General Least Squares

• Can handle combined

observations and parameters in
General Least Squares Adjustment the condition equations
(Curve Fitting) • No restriction on the number of
observations per equation

General Least Squares General Least Squares

• May be used for correlated • In the form:
observations with unequal
precision
• An iterative process Av+B∆= f
• Best used in solving curve-fitting
problems and coordinate-
transformation problems
Av = f v+B∆= f
Adjustment of Adjustment of Indirect
Observations Only Observations

Derivation of the GLS Derivation of the GLS

• Given a mathematical model, the • If we want to include unknown
minimum number of observations for parameters into the adjustment, we
its unique determination is n0 have to write an additional condition
• When n measurements for the model equation for each parameter
are acquired, such that n>n0, the • If u parameters are to be included,
redundancy is: the number of condition equations
r = n – n0 will be:

for n observational variables, c=r+u
there exist r independent condition
equations

1
 Derivation of the GLS Derivation of the GLS
• Lower limit: • Therefore,
-- u is equal to zero (u = 0)
-- number of condition equations is
equal to redundancy (c = r)
Lower Limit
0 ≤ u ≤ n0 Upper Limit

• Upper limit:
-- u is equal to n0 (u = n0)
-- number of condition equations is:
Adjustment of
Observations
r≤c≤n Adjustment of
Indirect
c = r + n0 = n Only Observations
 maximum number of
conditions is equal to the number of
observations

Derivation of the GLS The Concept of Weight

• In many problems, however, the 1 The smaller the variance,
precision α
σ2 the higher the precision
number of parameters in the
adjustment is neither zero nor n0

the two special cases will not 1
weight α
be directly suitable σ2

we will need to use the
GENERAL LEAST SQUARE Where:
TECHNIQUE σ 02
weight = W = σ02 = “reference variance”
σ2 = value that will make
the largest weight equal to 1

The Concept of Weight The Concept of Weight

 w1 0 0  σ 02 / σ 12 0 0 
  W = σ 02Σ −1
W =0 w2 
0 = 0 σ0 / σ2
2 2
0 
 
 0 0 wn   0 0 σ 02 / σ n2 
σ 12 0 0
 
Σ = covariance matrix =  0 σ 2 0 
2

1 / σ 12 0 0  0
   0 σ n2 
W =σ  0
2
0 1 / σ 22 0 
 0 0 1 / σ n 
2

1
Q= Σ Cofactor matrix (Q)
σ 02

2
 Example Example
• For any of the points, the general
A straight line, y = ax + b, must be fitted
through three points. The following data are
equation of the straight line is:
given: F = ( y + v y ) − a ( x + vx ) − b = 0
Point X (cm) Y (cm) σx2 (cm2) σy2 (cm2)
1 2.00 3.20 0.04 0.10
non-linear in the unknowns!
2 4.00 4.00 0.04 0.08
least squares adjustment
3 6.00 5.00 0.04 0.08 cannot be used directly
Find the least square estimates of the
equation needs to be
parameters a and b, using GLS. linearized first…

Example Example
• The linearized form of the condition • In matrix form:
equation is:

∂F ∂F ∂F ∂F  vx  δ a 
( y − a0 x − b0 ) + vx + vy + δa+ δb = 0 [ −a0 1]   + [ − x −1]   = (− y + a0 x + b0 )
∂x ∂y ∂a ∂b v
 y δ b 

(−a0 )vx + (1)v y + (− x)δ a + (−1)δ b = (− y + a0 x + b0 )

Example Example
• The linearized equations (matrix • The initial approximations for a0 and
form) for all three points can be b0 can be obtained by using the
written as: equation of the straight line that
 vx1 
passes through any two of the three
  points.
 v y1 
 − a0 1 0 0 0 0    − x1 −1  ( − y1 + a0 x1 + b0 )  • Using points 2 and 3 for instance,
 0 vx2 δ a 
 0 −a0 1 0 0   +  − x2 −1   =  (− y2 + a0 x2 + b0 ) 
vy  δb
 0 0 0 0 − a0 1   2   − x3 −1    (− y3 + a0 x3 + b0 )  y2 = a0 x2 + b0
 vx3 
 
 v y3  y3 = a0 x3 + b0

This is in the form: Av + B∆ = f

3
 Example Example
• Solving for a0 and b0, • The initial matrices are therefore,
 −0.5 1 0 0 0 0
y − y2 5.00 − 4.00
a0 = 3 = = 0.500 A =  0 0 −0.5 1 0 0 
x3 − x2 6.00 − 4.00
 0 0 0 0 −0.5 1 
b0 = y2 − a0 x2 = 4.00 − 0.500(4.00) = 2.00
 −2 −1  −3.20 + 0.5(2.00) + 2.00   −0.20 
B =  −4 −1 f =  −4.00 + 0.5(4.00) + 2.00  =  0 
 −6 −1  −5.00 + 0.5(6.00) + 2.00   0 

Example Example
• The covariance matrix is: • From the general law of propagation
σ x21 0 0 0 0 0  of variances and covariances, the
 
0 σ y2 1
0 0 0 0  covariance matrix is transformed to:
 
0 0 σ x2 0 0 0 
Σ= 2

0

0 0 σ y2 2
0 0 

 0.11 0 0 
0 0 0 0 σ x2 0  
Σ e = AΣA =  0
t
0.09 0  cm 2
 
3

 0 0 0 0 0 σ y23 
0.04 0 0 0 
0 0
 0 0 0.09 
 0 0.10 0 0 
0 0

 0 0 0.04 0 0 0  2
Σ=  cm
• If the reference variance is σ 0 = 0.09
2
 0 0 0 0.08 0 0 
 0 0 0 0 0.04 0 
 
 0 0 0 0 0 0.08 

Example Example
• The equivalent weight matrix is To solve for the unknowns, we use the
therefore, formula:

We = σ 02Σ e−1 ( B W B) ∆ = B W f
T
e
T
e

 1 
 0.11 0 0 
  0.818 0 0 
We = (0.09)  0

1
0.09
0  =  0

1 0  N∆ = t
   0 0 1 
 0 1  
0
 0.09 
TRAINING CENTER FOR APPLIED GEODESY AND PHOTOGRAMMETRY
COLLEGE OF ENGINEERING, UNIVERSITY OF THE PHILIPPINES, DILIMAN

4
 Example Example
• The solved matrices are therefore, • The corrected parameter values are:

55.272 11.636 
N =
11.636 2.818 
 a1 = a0 + δ a = 0.5 − 0.0482
0.3272 
t= 
b1 = b0 + δ b = 2.0 + 0.2571
0.1636 
 0.1384 −0.5715 a1 = 0.4518
N −1 =  
 −0.5715 2.7147  b1 = 2.2571
 −0.0482 
∆ = N −1t =  
these are only the results from the 1st
 0.2571  iteration.

Example Example
After second iteration: To calculate the adjusted coordinates, we
use the derived equation for the residuals v:
δa= 3.58242E-05
δb= -0.000191062
v = QAtWe ( f − B∆ ) *
a= 0.451822
b= 2.256952
The equation of the line therefore is:
* For reference, you may read the derivation in the book
“Analysis and Adjustment of Survey Measurements”,
y = 0.4518x+2.2700 Mikhail and Gracie. Von Nostrand Reinhold: 1981.
pp. 242-246

Example Example
Adjusted coordinates of the three points:
 0.01  2.00  0.01   2.01
− 0.04 3.20 − 0.04 3.16 
       
POINT x(cm) y(cm)
 − 0.01 ^ 4.00  − 0.01 3.99  1 2.01 3.16
v=  l =l+v= + =   2 3.99 4.06
 0.06  4.00  0.06   4.06
 0.01  6.00  0.01   6.01 3 6.01 4.97
       
5.00  − 0.03 4.97 
− 0.03
Verify if these adjusted positions lie on
the line y = 0.4518x+2.2570