. Finally, example 7 breaks well ISCWSA
ISCWSA No. 3: a designer well in the Bass Strait, incorporat- No. 3 into three depth intervals, with the basic and interference-
ing a number of difficult hole directions and geometries. corrected models being applied alternately. This example is in-
Figs. 2 and 3 illustrate the test profiles in plan and section. cluded as a test of error term propagation.
Their full definition, given in Table 2, includes location, gravity The results in Table 3 were computed by the author, and have
and magnetic fields, survey stations, toolface angles and depth been independently verified by Anne Holmes, Steve Grindrod,
units. and Andy Brooks. Exact duplication of these results is a powerful

228 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
 TABLE 3 CALCULATED POSITION UNCERTAINTIES AT 1 s.d.. UNCERTAINTY AT THE TIE LINE MD0 IS ZERO; STATIONS INTERPOLATED AT WHOLE MULTIPLES
OF STATION INTERVAL USING MINIMUM CURVATURE; WELL PLAN WAY POINTS INCLUDED AS ADDITIONAL STATIONS; INSTRUMENT TOOLFACEBOREHOLE TOOLFACE

Uncertainties Along Correlations Between Survey Bias Along

Borehole Axes Borehole Axes Borehole Axes
Well
No. Depth interval(s) Model Option H L A HL HA LA bH bL bA
1 1 0 to 8000 m basic S, sym 20.116 m 84.342 m 8.626 m 0.016 0.676 0.004
2 1 0 to 8000 m ax-int S, sym 20.116 m 196.390 m 8.626 m 0.005 0.676 0.005
3 2 0 to 12,500 ft basic S, sym 16.185 ft 29.551 ft 10.057 ft 0.030 0.613 0.049
4 2 0 to 12,500 ft basic D, sym 16.185 ft 29.551 ft 9.080 ft 0.030 0.429 0.073
5 2 0 to 12,500 ft basic S, bias 15.710 ft 27.288 ft 8.526 ft 0.050 0.607 0.145 6.788 ft 12.411 ft 11.698 ft
6 2 0 to 12,500 ft basic D, bias 15.710 ft 27.288 ft 8.419 ft 0.050 0.574 0.148 6.788 ft 12.411 ft 4.758 ft
(1) 0 to 1380 m basic S, sym 2.013 m 4.703 m 0.919 m 0.007 0.633 0.006 (results at 1380 m)
7 3 (2) 1410 to 3000 m ax-int S, sym 3.239 m 3.646 m 7.890 m 0.172 0.623 0.665 (results at 3000 m)
(3) 3030 to 4030 m basic S, sym 5.604 m 5.892 m 9.594 m (results at TD4030 m)

H. S. Williamson: Accuracy Prediction for MWD

0.180 0.590 0.302
Key to error models: basic Basic MWD
ax-int Basic MWD with axial interference correction
Key to calculation options: S, sym Uncertainty at survey station, all errors symmetric (i.e., no bias)
S, bias Uncertainty at survey station, selected errors modeled as biases (see Table 4)
D, sym Uncertainty at assigned depth, all errors symmetric (i.e., no bias)
D, bias Uncertainty at assigned depth, selected errors modeled as biases (see Table 4)

els described in this paper.

error models in well ISCWSA No. 1.

axis in a fish-hook well, ISCWSA No. 2.

Conclusions and Recommendations

SPE Drill. & Completion, Vol. 15, No. 4, December 2000

229
Fig. 5Variation of lateral uncertainty and ellipsoid semimajor
Fig. 4Comparison of basic and interference corrected MWD

directional MWD service, Table 4, a set of well profiles for in-

and demanding test for implementations of the method and mod-

modeling. The standardized elements are a nomenclature see be-

lishes a common starting point for wellbore position uncertainty
This paper, and the collaborative work which it describes, estab-

of position uncertainty calculation, an error model for a basic

low, a definition of what constitutes an error model, mathematics
 TABLE 4 SUMMARY OF BASIC MWD ERROR MODELS

Weighting With Axial Propagation

Function Basic Model Correction Mode

Sensors

ABX 0.004 ms2 S

ABY 0.004 ms2 S
ABZ 0.004 ms2 S
ASX 0.0005 S
ASY 0.0005 S
ASZ 0.0005 S
MBX 70 nT S
MBY 70 nT S
MBZ 70 nT S
MSZ 0.0016 S
MSY 0.0016 S
MSZ 0.0016 S
ABIX 0.004 ms2 S
ABIY 0.004 ms2 S
ABIZ 0.004 ms2 S
ASIX 0.0005 S
ASIY 0.0005 S
ASIZ 0.0005 S
MBIX 70 nT S
MBIY 70 nT S
MSIX 0.0016 S
MSIY 0.0016 S
Misalignment
SAG 0.2 0.2 S
MX 0.06 0.06 S
MY 0.06 0.06 S
Axial magnetic interference
AZ 0.25 S
AMID 0.6 S or B*
Declination
AZ 0.36 0.36 G
DBH 5000nT 5000nT G
Total magnetic field and dip angle
MDI 0.20 G
MFI 130 nT G
Along-hole depth
DREF 0.35 m 0.35 m R
DSF 2.4104 2.4104 S
DST 2.2107 m1 2.2107 m1 G or B**

*When modeled as bias: 0.33, 0.5.

**When modeled as bias: 4.4107 m1, 0.

vestigating error models, and a set of results for testing software I wellbore inclination, deg
implementations. A wellbore azimuth, deg
The future work which these standards were designed to facilitate A m wellbore magnetic azimuth, deg
includes toolface angle, deg
establishment of agreed error models for other survey ser- magnetic declination, deg
vices, including in-field referencing and gyroscopic tools and magnetic dip angle, deg
interchangeability of calculated position uncertainties among B magnetic field strength, nT
survey vendor, directional drilling company, and operator. G gravity field strength, ms2**
Useful though this work is, it is only a piece in a large jigsaw X,x,Y,y,Z,z tool reference directions; see Fig. 1
puzzle. Taking a wider view, the collaborative efforts of the ex- *Adopted by ISCWSA participants as a standard for all technical
tended survey community should now be directed towards stan- correspondence.
dardization of quality assurance measures, strengthening the link **The international standard value for G of 9.806 65 ms2 was
between quality assurance specifications and error model param- used in the calculation of the results in Table 3.
eters, and better integration of wellbore position uncertainty with
the other aspects of oil field navigation. Special Nomenclature
b component of wellbore position bias vector
Nomenclature
B estimated magnetic field strength
ISCWSA Nomenclature* B H horizontal component of magnetic field
D along-hole depth, m, ft strength

230 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
 C wellbore position uncertainty covariance matrix 4. Dubrule, O. and Nelson, P.H.: Evaluation of Directional Survey
e 1 s.d. vector error at an intermediate station Errors at Prudhoe Bay, SPEDE September 1987 257.
e* 1 s.d. vector error at the station of interest 5. Grindrod, S.J. and Wolff, J.M.: Calculation of NMDC Length Re-
quired for Various Latitudes Developed From Field Measurements of
E sum of vector errors from slot to station of in-
Drill String Magnetisation, paper 11382 presented at the SPE 1983
terest IADC/SPE Drilling Conference, New Orleans, 2023 February.
particular value of a survey error 6. Macmillan, S. et al.: Error estimates for geomagnetic field values
H,L used in calculation of the toolface computed from the BGGM, British Geological Survey Technical
m bias vector error at an intermediate station Report No. WM/93/28C 1993.
m* bias vector error at the station of interest 7. Ekseth, R.: Uncertainties in Connection with the Determination of
M wellbore position bias vector Wellbore Positions, PhD dissertation, Norwegian U. of Science and
mean of error value Technology, Trondheim, Norway 1998.
standard deviation of error value, component of
wellbore position uncertainty
p survey measurement vector (D,I,A)
P,Q,R intermediate calculated quantities Appendix A: Mathematical Description of Propagation
r wellbore position vector Model
rk increment in wellbore position between stations The total position uncertainty at a survey station of interest, K in
k1 and k survey leg L is the sum of the contribution from all the active
correlation coefficient error sources. It is convenient computationally to group the error

estimated magnetic dip angle sources by their propagation type and to sum them separately.
v along-hole unit vector
w factor relating error magnitude to uncertainty in Vector Errors at the Station of Interest. Recall that the vector
measurement error due to the presence of error source i at station k is the sum of
the effect of the error on the preceding and following survey dis-
Subscripts and Counters placements:

hla,HLA
i


borehole referenced frame
a survey error term
ei,l,k i,l drk drk1 pk
dpk

dpk i
. A-1
k a survey station
K survey station of interest Evaluating this expression using the minimum curvature well
Kl number of stations in the lth survey leg trajectory model is cumbersome. There is no significant loss of
accuracy in using the simpler balanced tangential model:

l a survey leg
L survey leg containing the station of interest sin I j1 cos A j1 sin I j cos A j
nev earth-referenced frame D j D j1
r j sin I j1 sin A j1 sin I j sin A j . A-2
2
cos I j1 cos I j
Superscripts
The two differentials in the parentheses in Eq. A-1 may then be
dep at the along-hole depth assigned to the survey expressed as
station
rand random propagation mode
svy at the point where the survey measurements
dr j
dpk

dr j dr j dr j
dD k dI k dA k jk,k1 , A-3

were taken
syst systematic propagation mode sin I k1 cos A k1 sin I k cos A k
well well by well or global propagation mode drk 1
sin I k1 sin A k1 sin I k sin A k , A-4a
dD k 2
cos I k1 cos I k

Acknowledgments
The author thanks all participants in the ISCWSA for their enthu- sin I k cos A k sin I k1 cos A k1
drk1 1
siasm and support over several years and in the review of this sin I k sin A k sin I k1 sin A k1 , A-4b
dD k 2
paper. cos I k cos I k1

Particular contributions to the MWD error model were made by
John Turvill and Graham McElhinney, both now with PathFinder D j D j1 cos I k cos A k
dr j 1
Energy Services, formerly Halliburton Drilling Systems; Wayne D j D j1 cos I k sin A k jk,k1 , A-5
dI k 2
Phillips, Schlumberger, Paul Rodney and Anne Holmes, Sperry- D j D j1 sin I k

Sun Drilling Services; and Oddvar Lotsberg, formerly of Baker
Hughes INTEQ. D j D j1 sin I k sin A k
dr j 1
Participants in the working group on error propagation were D j D j1 sin I k cos A k jk,k1 . A-6
David Roper, Sysdrill Ltd.; Andy Brooks and Harry Wilson, dA k 2
0
Baker Hughes INTEQ; and Roger Ekseth, formerly of Statoil.
The author also wishes to thank BP for their permission to For the purposes of computation, the error summation terminates
publish this paper. at the survey station of interest. Vector errors at this station are
therefore given by
References drK pK
1. Wolff, C.J.M. and de Wardt, J.P.: Borehole Position Uncertainty * i,L
ei,L,K . A-7
dpK i
Analysis of Measuring Methods and Derivation of Systematic Error
Model, JPT December 1981 2339. * indicates that a measurement error at this sta-
The notation ei,L,K
2. Thorogood, J.L.: Instrument Performance Models and Their Appli- tion affects only the preceding survey displacement. In what fol-
cation to Directional Survey Operations, SPEDE December 1990
294.
lows we reserve the notation ei,l,k for vector errors at intermediate
3. Brooks, A.G. and Wilson, H.: An Improved Method for Computing stations, which affect both the preceding and following displace-
Wellbore Position Uncertainty and Its Application to Collision and ments.
Target Intersection Probability Analysis, paper 36863 presented at
the SPE 1996 SPE European Petroleum Conference, Milan, Italy, Undefined Weighting Functions. For some combinations of
2224 October. weighting function and hole direction, one component of the mea-

H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000 231
 TABLE A-1 ERROR VECTORS IN VERTICAL HOLE WHERE WEIGHTING FUNCTION
IS SINGULAR

Sensor errors (with or without axial interference correction)

ABX
or
ABIX
ei , l , k
i , l D k 1 D k 1
2G
cos
sinA
A
0

ABY
or
ABIY
ei , l , k
i , l D k 1 D k 1
2G
cosA
sinA
0

Misalignment errors

MX ei , l , k
i , l D k 1 D k 1
2 sinA
cosA
0
MY ei , l , k
i , l D k 1 D k 1
2 cosA
sinA
0

surement vector usually the azimuth is highly sensitive to

changes in hole direction and the vector p / i is apparently
undefined. There are two cases.
syst
C i,l
Kl

k1
ei,l,k "
Kl

k1
ei,l,k
T

, A-12

Vertical Hole. In this case, dr/dp is zero but the vectors ei,l,k and the total contribution over all survey legs is
* are still finite and well defined. They may be computed
and ei,L,K
L1
by forming the products of Eqs. A-1 and A-7 algebraically and
evaluating them as a whole. Take as an example the weighting
syst
C i,K C
l1
syst
i,l
function for an X-axis radially symmetric misalignment. Substitut-
ing the expression for p/ M X , Eq. 27, and the well trajectory
model equations, Eqs. A-3 to A-6, into Eqs. A-1 and A-7, and
setting I equal to zero gives
K1

k1
ei,L,k ei,L,K
* " K1

k1
ei,L,k ei,L,K
* T

. A-13

sin A Well by Well and Global Errors. Each of these error types is
i,l D k1 D k1 systematic among all stations in a well. The individual vector
ei,l,k cos A A-8
2 errors can therefore be summed to give a total vector error from
0 slot to station:
and L1

e Kl

K1

Ei,K i,l,k ei,L,k ei,L,K
* . A-14
sin A l1 k1 k1
i,L D K D K1
*
ei,L,K cos A . A-9 The total contribution to the uncertainty at survey station K is
2
0
well
C i,K Ei,K "Ei,K
T
. A-15
There are similar expressions for Y-axis axial misalignment and
X- and Y-axis accelerometer biases. These are given in Table A-1. Total Position Covariance. The total position covariance at sur-
Equivalent expressions may be used for evaluating bias vectors in vey station K is the sum of the contributions from all the types of
* , i,l and i,L substituted for
the vertical hole, with mi,l,k , mi,L,K error source:
* and i,l and i,L , respectively.
ei,l,k , ei,L,K
Other Hole Directions. Some error sources really are un- C Ksv y C
iR
i,K
rand
C
iS
i,K
syst

i W,G
well
C i,K , A-16
bounded in certain hole directions. The examples in this paper are
sensor errors after axial interference correction in a horizontal and where the superscript svy indicates the uncertainty is defined at a
magnetic east/west wellbore, a so-called 90/90 well. In such survey station.
cases, the assumptions of linearity break down, and computed
position uncertainties are meaningless. Software implementations Survey Bias. Error vectors due to bias errors are given by expres-
should include an error-catching mechanism for this case. sions entirely analogous with Eqs. A-1 and A-7:

Summation of Errors. Vector errors are summed into position

uncertainty matrices as follows.
mi,l,k i,l drk drk1 pk
dpk

dpk i
, A-17
Random Errors. The contribution to survey station uncertainty
from a randomly propagating error source i over survey leg l not drK pK
* i,L
mi,L,K . A-18
containing the station of interest is dpK i
Kl The total survey position bias at survey station K, MKsv y , is the
rand
C i,l e
k1
i,l,k " ei,l,k
T
, A-10 sum of individual bias vectors taken over all error sources i, legs
l and stations k:
and the total contribution over all survey legs is
L1 K1
MKsv y
i
L1

l1
Kl

kl
mi,l,k K1

k1
mi,L,k mi,L,K
* . A-19
rand
C i,K C
l1
i,l
rand
e
k1
i,L,k " ei,L,k
T
Position Uncertainty and Bias at an Assigned Depth. Defining
the superscript dep to indicate uncertainty at an assigned depth, it
* " ei,L,K
ei,L,K * T. A-11 may be shown that
Systematic Errors. The contribution to survey station uncer- ei,L,K * s v y i,L w i,L,K vK ,
* de p ei,L,K A-20
tainty from a systematically propagating error source i over sur-
vey leg l not containing the station of interest is
de p
ei,l,k ei,l,k
svy
, A-21

232 H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000
 Relative Uncertainty Between Wells. When calculating the un-
certainty in the relative position between two survey stations
(K A ,K B ) in wells A,B, we must take proper account of the cor-
relation between globally systematic errors. The uncertainty is
given by

C s v y rK A rK B C Ks vAy C Ks vBy E
iG
i,K A " Ei,K B
T

Ei,K B " Ei,K A T . A-24

The relative survey bias is simply
Ms v y rK A rk B MKs vAy MKs vBy . A-25
Substitution of Eqs. A-20 to A-23 into these expressions gives
the equivalent results at the along-hole depths assigned to the
stations.

Transformation Into Borehole Reference Frame. The results

derived above are in an Earth-referenced frame north, east, ver-
tical, subscript nev. The transformation of the covariance matri-
ces and bias vectors into the more intuitive borehole referenced
frame highside, lateral, along hole, subscript hla is straightfor-
Fig. A-1Vector errors at the last station point of interest due ward:
to an along-hole depth error at the last station. C hla T T C ne v T , A-26

where w i,L,K is the factor relating error magnitude to depth mea-

surement uncertainty and vK is the along-hole unit vector at sta-
tion K. Figs. A-1 and A-2 illustrate these results. Substituting
bH
b L Mhla T T Mne v ,
bA
A-27

these expressions into Eqs. A-12 to A-16 yields the position un- where
certainty at the along-hole depth assigned to each survey station. cos I K cos A K sin A K sin I K cos A K
Survey bias at an assigned depth is calculated by substituting
the following error vectors into Eq. A-19: T cos I K sin A K cos A K sin I K sin A K A-28
sin I K 0 cos I K
is a rotation matrix. Uncertainties and correlations in the principal
borehole directions are obtained from
H C hla 1,1 etc., A-29
C hla 1,2
HA etc. A-30
H L
Appendix B: Calculation of Toolface Angle
The following formulas may be used to calculate borehole tool-
face angle from successive surveys:
H K sin I K cos I K1 sin I K1 cos I K cos A K A K1 , B-1
L K sin I K1 sin A K A K1 , B-2
1
if H K 0, K tan L K /H K , B-3
1
if H K 0, K tan L K /H K 180, B-4
if H K 0, K 270, 0 or 90 as L K
0, L K 0 or L K 0. B-5

SI Metric Conversion Factors

Fig. A-2Vector errors at the last station point of interest due ft 3.048* E01 m
to an along-hole depth error at an earlier station.
*Conversion factor is exact. SPEDC

mi,L,K * s v y i,L w i,L,K vK ,

* dep mi,L,K A-22 Hugh. S. Williamson is a well positioning specialist with BPs Up-
stream Technology Group in Sunbury-on-Thames, U.K. e-mail:
williahs@bp.com. He holds a degree in mathematics from
Cambridge U. and a degree in engineering surveying and ge-
dep
mi,l,k mi,l,k
svy
. A-23 odesy from Nottingham U.

H. S. Williamson: Accuracy Prediction for MWD SPE Drill. & Completion, Vol. 15, No. 4, December 2000 233