Part II: Survey Observations and Adjustments

2.1 Introduction

While the observations made in a survey are important, it is more important to analyze the results
made from that survey. The analysis can be superficial in that the surveyor looks and makes a
judgement on the observations or it can be elaborate based on statistical principles from a least
squares adjustment. This latter approach to analysis is the method that will be used for MDOT
projects and contracts. The main advantage of least squares lies in the ability to statistically reach
a conclusion as to the quality of the observations within a survey and to test to see if these
measurements meet recognized standards in the profession.

2.2. Planning and Design of the Survey

One of the most important aspects of any survey project is the planning and design of that survey.
With the advent of the global positioning system (GPS), planning has become more common place.
The idea behind planning is to optimize the geometry of the survey thereby increasing the resulting
accuracy of that survey. For example, in conventional traversing it is important to stay away from
“thin” or “skinny” triangles. These kinds of geometric figures create weaknesses within the solution.
Small errors in the triangles become magnified in the adjustment process. The same situation applies
to GPS. Here the geometry involves the location of the satellites in the sky. As shown in Part V
(10.8), poor satellite geometry can result in a dilution of precision that is unacceptable for the GPS
survey. Figure 2.1 shows good GPS satellite geometry whereas Figure 2.2 shows poor geometry
since all of the satellites are located within one quadrant of the sky creating thin angles at the
receiver station.
Figure 2.1
Example of good satellite geometry

2.1
 Figure 2.2
Poor satellite geometry.

Planning the survey also involves identifying the land owners upon whose land the survey will be
performed. Prior to the execution of any survey, these individuals must be contacted and permission
obtained to access survey points on their property. Try to identify the location of survey points to
be as unobtrusive as possible and situated in locations that minimize damage to the owners property.
Make sure that the owners know when the survey will be performed and the purpose for the survey.

2.3. Pre-Analysis

The ability to perform pre-analysis on a particular survey is a very helpful tool in developing the
necessary specifications to support the accuracy needs of the survey. By careful planning, apriori
estimates of the measured quantities can be determined and then, through error propagation, initial
results based on that survey design, can be determined. The pre-analysis will help identify
weaknesses in either the geometry or in the instrumentation/survey techniques that are applied to the
survey.

When using mixed survey equipment for a survey, it is important that the precision of those tools
be compatible to each other. For example, when using an EDM with a theodolite the relative
precision of both instruments must be consistent with each other in order to achieve the desired

2.2
accuracy for that particular survey. For example, Table 2.1 shows the allowable angular error for
a given linear accuracy. Table 2.2 gives a comparable example identifying the allowable linear error
for a given angular error.
Table 2.1
2.4. Observations and Errors Allowable Angular Error for a
Given Linear Error
It is a well recognized fact that all
measurement, except counting, contain errors. Accuracy of Linear Allowable Angular
These are inherent in any measurement Measurements Error
because of residual errors in the manufacture
of the instruments, human observations, and 1:500 6' 53"
environmental conditions. A well designed 1:1,000 3' 26"
survey will eliminate blunders and mistakes,
account for any systematic errors within the 1:5,000 0' 41"
measurement process, and minimize the effects
1:10,000 0' 21"
of random errors that will still exist after
blunders and systematic errors have been 1:50,000 0' 04"
either eliminated or corrected.
1:100,000 0' 02"
2.4.1. Blunders 1:1,000,000 0' 00.2"
Blunders are mistakes. They must be
e l i m i n at ed from t h e
observational set by re- Table 2.2
observation. Blunders Allowable Linear Error for a Given Angular Error
cannot be corrected for
because they are not errors. Least Allowable Linear Error In
Typical blunders include Reading of
writing down the wrong Instrument 100' 500' 1,000' 5,000'
number for an observation
5' 0.145 0.727 1.454 7.272
such as recording an angle of
48b 51' 25" when the 1' 0.029 0.145 0.291 1.454
instrument operator actually
read 84b 51' 25". Blunders 30" 0.015 0.073 0.145 0.727
are caused by carelessness 20" 0.010 0.049 0.097 0.485
and sloppy field procedures.
A common blunder found 10" 0.005 0.024 0.049 0.242
today with GPS is the
erroneous recording of the 5" 0.002 0.012 0.024 0.121
height of the receiver. 2" 0.001 0.005 0.010 0.048
Blunders can be minimized
by establishing field 1" 0.002 0.005 0.024
procedures in which checks

2.3
on the measurements are incorporated within the survey procedure. These can include very simple
tasks like recording the height of the receiver in both meters and feet and checking these
measurements before making the observations, or having the note keeper repeat the measurement
values back to the observer before writing them in the notebook.

2.4.2. Systematic Errors

Systematic errors are those errors caused by either the instrument, operator, or environment that
affect a survey measurement. They follow some known physical law and their effects can be
corrected for in subsequent processing of the data. For example, in taping, systematic errors can
occur in the tape itself in that the nominal length is different than the calibrated length. While a 30
meter tape may be graduated throughout, careful calibration may show that the actual tape length is
29.989 meters. Thus, every tape length results in an error of -0.011 meters. Subtracting the error
from the measurement, or adding a correction, results in the correct length. The tape can also be
affected by the environment. Tapes expand and contract as the temperature changes. Thus, if the
tape is used when the temperature is 5bC, each tape length will again be too short. The correction
can be found by taking into account the difference in the field temperature and the standard
temperature and applying this to the coefficient of thermal expansion. The tape can also be affected
by the way the operator uses that tape in the field measurement. If the distances were measured
along the slope of the ground, then the measurements must be reduced to the corresponding
horizontal length. If the tape was suspended above the points in the measurement then the correction
for the effects of sag must be considered in the reductions. Finally, if the applied tension was
different than the standard tension this also results in an erroneous measurement. The key thing in
all of these examples is that there exists mathematical formulas in which the actual tape
measurements can be corrected for in subsequent reductions.

Another principle should be evident from the example of taping and this is that systematic errors
have a tendency to accumulate. All instruments and measurements have the opportunity to exhibit
systematic errors. The goal of the surveyor is to ascertain if a systematic error exists, determine its
magnitude, and correct the measurement accordingly.

2.4.3. Random Errors

After all of the blunders have been eliminated and the systematic errors accounted for, error still
remains. This error is referred to as random error. It exists because of the limitations in the
manufacture of the equipment, inconsistencies in repeated observations, and inappropriate modeling
of the environmental effects on the survey process. Lets look to the taping example to see the effects
of random error. In Figure 2.3(a) we see a portion of a tape graduated to 0.01 units with the mark
to be measured somewhere between 51.28 and 51.29. Because it looks to be in the middle, most
surveyors would record this as 51.28. In Figure 2.3(b), the same distance is now measured with a
tape graduated to 0.002 units. Here the measurement would be recorded as 51.284 units. Another
tape, graduated to smaller units would yield similar results. The difference of 0.004 units is a
random error due to the resolution of the measuring tool.

2.4
 Figure 2.3
Using two tapes to show the effects of random error.

Random errors are characterized as equally likely to be positive or negative. They also have a
tendency to cancel out their effects. For example, if the mark in Figure 2.3(a) was slightly to the
right, then the error would have been of opposite sign. It is the random errors that are evaluated in
the analysis of a survey. Random errors are unavoidable. While a careful set of measurement
procedures can be implemented that minimize their effect, these errors can never be completely
eliminated. The magnitude of these errors is unknown but they can be estimated through experience
or subsequent analysis. Random errors follow the laws of probability and compensation in that large
errors rarely occur, small errors are more likely than large errors, and there is an equal opportunity
of positive or negative error to occur.

2.5. Accuracy and Precision

Two measures of quality of a series of survey measurements are accuracy and precision. Although
many use these terms synonymously, they are different as shown in the targets depicted in Figure 2.4.
Accuracy is how close a set of observations are to the “true” value. For example, in Figure 2.4(a),
accuracy may be high when one considers all of the shots on the target. The average value may be
right in the center of the target. Precision is a measure of the repeatability of a set of measurements.
For example, Figure 2.4(b) shows that the difference in the 8 shots on the target is very small. In
other words, there is a high level of repeatability in each of the shots that were fired. Yet, none hit
the center of the target therefore while the precision was high, the accuracy was not. Finally, Figure
2.4(c) shows the same target with both high precision and high accuracy.

2.5
 Figure 2.4
Targets showing the relationship between accuracy and precision.

2.6. Variance, Covariance and Correlation

Analysis of the measurements made in a survey is done by looking at redundant observations. This
is very important because without these extra observations, no checks are possible within the survey.
Lets look at a very simple example, a triangle. It is well known that three elements of a triangle must
be known to solve for the other unknown elements. Thus, if two angles are known, the third can be
found by subtracting the sum of the two known angles from 180b. But, there is no guarantee that
the angle is correct. It only follows the basic rules of Euclidean geometry. On the other hand, if the
third angle is also measured then there is a check on the condition that the sum of the angles must
equal 180b. This is redundancy. Surveyors try to apply checks whenever possible. Angles and
distances are measured multiple of times to ensure that the measurements are “good”. Traverses and
level loops should form closed figures, closing on the beginning point or another control/bench
mark. Again, this provides redundancy and a basic check on the survey results. Redundancy is also
referred to as degrees of freedom. Using the triangle example, measuring all three angles (along with
one of the distances) yields one degree of freedom, or one redundant observation.

Analysis begins by looking at the errors that exist in a series of measurements. Theoretically, error
is the difference between an observation and its true value. Unfortunately, the true value is not
always known. Therefore, the error is estimated as the difference between the observed value and
the expected value of that measurement. This is often referred to as a residual. Thus, the residual,
v, can be shown to be

vi  xi  x̄

2.6
where: xi is the observed value, and
x̄i is the expected value (such as an average).

The precision of a set of observed data can be defined by the variance. For surveying applications,
it is essential to understand that the measurements represent only a sample. Thus, the sample
variance gives an unbiased estimate of the population and it is defined as:
n
2
Σ vi
n1
S2 
n  1
From this, the standard deviation can be found as the square root of the sample variance. Thus,

n
2
Σ vi
n1
S 
n  1

Lets look at an example. Assume that we have

Table 2.3
taken 25 readings from a 1" theodolite. These
Observations made with a 1" theodolite
are given in Table 2.3. The arithmetic mean of
recorded to the second only.
the observations is found by summing the
observations and dividing this sum by the
30.1 31.2 32.9 31.0 32.7
number of observations taken. Thus,
31.4 33.4 31.7 35.4 30.4
n
35.9 30.9 34.0 31.0 33.1
ˆ xi
i1
x̄  31.9 30.5 32.3 31.8 32.6
n
31.7 35.3 33.0 31.1 33.2
5
808.  32.3
25

The sample variance becomes

(30.1  32.3)2  (32.7  32.3)2 (31.1  32.3)2
S2   2.4917
25  1

Taking the square root results in a standard deviation of S = ±1.58. Note that sometimes the variance
and standard deviation are designated using the Greek symbol σ. Theoretically σ represents the
population standard deviation.

Correlation exists when two or more observations are jointly involved. An excellent example of

2.7
correlation is the use of coordinates. Correlation is measured statistically by using the covariance.
Correlation values range from -1 to +1. When the value is close to +1, then the two measurements
are said to be highly correlated. On the other hand, when the correlation coefficient is close to -1
then the two variables are inversely related, such as one finds with barometric leveling where
elevation is inversely proportional to atmospheric pressure. The correlation coefficient is given as
sxy
ρ̂ 
sx s y
where:
(x  x̄ ) (y  ȳ )
sxy  ˆ
n  1

2 (x  x̄ )2 2 (y  ȳ )2
sx  ˆ sy  ˆ
n  1 n  1

2.7. Error Ellipses

Errors within a survey have a tendency to accumulate through the length of the survey. Lets look
at a simple example of conventional traversing (figure 2.5). A target is centered over the backsight,
A, while the theodolite is centered over another known point B. Centering is done using optical
plummets. The unknown point, C, also has a target centered over its position and the angle is
measured at point B. In this scenario, there are a number of potential sources of error affecting the
measured angle. First, the optical plummets may not be perfectly adjusted resulting in the targets
and instrument being displaced over the point by some small amount. Second, when the instrument
operator turns the angle there are potential errors in both pointing the instrument on the target and
in reading of the angle. Finally, the theodolite has certain inherent errors that will affect the
measured angle. These instrument errors are described in Appendix D. The end result is that the
“true”position of point C actually lies within a circle of uncertainty. Now, if the instrument is moved
to point C and a new point D is subsequently measured, the uncertainty at D contains not only the
uncertainty in the angle measurement at C but also the uncertainty in the position of point C. It is
the surveyor’s responsibility to measure how this error propagates throughout the survey.

A measure of this circle of uncertainty is called the error ellipse. A typical error ellipse is depicted
in figure 2.6. If the standard deviations in the x and y directions are identical then the ellipse will
form a circle. In addition, seldom will the u, v axes of the error ellipse follow the x and y axes of
the coordinate system. The angle t is the rotation angle between the ground coordinate system and
the axis of the ellipse.

2.8
 Figure 2.5
Example of Error Sources in an Angle Measurement

Figure 2.6
Typical error ellipse.

2.9
The semimajor and semiminor axes of the error ellipse can be computed as follows:

1/2
2 2 2 2 2
2 Sx  Sy Sx  Sy 2
Sx    Sxy
2 4
1/2
2 2 2 2 2
2 Sx  Sy Sx  Sy 2
Sx    Sxy
2 4

There are several advantages of the error ellipse in analysis. It provides information on the precision
of the adjusted positions of the stations. In addition, the user can make visual comparisons of the
relative precisions between any stations by looking at the shape of the ellipses, their sizes, and the
orientation of each ellipse. These comparisons, though, need to be tempered in that the constraints
used in the adjustment, the precision of the observations, and the geometry of the survey will all
affect the error ellipse. Moreover, it is general practice to enlarge the ellipse for visual purposes.
Finally, the error ellipses can be used in pre-analysis to help select the equipment and the field
techniques that will be used to complete the survey.

2.8. Error Propagation

As we have seen in the previous section, error propagates through a survey affecting the adjustment
of subsequent points within the survey. Error propagation does not necessarily imply that a least
squares solution to a survey problem is necessary, although it is a convenient byproduct of such an
adjustment. To determine the effects of errors made in a survey, one must propagate that error
through the survey.

Assume that we have a function Z. Then, the law of propagation of variances can be shown to
follow this simple rule:

2 2 2

Z
Z
Z
SZ  S  S    S

X1 x1
X2 x2
Xp x p

The basic assumption in this formula is that no correlation exists between the different measurements
Sx y  0 . Using this basic formulation, some frequently used functions can be defined. The
i i
standard deviation of the sum can be found to be

2 2 2
Ssum  S1  S2    S n

The standard deviation for a series is

2.10
 Sseries  Sx n

Finally, the standard deviation of the mean can be shown to be

2 2 2
1 1 1 nS 2 S
Smean  S  S    S  
n x̄1 n x̄2 n x̄ n n 2
n

In a more general sense, one can define the variance-covariance matrix for a function Z of n
independently measured quantities as ZZ. This will be defined as:

Z1
Z1
Z1
Z1
Z2
Zn

 

x1
x2
xn Sx 2 Sx x  Sx x
x1
x1
x1
1 1 2 1 n

Z2
Z2
Z2
Z1
Z2
Zn
 Sx x S x 2  Sx x 
ˆZZ 
x1
x2
xn
x2
x2
x2
1 2 2 2 n

   
       
Sx x Sx x  Sx 2

Z n
Zn
Z n 1 n 1 n n
Z1
Z2
Zn
 

x1
x2
xn
xn
xn
xn

In matrix notation, this is

T
ˆZZ  A ˆ A

where  is the variance-covariance matrix of the observations. This form does show the general law
of propagation of variances where correlation in the observations may exist.

2.9. Quality Checks/Quality Assurances

After the least squares adjustment has been performed, the next step is to evaluate the results
statistically. Since most survey measurements represent only a small sample of the total population,
analysis of a normal distribution is not the most appropriate measure of the quality of the
performance. Instead, two main statistical distributions are used: Chi-square (χ2) and the t or student-
t distributions.

2.11
2.9.1. Chi Square (χ2)

The χ2 distribution is used to test the sample variance to see if it is in agreement with the population
variance. A confidence region is established in which the sample variance should lie between. The
actual location of the sample variance is based on some specific percentage probability, and the
sample set variance, and the degrees of freedom for the particular problem. It is defined by
(n  1) S 2
χ2 
σ2

To test the variance, a confidence region is created.

df S 2 df S 2
< σ2 <
2 2
χα/2 χ1  α/2

where df is the degrees of freedom and (1 - α) is the confidence interval. Lets look at an example.
An angle was measured 10 times and found to be 42b 12' 14.6" with a standard deviation, s, of ±3.7".
Perform a chi-square test to see if the standard deviation is 2.0" at the 5% confidence level versus
the alternative that it is not 2.0".

Modify the previous equation and look in the χ2 table for the appropriate χ2 values at the 0.975 and
0.025 confidence levels with 9 degrees of freedom. Thus,

2
χ0.975σ2 (2.80) (2.0)2
  1.20
n  1 9
2
χ0.025σ2 (19.0) (2.0)2
  8.44
n  1 9

Since the sample falls within the range, the hypothesis is accepted.

2.9.2. Student-t

The second important statistical test is the t- or student-t distribution which is used to compare the
sample mean to the population mean. This distribution is similar, but more variable or flatter, to the
normal distribution. The t-test is used in surveying because of the small sample of observations that
are made within our surveys. The distribution is shown as:
x̂  µ
t
S/ n
As an example, refer to the theodolite observations that were presented in Table 2.3. Recall that the

2.12
arithmetic mean was 32.3" and the standard deviation was computed to be ±1.58". A total of 25
pointings of the theodolite were made. Then, lets compare this standard error to the population mean
at the 95% confidence level.

We begin by setting up the confidence interval for the t-distribution as

S S
x̄  t α < µ < x̄  t α
2 n 2 n

where: µ is the arithmetic mean for the population and

α is the percentage for the confidence level desired.

Because the t-distribution shows the area under the distribution curve at the ends, the confidence
level is (1 - α) where α in this case is 0.05. Since 25 observations were made, there are 24 redundant
observations which means that the degrees of freedom (df) is 24. Using a table for the t-distribution,
t α  t 0.05  2.064 . Then
2 2

1.58 1.58
32.3  2.064 < µ < 32.3  2.064
24 24

which results in

31.6 < µ < 33.0

This means that the population average will fall, at the 95% confidence level, between 31.6" and
33.0".

If we took just the first column of the data in Table 2.3, we would have the following:

x  32.2
S = ± 2.18"
df = 4

Then,
2.18 2.18
32.2  2.776 < µ < 32.2  2.776
4 4
yielding
29.2 < µ < 35.2

This shows that, with increasing observations, the student-t distribution gets closer to the normal
distribution.

2.13