Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

GdE 416 - 4 Error Propagation Law

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

4 ERROR PROPAGATION

Unknown values are often determined indirectly by making direct measurements of other quantities which are
functionally related to the desired unknowns. Examples in surveying include computing station coordinates from
distance and angle observations, obtaining station elevations from rod readings in differential leveling, and
determining the azimuth of a line from astronomical observations. Since all quantities that are measured directly
contain errors, any values computed from them will also contain errors. This intrusion, or propagation, of errors that
occurs in quantities computed from direct measurements is called error propagation. The errors in the
measurements are usually represented by the standard deviations, which, when propagated through
the functions, yield standard deviations of the computed quantities.

4.1 Basic Error Propagation Equation


Let z be a function consisting of x1, x2…xn uncorrelated variables having individual standard
deviations σ1,σ2…σn ,
𝑧 = 𝑎1 𝑥1 + 𝑎2 𝑥2 +. . . 𝑎𝑛 𝑥𝑛

then the variance of the computed quantity z,

𝜕𝑧 2 𝜕𝑧 2 𝜕𝑧 2
𝜎𝑧2 = (𝜕𝑥 ) 𝜎12 + (𝜕𝑥 ) 𝜎22 + ⋯ (𝜕𝑥 ) 𝜎𝑛2 (4.1)
1 2 𝑛

In matrix form:
𝜕𝑧
𝜕𝑥1
𝜎12 0 0
𝜕𝑧 𝜕𝑧 𝜕𝑧 𝜕𝑧
𝜎2 = [ ][ 0 𝜎22 0] 𝑠𝑖𝑛𝑔𝑙𝑒 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝑒𝑟𝑟𝑜𝑟 𝑝𝑟𝑜𝑝𝑎𝑔𝑎𝑡𝑖𝑜𝑛
𝜕𝑥1 𝜕𝑥2 𝜕𝑥3 𝜕𝑥2
0 0 𝜎32 𝜕𝑧
[𝜕𝑥3 ]
where:

𝜕𝑧 𝜕𝑧
, = partial derivatives of function f with respect to x1….xn
𝜕𝑥1 𝜕𝑥𝑛
𝜎12 … . 𝜎𝑛2 = variances of x1…. xn respectively

Case 1: z’s are a series of independently observed quantities with partial derivatives

𝜕𝑧 𝜕𝑧 𝜕𝑧 ⬚
= = =1
𝜕𝑥1 𝜕𝑥2 𝜕𝑥𝑛

having standard deviations of S1, S2 , . . . , Sn . Then

𝑆𝐴2 = 𝑆12 + 𝑆22 + ⋯ … . 𝑆𝑛2


𝑆𝐴 = √𝑆12 + 𝑆22 + ⋯ … + 𝑆𝑛2 𝑆𝑡𝑎𝑛𝑑𝑎𝑟𝑑 𝐷𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 𝑜𝑓 𝑎 𝑆𝑢𝑚 (4.3)

Example 4.1

A series of tape measurements with differing lengths and standard deviations.

x1,s1 x2, s2 x3,s3

Series of tape measurements

x1= 64.541 m s1 = ±.005 m


x2= 89.256 m s2 = ±.010 m
x3 = 100.281 m s3 = ±.015 m
GdE 416 4: Error Propagation page 2

Calculate the total distance and its standard deviation.

Solution
Total distance is the summation of all indvividual measurements:

f(x) = x1 + x2 + x3 = 254.078 m

the variance of the entire length is

𝜕𝑓 2 2 𝜕𝑓 2 2 𝜕𝑓 2 2
𝜎2 = ( ) 𝜎1 + ( ) 𝜎2 + ( ) 𝜎3
𝜕𝑥1 𝜕𝑥2 𝜕𝑥3

Since

𝜕𝑓 𝜕𝑓 𝜕𝑓
= = =1
𝜕𝑥1 𝜕𝑥2 𝜕𝑥3

therefore:
𝜎 2 = 𝜎12 + 𝜎22 + 𝜎32 (4.3)
𝜎 2 = 0.0052 + 0.0102 + 0.0152
𝜎 2 = 0.00035
𝜎 = ±0.019 𝑚

The standard deviation of the computed quantity is always greater than any of the standard deviation
of the observed quantities.

Example 4.2.
The area of a rectangular parcel of land is required together with its standard deviation. The length is
a = 100 m with a = 0.10 m and the width b = 40 m with  b= .08 m. Assume a and b to be
uncorrelated.

b = 40 m

a = 100 m.

Solution:
A = ab = 4000 m2
𝜕𝐴
=𝑏
𝜕𝑎
𝜕𝐴
=𝑎
𝜕𝑏
Therefore:
𝜕𝐴 2 2 𝜕𝐴 2 2
( ) 𝜎𝑎 + ( ) 𝜎𝑏 = 𝜎𝐴2
𝜕𝑎 𝜕𝑏
A2 = b2 a2 + a2b2 = 402(0.1)2 + (100) 2(.08)2 = 80m4
A = ±8.9 m2

This means that the area may deviate by ± 8.94 m2.(68% confidence);± 17.52 m2 (95%)
4000 m2 ± 9 m2
GdE 416 4: Error Propagation page 3

4.3 ERRORS IN ELECTRONIC DISTANCE OBSERVATIONS


All EDM observations are subject to instrumental errors that manufacturers list as constant, a, and scalar, b, error.
A typical specified accuracy is ±(a +b ppm). In this expression, a is generally in the range 1 to 10 mm, and b is a
scalar error that typically has the range 1 to 10 ppm. Other errors involved in electronic distance observations stem
from the target and instrument centering errors. Since in any survey involving several stations these errors tend to be
random, they should be combined using Equation (6.18). Thus, the estimated error in an EDM observed distance is

σD = √σ2i + σ2t + a2 + (Dxb ppm)2 (5.18)


where:
σD is the error in the observed distance D(km),
σi the instrument centering error,
σt the reflector centering error, and
a and b the instrument’s specified accuracy parameters.

Example 7.11 A distance of 138.375 m is observed using an EDM with a manufacturer’s specified accuracy of ±(5
mm + 10 ppm). The instrument is centered over the station with an estimated error of ±0.9 mm, and the reflector,
which is mounted on a handheld prism pole, is centered with an estimated error of ±3 mm. What is the error in the
distance? What is the E95 value?

SOLUTION
The scalar portion of the manufacturer’s estimated standard error is computed as

√52 + (10𝑥0.138)2 = 5.2 𝑚𝑚


therefore the distance error
σ = √0.92 + 32 + 52 + 5.22 = ±7.9 mm

Using the appropriate t value from Table , the 95% probable error is E95= 1.96σ=15.5 mm.

Notice in this example that the instrument’s constant error is the largest single contributor to the overall error in the
observed distance, and it is followed closely by the target centering error. Furthermore, since both errors are
constants, their contribution to the total error is unchanged regardless of the
distance. Thus, for this particular EDM instrument, distances under 200 ft (60 m) could probably be observed more
accurately with a calibrated steel tape. However, this statement depends on the terrain and the skill of the surveyors
in using a steel tape.

Example 4.3.

A triangular area is measured using a total station with a distance accuarcy of σ=±(2mm + 5 ppm x D)
and an angle accuracy of 5”. Corner C was occupied by the total station and measured a, b and angle
C. Calculate the area and its standard deviation. Assume a, b and C to be uncorrelated. The result of
the measurement is

a = 254.174 m.
b = 327.812 m.
C = 52o 42’ 21”

Solution
c
a a. Calculate the standard error of distance observations
𝜎𝑎 = √22 + (0.254174 𝑥5)2 = ±2.4 𝑚𝑚

A 327.812𝑥5 2
C
b 𝜎𝑏 = √22 + ( ) = ±2.6 𝑚𝑚
1,000

b. Calculate the area


1 1
𝐴 = 𝑎𝑏 sin 𝐶 = (254.174)(327.812) sin 52𝑜 42′21"
2 2
𝐴 = 33,142.5 𝑚2

b. Calculate the standard deviation of the area


GdE 416 4: Error Propagation page 4

𝜕𝐴 2 2 𝜕𝐴 2 2 𝜕𝐴 2 2
𝜎𝐴2 = ( ) 𝜎𝐴 + ( ) 𝜎𝐵 + ( ) 𝜎𝐶
𝜕𝑎 𝜕𝑏 𝜕𝐶

𝜕𝐴 1
= 𝑏 sin 𝐶 = 130.393 𝑚 𝜎𝑎2 = 5.76𝑥10−6 𝑚2
𝜕𝑎 2
𝜕𝐴 1
= 𝑎 sin 𝐶 = 101.102 𝑚 𝜎𝑏2 = 6.76𝑥10−6 𝑚2
𝜕𝑏 2
𝜕𝐴 1
𝜕𝐶
= 2 𝑎𝑏 cos 𝐶 = 25,242.5 𝑚2 𝜎𝐶2 = (5 x sin 1")2 = 5.88𝑥10−10 (𝑟𝑎𝑑)2

𝜎𝐴2 = (130.393)2 𝑥(5.76𝑥10−6 ) + (101.102)2 𝑥(6.76𝑥10−6 ) + (25,242.5 )2 𝑥(5.88𝑥10−10 )

𝜎𝐴2 = 0.543 𝑚4

𝜎𝐴 = ± 0.737 𝑚2 (68 %) 1.44 m2 (95%)

Example 4.4 The dimensions of the rectangular tank shown below:

are measured as
L= 40.00 ft SL = 0.05 ft
W = 20.00 ft SW =0.03 ft
H = 15.00 ft SH =0.02 ft
Find the tank’s volume and the standard deviation in the volume using the measurements above.

SOLUTION The volume of the tank is found using the formula

V= LWH = 40.00(20.00)(15.00) = 12,000 ft3

𝜕𝑉 2 2 𝜕𝑉 2 2 𝜕𝑉 2
( ) 𝜎𝐿 + ( ) 𝜎𝑊 + ( ) 𝜎𝐻2 = 𝜎𝑉2
𝜕𝐿 𝜕𝑊 𝜕𝐻
𝜕𝑉 𝜕𝑉 𝜕𝑉
𝜕𝐿
= 𝑊𝐻 = 300; 𝜕𝑊
= 𝐿𝐻 = 600; 𝜕𝐻
= 𝐿𝑊 = 800

Sv2 = (300)2 SL 2 + (600)2 SW 2 + (800)2 SH 2

Sv2 = 805 ft 6
Sv = ±28 ft 3 (68%) 55 ft3 95%

3. [20] The area of a lot shown below was being measured by pacing. Sides 1-2 and 2-3 were
both measured to be 56 paces and side 3-1 at 64 paces. A pace is 0.68 m having a standard
error of ±0.05 m/pace. Calculate the area of the lot and its standard error at 95% confidence.

2
a
1
h a

b
3

Solution
1-2=56 pacesx0.68 =a=38.08 m; 𝑠𝑎 = 0.05√56=0.37 m
2-3 = 56 pacesx0.68=a= 38.08 m𝑠𝑎 = 0.37 𝑚
3-1 = 64 pacesx0.68 = b=43.52 m sb = 0.05√64 = 0.4
GdE 416 4: Error Propagation page 5

Since the triangle is an isosceles triangle the area consist of two equal right triangles. The
total area is

1
Area = bh
2

To calculate the height h:


𝑏 2 43.52 2
ℎ = √𝑎2 − ( ) = √38.082 − ( ) = 32.12 𝑚
2 2
Therefore the area is:
1
Area = (43.52)(32.12) = 698.93 𝑚2
2
To determine the standard error of the area, the law of error propagation is applied:

𝜕𝐴 2 2 𝜕𝐴 2
𝑠𝐴2 = ( ) 𝑠𝑏 + ( ) 𝑠ℎ 2
𝜕𝑏 𝜕ℎ
Where:

𝜕𝐴 2 1 2 32.12 2 2
( ) = ( ℎ) = ( ) = 114.63 𝑚; 𝑠𝑏2 = (. 05√64) = 0.16 𝑚2
𝜕𝑏 2 2
𝜕𝐴 2 1 2 43.52 2 𝜕ℎ 2 2 𝜕ℎ 2
( ) = ( 𝑏) = ( ) = 210.44 𝑚; 𝑠ℎ2 = ( ) 𝑠𝑎 + ( ) 𝑠𝑏2
𝜕ℎ 2 2 𝜕𝑎 𝜕𝑏
−1⁄2 2 −1⁄2 2
𝜕ℎ 2 1 𝑏 2 43.52 2
( ) = ( [𝑎2 − ( ) ] 2𝑎) = [38.08 (38.08 − ( 2
) ) ] = 1.48𝑚2
𝜕𝑎 2 2 2
−1⁄2 2 −1⁄2 2
𝜕ℎ 2 1 𝑏 2 43.52 2
( ) = ( [𝑎2 − ( ) ] − 𝑏) = [−21.76 (38.08 − ( 2
) ) ] = 0.48𝑚2
𝜕𝑏 2 2 2
𝑠ℎ2 = 1.48𝑠𝑎2 + 0.48𝑠𝑏2 = 1.48(. 37)2 + 0.48(0.28)2 = 0.24𝑚2

𝑠𝐴2 = 114.63 ∗ 0.16 + 210.44 ∗ 0.24 = 68.85𝑚4


𝑠𝐴 = 8.3 𝑚2

4.4 General Error Propagation Equation


For two functions x and y where the variables are uncorrelated, a co-variance matrix is formed for the two
functions

𝜕𝑥 𝜕𝑥 𝜕𝑥 𝜕𝑦
⋯ 𝜎𝑥21 ⋯ 0 𝜕𝑥1 𝜕𝑥1
𝜎𝑥2 𝜎𝑥𝑦 𝜕𝑥1 𝜕𝑥𝑛
∑ 𝑥, 𝑦 = [ ]= [( ⋮ ⋱ ⋮ )] ⋮ ⋮
𝜎𝑥𝑦 𝜎𝑦2 𝜕𝑦 𝜕𝑦
0 ⋯ 𝜎𝑥2𝑛 𝜕𝑥 𝜕𝑦

[𝜕𝑥1 𝜕𝑥𝑛 ] [𝜕𝑥𝑛 𝜕𝑥𝑛 ]
or
∑ = 𝐽∑ 𝐽𝑇
12 𝑥

The term 𝜎𝑥𝑦 is known as the covariance. The covariance expresses the mutual interrelation
betweeen the two random variables The above system of equations shows a second order error
propagation.

Example 4.4:
The position of point A is determined by the radial distance r = 100m, with sr = 0.05 m and the
azimuth angle θ = 60, with sθ = 2’.Calculate the covariance matrix for point A. Assume r and  to be
uncorrelated.
GdE 416 4: Error Propagation page 6

Y
x = r cos θ x A
y = r sin θ
r = 100.00 m
sr= .05 m r y
θ = 60°
sθ = 2’=120” θ
X

Solution
𝜕𝑥
= cos 𝜃
𝜕𝑟
𝜕𝑥
= −𝑟 sin 𝜃
𝜕𝜃
𝜕𝑦
= sin 𝜃
𝜕𝑟
𝜕𝑦
= 𝑟𝑐𝑜𝑠𝜃
𝜕𝜃

 x x 

J =  r   = cos − r sin    0.5 − 86.60
=
y y   sin  r cos  0.866 50.00 
 
 r  

The covariance matrix of the known variables, r and , is a diagonal matrix because they are
uncorrelated. It is equal to :

𝒔𝟐𝒓 𝟎 . 𝟎𝟎𝟐𝟓 𝟎
𝜮𝒓𝜽 = [ 𝟐] = [ ]
𝟎 𝒔𝜽 𝟎 𝟑. 𝟒𝒙𝟏𝟎−𝟕

𝑠𝑥2 𝑠𝑥𝑦 .5 −86.6 . 0025 0 .5 . 866 . 0031 −.00035


[ ]=[ ][ ][ ]=[ ]
𝑠𝑥𝑦 𝑠𝑦2 . 866 50 0 3.38𝑥10−7 −86.6 50 −.00035 . 0027

The standard deviations are:


𝑠𝑥 = √. 0031 = 0.06
𝑠𝑦 = √. 0027 = 0.05
𝑠𝑥𝑦 = −0.0004
𝑠𝑥𝑦 0.0004
𝜌𝑥𝑦 = = = 0.13
𝑠𝑥 𝑠𝑦 0.06 𝑥 0.05
Example 4.6
Horizontal distance and difference in elevation can be computed from stadia observations by using
the following equations based on the figure below,

stadia hair

h.i

A H

Stadia Surveying
GdE 416 4: Error Propagation page 7

H = K s sin2 z
V = ½ K s sin 2z + h.i.
where: K = stadia interval factor
s = stadia interval
z = zenith angle
Let K=100 , s=1.376, s= 0.010 m, z= 65 23' 16" , = 5".
Assume s and z are uncorrelated
1. Calculate the H and V and the covariance matrix H and V
2. Calculate the co-relation coefficient of H and V
Solution
𝜕𝐻 𝜕𝐻 𝜕𝐻 𝜕𝑉
𝜎2 𝜎𝐻𝑉 2
𝜕𝑧 ] [𝜎𝑠 0 𝜕𝑠
[ 𝐻 ] = [ 𝜕𝑠 ][ 𝜕𝑠 ]
𝜎𝐻𝑉 𝜎𝑉2 𝜕𝑉 𝜕𝑉 0 𝜎𝑧2 𝜕𝐻 𝜕𝑉
𝜕𝑠 𝜕𝑧 𝜕𝑧 𝜕𝑧
𝜕𝐻 𝜕𝐻
= 𝐾𝑠𝑖𝑛2 𝑧 = 82.65 = 2𝐾𝑠 sin 𝑧 cos 𝑧 = 𝐾𝑠 sin 2𝑧 = 104.20 𝑚 𝜎𝑠2 = 100𝑥10−6 = .0001 𝑚2
𝜕𝑠 𝜕𝑧
𝜕𝑉 1 𝜕𝑉
= 𝐾 𝑠𝑖𝑛2 𝑧 = 37.86 = 𝐾𝑠 cos 2𝑧 = −89.87 𝑚 𝜎𝑧2 = 5.88𝑥10−10 𝑟𝑎𝑑
𝜕𝑠 2 𝜕𝑧
𝜎𝐻2 ⬚
𝜎𝐻𝑉 82.65 104.20 0.0001 0 82.65 37.86
[ ⬚ ]=[ ][ −10 ] [ ]
𝜎𝐻𝑉 𝜎𝑉2 37.86 −89.87 0 5.88𝑥10 104.20 −89.87

𝜎2 𝜎𝐻𝑉 0.6831 0.3129


[ 𝐻 ]=[ ]
𝜎𝐻𝑉 𝜎𝑉2 0.3129 0.1433
𝜎𝐻 = ±0.826 𝑚
𝜎𝑉 = ±0.378 𝑚
𝜎𝐻𝑉 = 0.3129 𝑚
0.3129
𝜌= =1
. 826 ∗ .378

4.3.1 Correlation Coefficient


This term relates the variances and its covariance to each other where σx and σy are the standard
deviations of x and y respectively.
𝝈𝒙𝒚
𝝆𝒙𝒚 =
𝝈 𝒙 𝝈𝒚

4.6. ERROR ELLIPSES

In two dimensional problems, such as the horizontal position of a point, error ellipses may be
established around the point to determine the uncertainty region for a point. The orientation of the
ellipse relative to the x, y axes system depends on the correlation between x and y. If they are
uncorrelated the ellispse axes will be parallel to x and y. If the two coordinates are of equal precision,
or x = y, the ellipse becomes a circle.
GdE 416 4: Error Propagation page 8

y
σy
x

θ
b a
σx
A x
r

Assume that estimated standard deviation in a distance r and direction  are r and , respectively

The semi-major and semi-minor axes of the corresponding ellipse are computed in the following:
1. A characteristic second degree polynomial is set up using the elements of the
covariance matrix as:
𝜎𝑥2 𝜎𝑥𝑦 𝜆 0
[ 2 ] = [0 𝜆 ]
𝜎𝑥𝑦 𝜎𝑦
Then
𝜎𝑥2 − 𝜆 𝜎𝑥𝑦
[ ]=0
𝜎𝑥𝑦 𝜎𝑦2−𝜆

And the characteristic polynomial is

(𝜎𝑥2 − 𝜆)(𝜎𝑦2 − 𝜆) − 𝜎𝑥𝑦


2
=0

or
2 − (x2+ y2)+ (x2y2 −xy2 ) = 0

2. The two roots 1,2 called eigenvalues of  are computed and their square roots are the
semi-major and semi-minor axes of the standard error ellipse.
Note:
Quadratic formula:
−𝑏 ± √𝑏 2 − 4𝑎𝑐
𝜆=
2𝑎
√𝜆1 = 𝑎

√𝜆2 = 𝑏
3. The orientation of the ellipse is determined by computing  between the x axis and the
semi-major axis from
2𝜎𝑥𝑦
tan 2𝜃 =
𝜎𝑥2
− 𝜎𝑦2
Example 4.6

Construct the error ellipse of the covariance matrix.


  x2  xy   0.3767 − 0.2195
xy   2  = − 0.2195 0.630 
=
 xy y   

0.3767 − 𝜆 −0.2195
[ ]=0
−0.2195 0.630 − 𝜆

(0.3767 − 𝜆)(0.630 − 𝜆) − (−0.2195)2 =0


GdE 416 4: Error Propagation page 9

The characteristic polynomial is 2 - (0.3767+0.63) + (0.3767)(0.6300)-(-0.2195)2 = 0


2 - 1.0067+ 0.189141 = 0
Applying quadratic formula, 1= 0.76 2=0.25
The semi major axis is a=1 =0.87 m b=2=0.50 m

To get the orientation of the axes,


2(−0.2195)
tan 2𝜃 = = 1.732
(0.3767 − 0.6300)
 = 1/2 tan−1 1.732
 =30
It is impossible to maintain an exact equality between these two relative accuracies; but with some
exceptions, to be considered presently, surveys should be conducted so that the difference between
angular and distance accuracies is not great.
HW:
Construct the error ellipse of the co-variance matrix below:

. 0031 −.00035
[ ]
−.00035 . 0027

If a total station is used having an angular error of 5” and distance error of (3mm + 5ppm), what would be the
size of the ellipse at 95% probability.

Example 4.7
Horizontal distance and difference in elevation can be computed from stadia observations by using the ff. equations:
H = K s cos2 
V = ½ K s sin 2

where: K = stadia interval factor s = stadia interval  = vertical angle

1. Calculate the H and V if K=100, s = 1.376 and =35 23' 16"


2. Construct co-variance matrix of H and V if the estimated standard deviation of the stadia interval s= 0.02 m and
the standard deviation of the vertical angle is = 5".
3. Calculate the co-relation coefficient of H and V and assume s and  are uncorrelated.
4. Construct the error ellipse having a probability of 50 %.

4.6.1 Standard Error Ellipse

In the previous lesson of one dimensional accuracy, the probability that the true value was
within ±σ was 68%, but in the ellipse, the probability that the true value was within the ellipse region
was 39.4% It should be obvious that the larger the size of the ellipse the larger is the probability.
Using the standard ellipse as the base the table below gives the scale multiplier k to enlarge the
ellipse and the corresponding probability

Point Estimate probability values


2-D 3-D
P K p k
0.394 1.000 0.199 1.00
0.500 1.177 0.500 1.538
0.632 1.414 0.900 2.500
0.900 2.146 0.950 2.700
0.950 2.447 0.990 3.368
0.990 3.035

4.6.2 Standard error circle approximations

If we let a=σmax and b=σmin, then the value of the ratio σmin/σmax determines the relationship used to
calculate σC. The aim is to make the ellipse approximately a circle.
GdE 416 4: Error Propagation page 10

𝜎
Case 1 𝑓𝑜𝑟 0.8 < 𝜎 𝑚𝑖𝑛 < 1.0, 𝑡ℎ𝑒𝑛 𝑡ℎ𝑒 𝑚𝑒𝑎𝑛 𝑎𝑐𝑐𝑢𝑟𝑎𝑐𝑦 𝜎̃𝑐 𝑖𝑠
𝑚𝑎𝑥

1⁄ 1⁄
1 2 1 2
𝜎̃𝑐 = [ (𝑎2 + 𝑏 2 )] = [ (. 052 +. 0572 )] = 0.05
2 2
1⁄
1 2
𝜎̃𝑐 = [ (𝜆1 + 𝜆2 )]
2
. 05
= 0.9
. 057

𝜎
Case 2: 𝑓𝑜𝑟 0.6 < 𝜎 𝑚𝑖𝑛 < 0.8 , 𝑡ℎ𝑒𝑛
𝑚𝑎𝑥

𝜎𝑐 = 0.522𝜎𝑚𝑖𝑛 + 0.477𝜎𝑚𝑎𝑥
Case 3: for 0.2 ≤ σmin/σmax < 0.6 then
𝜎𝑐 = 0.5(𝑎 + 𝑏)
This would yield a slightly larger value of σc.

Example 4.7
The position of point A is determined by the radial distance r=100 m with σ r = 0.5 and the azimuth angle α = 60o
with σα= 30’. Compute the rectangular coordinates x and y and the associated covariance matrix for point A.
Assume r and α to be uncorrelated. Then calculate the semimajor and semiminor axes of the standard error
ellipse and its orientation.

Solution
x =r sin α = 86.60
y = r cos α = 50.00

𝜕𝑥 𝜕𝑥
[𝜕𝛼 𝜕𝑟 ] = [ 𝑟 cos 𝛼 sin 𝛼
]=[
50 0.866
]
𝜕𝑦 𝜕𝑦 −𝑟 sin 𝛼 cos 𝛼 −86.6 0.5
𝜕𝛼 𝜕𝑟

The co-variance matrix of the known variables r and α is a diagonal matrix because they are uncorrelated. It is
equal to

𝜎𝛼2 0 0.00872 0 ]
∑ =[ 2] = [
0 𝜎𝑟 0 0.52

The co-variance matrix of the Cartesian coordinates is



50 0.866 0.7569𝑥10−6 0 ] [ 50 −86.6 0.3767 −0.2195
∑ =[ ][ ]=[ ]
𝑥,𝑦 −86.6 0.5 0 0.25 0.866 0.5 −0.2195 0.6300

The characteristic polynomial is

0.3767 − 𝜆 −0.2195
[ ]=0
−0.2195 0.6300 − 𝜆

(0.3767 − 𝜆)(0.6300 − 𝜆) − 0.21952 = 0

𝜆2 − 1.0067𝜆 + 0.189141 = 0

From which the two roots are:

λ1= 0.76 λ2=0.25

The semi-major axis is a=√𝜆1 =0.87 m and the semi-minor axis is 𝑏 = √𝜆2 = 0.50 𝑚. To get the orientation of
the semi-major axis,
GdE 416 4: Error Propagation page 11

2𝜎𝑥𝑦 −2(0.2195) −0.4390


tan 2𝜃 = = = = 1.732
𝜎𝑥2 − 𝜎𝑦2 0.3767 − 0.6300 −0.2533

Since both sin 2θ and cos 2θ are both negative, 2θ is in the third quadrant and θ= 105o. (second quadrant).

a
b

Error ellipse for rectangular coordinates

The semi-minor axis is along the extension of r and equal to σ r = 0.5 m and the semi-major axis is oriented
normal to r and is equal to rσθ=100(0.0087)=0.87 m. Therefore the error ellipse axes always are oriented in the
direction where the variables are uncorrelated, as in this case r and α. Along any other pair of directions the
variables are correlated, as in the case of x and y, for which the correlation coefficient ρxy may be computed as
−0.2195
𝜌𝑥𝑦 = = 0.45
(0.6138)(0.7937)

For this example σmin=b=0.5 m and σmax = a = 0.87 m and the ratio is σmin/σmax = 0.57. The radius of the
standard error circle may be computed as

σc=0.5(a + b) = 0.5(0.5 + 0.87) = 0.685 m

If the ratio σmin/σmax is not below 0.2 then the advantage is that there is no need to compute the eigenvalues,
and the equation produces a slightly larger error circle.

Example 4.8
For uncorrelated variables r, λ and φ construct and error ellipsoid for coordinates x,y,z.

Z
φ = 30° σφ = 2’
λ = 60° σλ = 2’ P(x,y,z)
r = 100 m σr = .05 m
r
z
φ
Y
λ
x
X y

Solution.

x = r cos φ cos λ
y = r cos φ sin λ
z = r sin φ

The Jacobian matrix is:


GdE 416 4: Error Propagation page 12

𝜕𝑥 𝜕𝑥 𝜕𝑥
𝜕𝑟 𝜕𝜙 𝜕𝜆
𝜕𝑦 𝜕𝑦 𝜕𝑦 𝑐𝑜𝑠 𝜙 𝑐𝑜𝑠 𝜆 −𝑟 𝑠𝑖𝑛 𝜙 𝑐𝑜𝑠 𝜆 −𝑟 𝑐𝑜𝑠 𝜙 𝑠𝑖𝑛 𝜆 0.433 −25.0 −75.0
𝐽= = [ 𝑐𝑜𝑠 𝜙 𝑠𝑖𝑛 𝜆 −𝑟 𝑠𝑖𝑛 𝜙 𝑠𝑖𝑛 𝜆 𝑟 𝑐𝑜𝑠 𝜙 𝑐𝑜𝑠 𝜆 ] = [ 0.75 −43.3 43.3 ]
𝜕𝑟 𝜕𝜙 𝜕𝜆
𝑠𝑖𝑛 𝜙 𝑟 𝑐𝑜𝑠 𝜙 0 0.5 86.6 0
𝜕𝑧 𝜕𝑧 𝜕𝑧
[ 𝜕𝑟 𝜕𝜙 𝜕𝜆]
The covariance matrix of r, φ, and λ is a diagonal matrix since they are uncorrelated.
 r2 0 0
 
 r =  0   0 
2

0 0  2 

𝛴𝑥𝑦𝑧 = 𝐽𝛴𝑟𝜑𝜆 𝐽𝑇

0.433 −25 −75 . 0025 0 0 . 433 . 75 .5


𝛴𝑥𝑦𝑧 = [ 0.75 −43.30 43.30] [ 0 3.4𝑥10−7 0 ] [ −25 −43.3 86.6]
0.5 86.6 0 0 0 3.4𝑥10−7 −75 43.3 0
0.0026 0.0001 −0.0002
𝛴𝑥𝑦𝑧 = [ 0.0001 0.0027 −0.0003]
−0.0002 −0.0003 0.0032

𝜎𝑥2 𝜎𝑥𝑦 𝜎𝑥𝑧


[𝜎𝑥𝑦 𝜎𝑦2 𝜎𝑦𝑧 ] = 0
𝜎𝑥𝑧 𝜎𝑦𝑧 𝜎𝑧2

Standard deviations are: Co-variance are:


σx=± 0.051 m σxy = 0.0001
σy=± 0.052 m σyz = -0.0003
σz=± 0.056 m σ xz = -0.0002

Correlation Coefficient
This term relates the variances and its covariance to each other

𝜎𝑥𝑦 0.0001
𝜌𝑥𝑦 = = = 0.038
𝜎𝑥 𝜎𝑦 (0.051)(0.052)

𝜎𝑦𝑧 −0.0003
𝜌𝑦𝑧 = = = 0.103
𝜎𝑦 𝜎𝑧 (0.052)(0.056)

𝜎𝑥𝑧 −0.0002
𝜌𝑥𝑧 = = = 0.07
𝜎𝑥 𝜎𝑧 (0.051)(0.056)

4.8 Standard Error Ellipsoid


For a three dimensional space with uncorrelated observables r, λ, and φ and the resulting computed quantities
x,y,z a standard error ellipsoid may form with

a=√λ1, b=√λ2, c=√λ3

where λ1, λ2, λ3 are the eigenvalues of the covariance matrix of the coordinates x,y,z , then the standard error
ellipsoid may be replaced by a standard error sphere
1
𝜎𝑠 = (𝑎 + 𝑏 + 𝑐) 𝑓𝑜𝑟 0.35 ≤ 𝜎𝑚𝑖𝑛 ⁄𝜎𝑚𝑎𝑥 ≤ 1.0
3
or
1⁄2
1
𝜎𝑠 = [ (𝜎𝑥2 + 𝜎𝑦2 + 𝜎𝑧2 )] 𝑓𝑜𝑟 0.9 ≤ 𝜎𝑚𝑖𝑛 ⁄𝜎𝑚𝑎𝑥 ≤ 1.0
3

The spherical concept is not recommended when σmin/σmax is less than 0.35.
GdE 416 4: Error Propagation page 13

Derivatives of Trigonometric Functions


Let u be a differentiable function.

𝑑 𝑑 1 𝑑𝑢
𝑠𝑖𝑛 𝜃 = 𝑐𝑜𝑠 𝜃 𝑠𝑖𝑛−1 𝑢 = ∙ −1<𝑢 <1
𝑑𝜃 𝑑𝑥 √1 − 𝑢2 𝑑𝑥
𝑑 𝑑 −1 𝑑𝑢
𝑐𝑜𝑠 𝜃 = −𝑠𝑖𝑛 𝜃 𝑐𝑜𝑠−1 𝑢 = ∙ −1<𝑢 <1
𝑑𝜃 𝑑𝑥 √1 − 𝑢2 𝑑𝑥
𝑑 𝑑 1 𝑑𝑢
𝑡𝑎𝑛 𝜃 = 𝑠𝑒𝑐2 𝜃 𝑢 ≠ 𝑛 𝜋 ⁄2 𝑡𝑎𝑛−1 𝑢 = 2

𝑑𝜃 𝑑𝑥 1 + 𝑢 𝑑𝑥
𝑑 𝑑 −1 𝑑𝑢
𝑐𝑜𝑡 𝜃 = − 𝑐𝑠𝑐2 𝜃 𝑢 ≠ 𝑛𝜋 𝑐𝑜𝑡−1 𝑢 = ∙
𝑑𝜃 𝑑𝑥 1 + 𝑢2 𝑑𝑥
𝑑 𝑑 1 𝑑𝑢
𝑠𝑒𝑐 𝜃 = 𝑠𝑒𝑐 𝜃 𝑡𝑎𝑛 𝜃 𝑢 ≠ (2𝜋 − 1) 𝜋⁄2 𝑠𝑒𝑐−1 𝑢 = ∙ |𝑢| > 1
𝑑𝜃 𝑑𝑥 |𝑢|√𝑢 − 1 𝑑𝑥
2
𝑑 𝑑 −1 𝑑𝑢
𝑐𝑠𝑐 𝜃 = − 𝑐𝑠𝑐 𝜃 𝑐𝑜𝑡 𝜃 𝑢 ≠ 2𝑛𝜋 𝑐𝑠𝑐−1 𝑢 = ∙ |𝑢| > 1
𝑑𝜃 𝑑𝑥 |𝑢|√𝑢2 − 1 𝑑𝑥

You might also like