Bivariate Normal Distribution

• Joint distribution of two random

variables
• Very useful when dealing with
Bivariate Normal Distribution
planimetric (x,y) positions in surveying
and Error Ellipses
• Density function is a bell-shaped surface
centered at x = µx and y = µy

Bivariate Density Function Bivariate Density Function

• The joint density function of two random variables (X

and Y) which have a bivariate normal distribution is:
  2
 x − µx   y − µy   y − µy   
2
1  −1  x − µx  
f (x, y) = exp  − 2ρ    + 
 
2(1− ρ )  σ x 
     
 σ x   σ y   σ y   
2
2πσ xσ y 1− ρ 2
 

where:
µx and σx = mean and standard deviation of X
µy and σy = mean and standard deviation of Y
ρ = correlation coefficient of X and Y σ xy
ρ=
σ xσ y

Marginal Density Functions Cutting Ellipse/Ellipse of Intersection

• Density functions for X and Y • When a plane parallel to the x,y coordinate plane cuts
• Components of the bivariate normal distribution at the the bivariate density surface at a height K, an ellipse is
X and Y axes formed
• The same as the usual density functions for individual • The equation of this ellipse is: (obtained by making
normally distributed random variables f(x,y) = K)
 1  x − µ  2  2 2

f ( x) =
1
exp  −  x  x − µx   x − µx   y − µy   y − µy 
    − 2ρ     +   = (1 − ρ )c
2 2
σ x 2π  2  σ x    σ x   σx  σ y   σy 
 1  y − µ 2 
1   where:
f ( y) = exp − 
y
   −1
c 2 = ln  4π 2 K 2σ x2σ y2 (1 − ρ 2 )  = a constant
σ y 2π  2  σ y  

1
 Example Solution

−1
The parameters of a bivariate normal distribution c 2 = ln  4π 2 (0.1) 2 (1) 2 (0.5) 2 (1 − (0.5) 2 )  = 2.60
are µx = 4, µy = 5, σx = 1, σy = 0.5, and ρxy = 0.5. (1 − ρ 2 )c 2 = (1 − 0.25)(2.60) = 1.95
A plane intersects the density function at K = 0.1
above the x,y coordinate plane. Evaluate the The equation of the ellipse is:
ellipse of intersection. 2 2
 x−4  x − 4  y −5   y −5 
  − 2(0.5)   +  = 1.95
 1   1   0.5   0.5 

Simplifying,
( x − 4) 2 − 2( x − 4)( y − 5) + 4( y − 5) 2 = 1.95

Error Ellipse Error Ellipse

• Produced when the bivariate probability • The corresponding equation for the cutting
distribution is centered at the origin (µx = µy = 0) ellipse in this case would be:
2 2
  2
 x   y   y  
2
 x   x  y   y 
1  −1  x 
         − 2ρ   +   = (1 − ρ )c
2 2
f ( x, y) = exp    − 2ρ   +    
 2(1− ρ )  σ x   σ x   σ y   σ y   σ σx  σ y  σ y
2
2πσ xσ y 1− ρ 2  x 

This equation is used if we want to represent the
This equation represents a family of error
random errors only ellipses centered on the origin

Standard Error Ellipse The Standard Error Ellipse

• When c=1, we get the equation of the standard

error ellipse:
2 2
 x   x  y   y 
  − 2 ρ     +   = (1 − ρ )
2

σ
 x σx σ y  σy 

• Represents the area of uncertainty for the location

of a control point
• Size, shape, and orientation of a standard error
ellipse are governed by the parameters σx, σy, and
ρ.

2
 Sample variants of the standard error
ellipse (by varying the parameters)
Standard Error Ellipse

• In general, the principal axes (x’ and y’)

do not coincide with the coordinate axes
(x and y)
• The major axis of the ellipse (x’) makes
an angle θ with respect to the x-axis

Positional Errors Orthogonal (Rotational) Transformation

• A positional error is expressed in the • The two vectors can be related by the equation:
x,y coordinate system by the random  X ′  cos θ sin θ   X 
vector  X   Y ′  =  − sin θ cos θ   Y 
Y    
 
• The same positional error is expressed • θ is the angle of rotation
in the x’,y’ coordinate system by the • Transformations from one coordinate system to
the other can be made using the above equation
random vector  X ′
Y′ 
correlated errors may be transformed to
 
uncorrelated errors using the equation

Covariance Matrices Covariance Matrices

• The covariance matrices for the random Recall: Σ yy = AΣ xx AT

vectors are: Applying this to the vector relationship, we get:
X   σ x2 σ xy  σ x2′ 0   cos θ sin θ   σ x2 σ xy  cos θ − sin θ 
Y     2 
=  
 0 σ y ′   − sin θ cos θ  σ xy σ y2   sin θ cos θ 
  σ xy σ y2 
σ x2′ 0  Solving the matrix:
 X ′
Y′   2  σ x2′ = σ x2 cos 2 θ + 2σ xy sin θ cos θ + σ y2 sin 2 θ
 0 σ y′ 
Eq. 1
 
σ y2′ = σ x2 sin 2 θ + 2σ xy sin θ cos θ + σ y2 cos 2 θ Eq. 2

X’ and Y’ are uncorrelated (they are the 0 = (σ y2 − σ x2 ) sin θ cos θ + σ xy (cos 2 θ − sin 2 θ ) Eq. 3
principal axes of the ellipse)

3
 Orientation of Error Ellipse Semimajor and Semiminor Axes

sin 2θ The first two equations (Eq. 1 and Eq. 2) become:

But sin θ cos θ = and cos θ − sin θ = cos 2θ
2 2

2 1/ 2
σ x2 + σ y2  (σ x2 − σ y2 ) 2 
σ x2′ = + + σ xy2 
The third equation (Eq. 3) therefore becomes: 2  4 
1/ 2
1 2 σ x2 + σ y2  (σ x2 − σ y2 ) 2 
(σ y − σ x2 ) sin 2θ + σ xy cos 2θ = 0
2 σ y2′ = − + σ xy2 
2  4 
Which can be further simplified as:
2σ xy
tan 2θ = σx’ = semimajor axis
σ x2 − σ y2
σy’ = semiminor axis

Standard Error Ellipse Standard Error Ellipse

Example: Solution:
The random error in the position of a survey σ xy = ρσ xσ y = ( 0.8 )( 0.22 )( 0.14 ) = 0.0246 m 2
σ x2 + σ y2 ( 0.22 ) + ( 0.14 )
2 2
station is expressed by a bivariate normal = = 0.0340 m 2
2 2
distribution with parameters µx = µy = 0, σx
( )
1/ 2
 (σ 2 − σ 2 ) 2
1/ 2
 0.22 2 − 0.14 2 2

 ( ) ( )

= 0.22 m, σy = 0.14 m, and ρ = 0.80.  x y
+ σ xy2  =
2
+ ( 0.0246 )  = 0.0285 m 2
 4  4
Evaluate the semimajor axis, semiminor axis,    
and the orientation of the standard error  (σ 2 − σ 2 ) 2
σ x2 + σ y2 
1/ 2

+ + σ xy2 
x y
ellipse associated with this position error. σ x'2 = = 0.0340 + 0.0285
2  4 
 
σ x' = 0.0625 m
2 2

Standard Error Ellipse Probability of Error Ellipse

1/ 2
 (σ 2 − σ
σ x2 + σ y2 )
2 2 
− + σ xy2  Assumption: Independent (Uncorrelated) random errors X
x y
σ y'2 = = 0.0340 − 0.0285
2  4  and Y.
 
σ y'2 = 0.0055 m 2
σ x' = 0.0625 = 0.25 m 2 2 2 2
 x   x  y   y  X  Y 
  − 2ρ   +   = (1 − ρ )c   +   = c
2 2 2
 
σ y' = 0.0055 = 0.074 m σx  σx  σ y  σy  σx  σy 
2σ xy 2 ( 0.0246 ) Given a point whose position is defined by the random
tan 2θ = =
σ x2 − σ y2 ( 0.22 ) 2 − ( 0.14 ) 2 errors (X and Y), the point will lie within the error
tan 2θ = 1.711
ellipse if:
2 2
X  Y 
2θ lies in the 1st quadrant since σxy and σx2 - σy2 are both   +   ≤ c
2

σx  σy 
positive. Therefore, 2θ = tan-1(1.711) = 59.7° and the
orientation of the error ellipse is θ = 29.8°

4
 Probability of Error Ellipse Probability of Error Ellipse

The random variable U such that:

2 2
For two degrees of freedom, the probability density
X  Y  function of U can be derived:
U =  + 

σx  σ y 
1
has a chi-square distribution with two degrees of freedom. f ( y) = y ( n / 2)−1e− y / 2
2 Γ( n / 2)
n/2

1
(Recall from the past lecture that Y has a chi-square f (u ) = 2 / 2 y (2 / 2) −1e − u / 2
distribution: 2 2 Γ(2 / 2)
Y = Z12 + Z 22 + ⋯ + Z n2

where Zn are independent standard normal random 1

f (u ) = e −u / 2 for u>0 Probability Density Function of U
variables) 2

Probability of Error Ellipse Probability of Error Ellipse

Therefore, the probability that the position given by X and Example:

Y lies on/within the error ellipse is: The random error in the position of a
 X  2  Y  2  survey station is expressed by a bivariate
P   +   ≤ c 2  = P U ≤ c 2  normal distribution with parameters µx =
 
 σ x   σ y  
  µy = 0, σx = 0.22 m, σy = 0.14 m, and ρ =
c2
1 0.80. Evaluate the semimajor and
∫2e
−u / 2
= du
0
semiminor axes of the error ellipse
within which it is 0.90 probable that the
 X   Y 2

2
 error in position will lie.
P   +   ≤ c  = 1 − e
2 − c2 / 2
Probability
 σ x   σ y  
 

Probability of Error Ellipse

Solution:

P U ≤ c 2  = 0.90
2
The semimajor and semiminor
1 − e−c /2
= 0.90 axes are:
ln [ 0.10 ] = ln e − c  cσ x' = 2.146 ( 0.25 ) = 0.54 m
2
/2
 
cσ y' = 2.146 ( 0.074 ) = 0.16 m
−c 2
−2.3026 =
2
4.6052 = c 2
c = 2.146