Error Analysis and Least Squares

CPSD #G100398


Presented by

The Office of Land Surveys

Division of Right of Way and Land Surveys















Developed By:
J eremy Evans, PLS
Psomas, Inc.

And the Office of Land Surveys
Introduction:

The Error Analysis and Least Squares class was under development for over three
years. The reasons for this are many. One was the discussion of the goals of the course.
Offices that used the Star*net least squares adjustment program wanted the course to
focus on the use of the Star*net program. Surveyors with a background in mathematics
saw this as essential knowledge for the modern surveyor, and pushed for an in-depth
classroom course of up to 40 hours.

The final decision was to create a course that would show how statistical theory and
practice could be applied to the daily job of a Caltrans surveyor. The Caltrans Surveys
Manual contains many procedures and standards, but very little background on how those
standards were set (and that is not the function of the Manual). This course is to give that
information to the surveyors who will use those guidelines every day in their job.

Another reason for the development time was the consideration of how to present
the class. Should we hire an outside consultant, or write the course ourselves? Until we
answered the question of goals and course length, no one could begin to even work on the
course materials. If we hired an outside consultant to develop a course to our
specifications, the costs would have been prohibitive. To create our own course from
scratch would take too much time.

J eremy Evan, PLS, had developed a course similar to what we wanted while
working as an instructor at Santiago Canyon Community College in Orange County. We
asked Mr. Evans to present his course to Caltrans personnel. His presentation has become
the basis for this course. We have added more classroom exercises, and tailored examples
to reflect Caltrans procedures, but the course structure is his. By combining his material
and ours, Caltrans was able to save a lot of duplicate effort in developing this course.

This is not a course in Star*net. Caltrans has purchased over 100 copies of the
Star*net program, and it is the most common least squares program available throughout
the state. It is also simple to learn, so the instructor doesnt have to spend a lot of time
teaching the software procedures. But the purpose of using the program is to present the
concepts of least squares, not to teach the program. The Star*net manual does have a very
comprehensive tutorial, and anyone who wishes to learn more is encouraged to use it.

The Error Analysis and Least Squares class was not meant to be a background
course. The students who take this class should use what they learn here to quickly
improve the quality of their work through an understanding of the sources of errors
present in all survey work. By utilizing the concepts of instrument calibration,
redundancy, precision ratios and error ellipses, a surveyor should immediately see an
improvement in the quality of their fieldwork. That understanding and application is the
core concept for this course.



1
MEASUREMENT ANALYSIS
AND ADJUSTMENT
Capital Project Skill Development Class
(CPSD)
GRW117
By Jeremy Evans, By Jeremy Evans,
P.L.S. P.L.S. Psomas Psomas
Supplemented by Supplemented by
Caltrans Staff Caltrans Staff
Introduce Instructor
The purpose of this class is to give Caltrans surveyors a clear
understanding of error analysis. This is so the typical surveyor in the
field or office knows how to determine the precision and accuracy
needed to perform a task. This applies not just to control work, but
throughout the lifetime of a project.
Understand that there are three values for any measurement: the
measured value, the adjusted value, and the true value. The true
value can never be known, but a surveyor should know how to
combine proper techniques, strength of figure, and adjustments so
that they are confident that the measured and adjusted values are
close to the true value
Areas of interest - boundary, control and adjustments, design
mapping.
2
Introduction
The dark side of surveying is the
belief that surveying is about
measurements, precisions and
adjustments. It is not and never
will be.
Dennis Mouland
P.O.B. Magazine
July, 2002
All measurements, no matter how accurate, are still subject to Boundary Laws and
common sense.
I once saw a Record of Survey that called 13 monuments out of position, and never
held a single one as good.
Surveying is the ART and Science of measurement! Over-reliance on numbers leads
to the Dark Side, it does.
3
Introduction
Much has been written lately about least
squares adjustment and the advantages it
brings to the land surveyor. To take full
advantage of a least squares adjustment
package, the surveyor must have a basic
understanding of the nature of
measurements, the equipment he uses, the
methods he employs, and the environment
he works in.
Measurement analysis is the first part of this presentation.
Surveyors should have the ability to evaluate the amount of error in their
measurements and / or control the errors in their measurements
An understanding of measurements gives the surveyor this control
This course will present the Least Squares adjustment LAST,
because a surveyor must understand a lot more about errors before
they accept a least squares adjustment.
5
Introduction
Measurements and Adjustments:
War Stories
Discuss projects/situations where misunderstanding of measurements has
caused problems
A level run between benchmarks has an error of 0.25 feet in three
miles. Is that caused by random error that can be adjusted? Or by a
single bad reading that leaves a 0.30 jump in elevation between
two TBMs?
The 100 Hubble Space telescope was ground and polished to the
smoothest finish of any large mirror ever built. Was it any good?
Answer: The mirror was ground to the wrong prescription! Two
shuttle space flights were needed to add corrective lenses.
The surveyors understanding of measurements and datums is
critical to the success of any project.
Especially as projects get bigger in geographical terms. In the first
example above, a bust can be hidden by the relative size of the
project. How do you know when to accept a weak adjustment, and
when to re-measure?
6
Class Outline
Survey Measurement Basics - A Review
Measurement Analysis
Error Propagation
Introduction to Weighted and Least
Squares Adjustments
Least Squares Adjustment Software
Sample Network Adjustments
Star*Net is one of many least squares adjustment software on the market
Terramodel
Intergraph Survey Select Cad
Trimnet or TGO
All of these softwares deliver correct results, however star*net was the
first I used, speaks surveyor, has the greatest flexibility, etc.
The Department is moving to Trimnet from Star*net
Caltrans owns 120 Star*net licenses, so easiest to use as an
example
7
Measure First,
Adjustment Last
Adjustment programs assume that:
Instruments are calibrated
Measurements are carefully made
Networks are stronger if:
They include Redundancy
They have Strength of Figure
Adjust only after you have followed
proper procedures!
Leica 1103 should be turned in for servicing every 18 months. HQ has received
broken units that havent be serviced ever! Thats FOUR years!
Its not only good practice to regularly service the equipment, but cheaper in the long
run.
Are your tribrachs adjusted? Did you check the plummet and level before you
picked up a sight?
Have you cross-tied any control monuments?
Did you turn more sets when you had a weak control scheme, such as straight along
a RR?
Did you avoid as many 180 degree turns as possible?
Every Caltrans Surveyor should know Figure 5.1 !
8
Survey Measurement Basics
A Review of Plumb Bob 101
Introduce books
Adjustment Computation by Wolf and Ghilani- more readable than most
star*net manual, good basics of adjustment theory and star*net
Random Errors chapter of Moffitt. Also the adjustment chapter has a good
section on weighted means
9
Surveying (Geospatial Services?)
Surveying That discipline which
encompasses all methods for measuring,
processing, and disseminating information
about the physical earth and our
environment. Brinker & Wolf
Surveyor - An expert in measuring,
processing, and disseminating information
about the physical earth and our
environment.
If surveyors want to be considered professionals, we need to know the theory
behind our procedures.
10
Measurement
The act or process of determining extent,
dimensions, etc.
The surveyor is an Expert in Measurement.
Websters definition
Remember! Surveying is an ART and SCIENCE.
Accurate chaining with a plumb bob is an Art.
Calibrating an EDM is SCIENCE.
11
Instrument Testing
Pointing error of typical total station
Lets start with everything that you dont know about modern survey instruments.
12
This is a typical brochure of a modern total station.
Before you accept any of the statements as gospel, you must understand exactly
what the manufacturer is telling you.
DIN 18723 is the international testing standards for survey instrument
accuracy. Other DINs may cover such things as food safety or strength of
fishing line.
DIN 18723 sets exact parameters for testing; such as temperature range
during testing, rigidity of setups, and other parameters that can only be
easily performed at the factory. None of theses tests are done while sighting
over AC pavement in 100 weather.
13
Instrument Specifications
Note that the 5601 is a nominal 1 gun, but can be set for readings of 0.1
The 5602, 5603, and 5605 have different accuracies, but can all be read to 1.
What this means is that the 5601 is built differently that the others. Typically, this
would mean that it has sensors that read the digital plate in four places (90
o
apart),
and then takes the average of the four readings.
The others instruments would read the plate in two places (180
o
apart), and then
takes the mean of those reading.
The 5605 is built to the same manufacturing tolerances as the 5602, but just didnt
test as well when checked at the factory.
The standard deviation of 5 means that EACH time you sight the instrument at a
target, the readout has a 68% chance of being within 5 of the true value. The less
precise the instrument, the more you need redundant measurements to achieve
accurate results.
REMEMBER: This is the accuracy of a calibrated instrument the day it leaves the
factory! If you dont keep it calibrated, those numbers dont mean anything!
It also doesnt account for poor work by the operator, such bad readings caused by
parallax.
Use 5 value (far right) for examples
14
Instrument Specifications
The 5602, 5603, and 5605 are all essentially the same instrument! After
manufacture, all instruments are the tested for accuracy (DIN 18723).
Those instruments that have a standard deviation of less than 2 are labeled
5602.
Those that have a standard deviation of more than 2 but less than 3 are
5603s.
The 5605 is built to the same manufacturing tolerances as the 5602, but
just didnt test as well when checked at the factory.
Note: The instrument companies always try to build their equipment to the
higher standard. If you tried to order a 5605 from Trimble, they might tell
you that they dont have any available right now, and arent making any due
to the high demand for the 5602s.
What they really mean is that the factory is doing a great job, and all
instruments are passing the 2 standard.
Use (3mm + 3ppm) value (far right)
Is the ppm value here the same as the ppm value that is dialed into the
instrument dealing with temperature and pressure?
15
Instrument Specifications
Distance Measurement

m
= (0.01 + 3ppm x D)
What is the error in a 3500 foot
measurement?

m
= (0.01+(3/1,000,000 x 3500)) =
0.021
Discuss setting PPM . If you are on the beach in So Cal, you might get away
with setting the PPM to zero. If you arent at sea level and 72, start
calibrating.
Sigma (lower case) denotes standard deviation

m
is Standard deviation of the mean, a measurement of accuracy. More on that
later.
Apply the standards for a 5605.
3 mm x 10,000 = 30m
So a single measurement less that 30m (100ft) will have an precision ration less
than 1/10,000.
Thats why we tie monuments twice!
16
Calibration or Dont shoot
yourself in the foot.
Leica instruments should be serviced
every 18 months.
EDMs should be calibrated every six
months
Tribrachs should be adjusted every six
months, or more often as needed.
Levels pegged every 90 days
The service contract with the Leica suppliers call for 18 mos. service intervals.
Servicing doesnt cost anything, but blunders do!
more often as needed means before a control survey, after being dropped,
or any rainy equipment day.
Using SECO Tribrach Adjusters
Tech Tip Number: 12
Created: January 1, 2001
Optical Plummet Adjustment Using: Tribrach Adjusting Cylinder #2001 or
Tribrach Adjuster #2002 (see illustration 1)

EQUIPMENT NEEDED: Tribrach Adjusting Cylinder #2001 or Tribrach Adjuster #2002, Tripod or
instrument stand, 2 tribrachs, target

1. Place the tribrach on the tripod. Put the adjuster in the tribrach. Place the tribrach to be
adjusted on top of the adjuster so that it is upsidedown looking at the target on the ceiling.
The distance between the tribrach and target should be between 4 and 5 feet.

2. Using the leveling screws of the bottom tribrach, point the crosshair of the tribrach being
tested to coincide with the target.

3. Rotate the tribrach being tested 180 degrees on the adjuster. Crosshairs will stay on the
target of an adjusted tribrach.

4. After 180 degree turn if the crosshair does not stay on the target, half the error should be
corrected with the adjusting screws provided by the manufacturer of the tribrach. The
remainder should be corrected with the leveling screws of the bottom tribrach.

5. Repeat steps 2, 3 & 4 until the crosshair stays on target at all positions.

Tribrach Circular Vial Adjustment Using: Tribrach Adjuster #2002 (see
illustration 2)

EQUIPMENT NEEDED: Tripod or instrument stand, Tribrach Adjuster #2002 and adjusting pins.

1. Place the tribrach on the tripod and fasten to the tripod.

2. Place the #2002 in the tribrach and level the tribrach using the vial on the #2002. Ignore the
circular vial on the tribrach.

3. To level the #2002: Point one end of the #2002 vial to any leveling screw and using that
screw bring the vial to center.

4. Now turn the #2002 90 degrees so that each end of the vial is as close as possible to the
other two leveling screws. Using these two leveling screws, center the vial.

5. Turn the vial 90 degrees back to the original leveling screw and level again if necessary.

6. Repeat 1,2 & 3 until the vial remains centered at both positions.

7. To test the adjustment of the #2002 vial at any centered position, rotate the #2002 180
degrees. The vial should stay centered within one graduation. If not, take half the error back
to the center with the vial mounting screw that is on the high side. 8. If the circular vial on the
tribrach is not centered, use the adjusting screws and bring to center.






17
Is It a Mistake or an Error?
Mistake - Blunder in reading, recording or
calculating a value.
Error - The difference between a measured
or calculated value and the true value.
Discuss true value. For a traverse, there is the measured value, adjusted
value, and true value.
True Value does exist but cannot be measured or known
The best that anyone can do is a mean value or most probable value
18
Blunder
a gross error or mistake resulting usually from
stupidity, ignorance, or carelessness.
Most blunders are caused by human error. If you are lucky, its someone elses error,
not yours. This is why we have specific field techniques, such as double tying
monuments, measuring all HIs, and closing traverses. If all of the procedures are
done properly, then blunders can be isolated and dealt with.
19
Setup over wrong point
Bad H.I.
Incorrect settings in equipment
Blunder
20
Types of Errors
Systematic
Random
Error is the difference between the measured value and the true value.
Its the job of a surveyor to reduce errors to a minimum. But always accept that
there will be minor errors, and not try to fix data that is within tolerance.
21
Systematic
an error that is not determined by chance
but is introduced by an inaccuracy (as of
observation or measurement) inherent in
the system
22
Each measurement made
with the tape is 0.1' shorter
than recorded.
Glass with wrong offset
Poorly repaired chain
Imbalance between level
sightings
Systematic
Systematic errors may be correctable, or may not be.
If the target offset is wrong, a correction factor can be applied.
If foresights and backsights on a level run arent balanced, the results may be poor,
but there is no correction factor, unless you used digital levels.
Systematic Errors are best reduced by following proper procedures.
23
Random
an error that has a random
distribution and can be attributed to
chance.
without definite aim, direction, or
method
True Random Errors are created by the inherent limitations of equipment, operator
skill, and field conditions. They are whats left after blunders have been eliminated,
and systematic errors reduced to a minimum.
24
Random
Poorly adjusted tribrach
Inexperienced Instrument
operator
Inaccuracy in equipment
Many tribrachs have a centering error of +/- 2mm. They dont have to be poorly
adjusted to introduce error. A poorly adjusted tribrach creates systematic error, a
properly adjusted one will still be a source of random error.
All equipment has inherent inaccuracy. Therefore, all measurements will contain
random error.
25
Nature of Random Errors
A plus or minus error will occur with the
same frequency
Minor errors will occur more often than
large ones
Very large errors will rarely occur (see
mistake)
A Normal Distribution Curve has all of these attributes:
1. It is symmetrical about the mean
2. More data is close to the mean that farther away
3. Very little data is found at the fringe
26
Normal Distribution Curve #1
A plus or minus error will occur with the same
frequency, so
Area within curve is equal on either side of the mean
The Normal Distribution curve is also known as the Bell Curve due to its shape.
Its was developed by an 18
th
century German mathematician and astronomer named
Karl Gauss.
If this was a chart of coin tosses, the chance of a coin land on heads is equal to the
number of coins landing tails. And the number of coins landing heads 6 out of 10
times is equal to the number of coins landing tails 6 out of 10 times, etc.
This is a Normal Curve! In real life, the data is often skewed.
27
Normal Distribution Curve #2
Minor errors will occur more often than large
ones, so
The area within one standard deviation () of
the mean is 68.3% of the total
In any population sample, most errors will be close to the mean.
Back to the coin toss example, a lot more tests will show a 6 to 4 split than will
show an 8 to 2 split
The distribution curve and standard deviation are best explained with large
population samples.
If this represented a sampling of coins tossed 100 times, differences within one
standard deviation would be considered minor.
These are just the basics. We will explain the formula for standard deviation later.
In surveying, we use much smaller samples, so the data rarely plots as a smooth
curve.
28
Normal Distribution Curve #3
Very large errors will rarely occur, so
The total area within 2 of the mean is 95%
of the sample population
In the previous slide we mentioned that a small number of data wont result in a
smooth curve.
In a random sampling of a general population, it usually takes a minimum of sample
of about 30 to see a true curve start to form.
With numbers less than thirty, its possible that there wont be any measurements
outside of two standard deviations.
Since surveyors only measure a sample of thirty or more with GPS equipment, ALL
conventional field measurements should fall within that limit.
If we go back to the example of the curve representing 10 coin tosses, the chance of
any person tossing 4 heads in a row is 2
4
(16:1) or 6.25%
The odds of 5 in a row is 32:1, or 3.125%.
So while it is possible to toss heads 10 times in a row (1024:1, or 0.1%),
measurements outside of 2 sigma of the mean arent usually relevant. For
measurement data, that means flawed.
29
Histograms, Sigma, & Outliers
4.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 -3.5 -4.0
Residuals
1 1
Outlier
\
M
E
A
N
2 2
Histogram: Plot of the Residuals
\
1 : 68% of residuals
must fall inside area
2 95 % of residuals
must fall inside area
Bell shaped curve
/
This data shows the precision of a set of turned angles.
A residual is the difference between the measured value and the most likely value
(usually the mean). Thats different from the definition of error, which is the
difference between measured and true. Since the true value isnt known, you cant
calculate error. But a residual is a value that can be calculated and used for
mathematical adjustments.
Notice the Outlier. A bell curve should help identify data that should be excluded
(Blunders)
All data within 2 sigma is significant data, even if it isnt precise.
It still has statistical value, and isnt weak data
In this example, measurements within 1.75 seconds of the mean will happen 68% of
the time. So even very good measurements have a measure of uncertainty.
30
Measurement Components
All measurements consist of two
components: the measurement and
the uncertainty statement.
1,320.55 0.05
The uncertainty statement is not a
guess, but is based on testing of
equipment and methods.
Uncertainty statement is usually a statement of accuracy
In the last slide, the measurement of one standard deviation was +/- 1.75. That was
the uncertainty statement
The second bullet originally read The uncertainty statement is not a guess, but is
based on testing of equipment, personnel, methods and the surveyors judgment.
Whats the difference between the two?
Answer: All properly adjusted equipment, used correctly, has random errors. The
human factors (personnel and judgment) introduce blunders and systematic errors.
The published instrument uncertainty statements (2.0 mm +/- 2 ppm) are the
expected instrument error. The actual field measurements include the systematic
errors beyond the manufacturers control.
31
Accuracy Vs. Precision
Precision - agreement among readings of the
same value (measurement). A measure of
methods.
Accuracy - agreement of observed values
with the true value. A measure of results.
Bullseye example
Bullseye is not a very good example in surveying. A tightly grouped set of
measurements (precision) that misses the Bullseye (accuracy) doesnt help
the surveyor.
Q. If several tightly grouped measurements miss the bullseye, how would
you know?
A. See Standard Deviation of the Mean
32
Measurement Analysis
Determining Measurement Uncertainties
Now well take a data sample and show how the Bell curve applies to
measurements.
33
Determining Uncertainty
Uncertainty - the positive and negative range
of values expected for a recorded or
calculated value, i.e. the value (the second
component of measurements).
34
Your Assignment
Measure a line that is very close to 1000 feet
long and determine the accuracy of your
measurement.
Equipment: 100 tape and two plumb bobs.
Terrain: Basically level with 2 high brush.
Environment: Sunny and warm.
Personnel: You and me.
If the instructor wishes to have the class perform this exercise, see the sample
instructions in the student work book.
Chaining Test Data Exercise

Equipment: 100 steel chain, 2 nails, 2 plumb bobs.

Setup: On level ground lay the chain out flat, and place two nails approximately 100 feet
apart. The site can be on grass, dirt, or pavement, as long as it is level.

Procedure: Have the class form 2-person teams, with each team making a single
measurement of the distance. Both chainmen should use a plumb bob, with the head
chainman holding the chain no more than waist high. If time permits, the trainees can use
a spring balance and thermometer, and adjust for sag and temperature. The tape
corrections would be part of eliminating systematic errors. If corrections for sag or
temperature arent made, students should still be aware of the correction procedures. You
can still use uncorrected measurements for the classroom exercise.

Measurements: At least 10 measurements should be made. If the class has fewer than
20 students (10 teams), then teams may switch off head and rear chainmen until a total of
10 measurements are obtained. There may be more measurements, but for simplicity it
shouldnt be much more than 10, and should be an even number. Each chaining team
should not reveal their results until all measurements have been made.

Calculations: After all measurements have been collected, the student will return to the
classroom, and use the data as shown in the PowerPoint to obtain mean, standard
deviation, and standard deviation of the mean

NOTE: After completing the exercise, DONT try to measure the distance using EDM
equipment! Students should be aware that they will never know the true value. There
are measurements that are close to the mean value, but there is no right answer. Even a
distance measured by modern equipment has its own random errors, and is not the true
value.

35
Planning the Project
Test for errors in one tape length.
Measure 1000 foot distance using same
methods as used in testing.
Determine accuracy of 1000 foot
distance.
36
Test Data Set
Measured distances:
99.96 100.02
100.04 100.00
100.00 99.98
100.02 100.00
99.98 100.00
Need to measure between two points approximately 1000 apart and need
the accuracy of the measurement
Discuss how measurements were made (chain, bobing up to waist high,
etc.)
Objective is to determine error per chain length by testing, then determine
the error in the 1000 distance.
37
Averages
Measures of Central Tendency
The value within a data set that tends to exist at
the center.
Arithmetic Mean
Median
Mode
Measures of Central Tendency is a corollary to the Nature of Random Errors #2.
Mean is the sum of measurements divided by the number of observations.
Median is the midpoint of the observations (half are less, half are greater).
Mode is the most common value.
38
Averages
Most commonly used is Arithmetic Mean
Considered the most probable value
n = number of observations
Mean = 1000 / 10
Mean = 100.00
n
mean
meas.

=
39
Residuals
The difference between an individual reading
in a set of repeated measurements and the
mean
Residual () = reading - mean
Sum of the residuals squared (
2
) is used in
future calculations
Residuals are also called variations. Thats why v is used in the formula
40
Residuals
Calculating Residuals (mean = 100.00):
Readings residual residual
2
99.96 -0.04 0.0016
100.02 +0.02 0.0004
100.04 +0.04 0.0016
100.00 0 0
100.00 0 0
99.98 -0.02 0.0004
100.02 +0.02 0.0004
100.00 0 0
99.98 -0.02 0.0004
100.00 0 0

2
= 0.0048
The determination of mean and the resulting residuals are the beginning
of a least squares adjustment. The sum of the residuals squared should be
smallest where the mean was properly calculated.
41
Standard Deviation
The Standard Deviation is the range
within which 68.3% of the residuals will fall
or
Each residual has a 68.3% probability of
falling within the Standard Deviation range
or
If another measurement is made, the
resulting residual has a 68.3% chance of
falling within the Standard Deviation range.
Standard Deviation is sometimes referred to as standard error.
Dont get the cart before the the horse, the slide makes it sound like the
definition of deviation is 68.3%.
In reality, the formula for Standard Deviation results in a 68.3%
probability, not the other way around.
42
Standard Deviation Formula
( )
1 n
deviation Standard
2

=

' 023 . 0
9
0048 . 0
= =
Sigma (lower case) denotes standard deviation
Sigma (upper case) denotes Summation
Vee (italics) denotes residual, the difference between individual measurements and
mean
n denotes number of measurements
The pure formula for standard deviation would have just n in the denominator,
not n-1
But you cant have a standard deviation from just one measurement. So n-1
represents the number of redundant measurements,
You can make only a single measurement if you wanted to. But you wouldnt be
able to calculate a standard deviation from a single measurement.
Note that the more redundant measurements (n-1) you have, the closer n-1
approaches n.
That is, if n=2, then n-1 is of n. But if n=100, then n-1 is 99% of n.
The more redundant measurements, the more accurate the standard deviation
43
Standard Deviation
Standard Deviation is a comparison of the
individual readings (measurements) to the
mean of the readings, therefore
Standard Deviation is a measure of.
44
Standard Deviation
Standard Deviation is a comparison of the
individual readings (measurements) to the
mean of the readings, therefore
Standard Deviation is a measure of.
PRECISION! PRECISION!
Draw a Bell Curve that is very tall and steep, and compare it to a very low and flat
curve.
Which curve represents a higher precision?
The closer the data are to the mean, the higher precision.
44
Standard Deviation of the
Mean
This is an uncertainty statement regarding the mean
and not a randomly selected individual reading as is
the case with standard deviation.
Since the individual measurements that make up the
mean have error, the mean also has an associated
error.
The Standard Deviation of the Mean is the range
within which the mean falls when compared to the
true value, therefore the Standard Deviation of the
Mean is a measure of .
46
Standard Deviation of the
Mean
This is an uncertainty statement regarding
the mean and not a randomly selected
individual reading as is the case with
standard deviation.
Since the individual measurements that
make up the mean have error, the mean
also has an associated error.
The Standard Error of the Mean is the
range within which the mean falls when
compared to the true value, therefore the
Standard Deviation of the Mean is a
measure of .
ACCURACY!
Draw a Bell Curve that is skewed, with one steep side and one gentle slope.
Is this more accurate than a symmetrical data set?
Q. What happens if you turn three sets of angles instead of two or four?
SEE EXERCISE FOR STANDARD DEVIATION OF THE MEAN
Exercise for Standard
Deviation of the Mean
Slide #39
An instrument man measures an angle
three times.
He gets the following results:

504538
504544
504538

Calculate the Standard Deviation and
Standard Deviation of the Mean for this
of three angles.
(Hint: just use the seconds as whole
numbers)


( )
1 n
deviation Standard
2

=

n
m

= ) ( Mean the of Deviation Standard

Not satisfied with the spread of the
measurements, the instrument man then
turns another set of angles:

504544
504538
504542
504536

Calculate the Standard Deviation and
Standard Deviation of the Mean for the
set of four angles.


Calculation of Standard Deviation

Meas.# Angle Residual V2
1 44 4 16
2 38 -2 4
3 38 -2 4
SUM 24

1 44 4 16
2 38 -2 4
3 42 2 4
4 36 -4 16
SUM 40


Calc Standard Deviation for set #1
alc Stand. Dev. of the Mean for set #1


C

Calc Standard Deviation for set #2

Calc Stand. Dev. of the Mean for set #2






Note that:

1. Both data sets have the same mean
40

2. The first set has a smaller spread than
the second 6 vs. 8, but isnt
symmetrical about the mean




3. The first asymmetrical set has a
smaller standard deviation (precision)
( ) 46 . 3
2
24
#1 set = =





4. The second symmetrical set has a
smaller standard deviation of the mean
(accuracy)



If your observations arent symmetrical,
it is better to take more observations
than guess which ones are better

When turning sets with a total station,
always turn an even number!
0 . 2 ) ( #1 = =
( )
3
m
46 . 3
set
65 . 3
3
40
#2 set = =
82 . 1
4
65 . 3
) ( #2 set = = m
47
Standard Deviation of the
Mean
Distance = 100.000.007
(1 Confidence level)
n
) ( Mean the of Error Standard m

=
' 007 . 0
10
023 . 0
m = =
Back to the chaining example.
For every measurement that you make, there are three values.
The first value is the measured value (in this example, each of the 10
measurements)
The second value is the adjusted value; i.e. the mean.
The third value is the True Value.
The Standard Deviation of the Mean is your confidence in the adjusted
value.
This calculation states I am confident that the true value lies within 0.007
of the adjusted value.
48
Probable Error
Besides the value of =68.3%, other
error values are used by statisticians
An error value of 50% is called Probable
Error and is shown as E or E
50

E
50
= (0.6745)
E
50
is used statistically for things like toxicity studies on lab rats and Air
Force bombing surveys
(The Doc says you got a 50-50 chance!)
49
90% & 95% Probable Error
A 50% level of certainty for a measure of
precision or accuracy is usually unacceptable.
90% or 95% level of certainty is normal for
surveying applications
) 6449 . 1 ( E90 =
) 96 . 1 ( 95 = E
n
E
E
90
m 90 =
n
E
E
95
m 95 =
Must calculate E90 or E95 before calculating E90m or E95m
50
95% Probable Error
Distance = 100.000.015
(2 Confidence Level)
' 046 . 0 ) 023 . 0 2 ( ) 2 ( 95 = = = E
' 015 . 0
10
046 . 0 95
95 = = =
n
E
E m
Note that the last slide showed the multiple for E
95
as 1.960, and this one
is 2. Technically two standard deviations and E
95
arent the same, but the
two terms are often used interchangeably in day-to-day usage. (Day to Day
by mathematicians, I dont know too many surveyors who talk like that)
Method to determine the E95m for each tape length
This now reads, Theres a 95% chance that any measurement of 100.00
feet is accurate within 0.046, and the average of a set of ten measurements
will be within 0.015 of the true value 95% of the time.
Now 0.046 doesnt sound very accurate, and it isnt.
But this is the 95% level. If you chain a distance of 100 feet, any one
chaining is probably (68.3%) within 0.023 feet of true,
But for any one measurement, you are certain that you are within 0.046.
Check the original data set. MOST measurements are within one standard
deviation, but ALL are within E
95.
51
Meaning of E
95
If a measurement falls outside
of two standard deviations, it
isnt a random error, its a
mistake!
Francis H. Moffitt
Were Surveyors, not statisticians. Random Errors that fall outside of E
95
arent
random errors.
Time to re-check you measurements and equipment.
52
How Errors Propagate
Error in a Series
Errors in a Sum
Error in Redundant
Measurement
53
Error in a Series
Describes the error of multiple measurements
with identical standard deviations, such as
measuring a 1000 line with using a 100
chain.
n E Eseries =
The error of a series is equal to the error of each measurement times the square root
of the number of measurements in the series.
Common Sense would say that if one measurement has an error of 0.02, and
you measured 10 lengths, the total error would be 0.20. But because errors can be
positive or negative, the chance of a random error being positive for 10
measurements in a row is remote. That is why the square root give a more reliable
answer.
What is the error in 1000 distance using error value determined before?
Using data from slide 43 and the error of series formula:
=0.015 (10
)
=0.047 = 0.05
Note that we are using the term Error and the variable E rather than Standard
Deviation and sigma.
If we replaced E with sigma, the formula would still be correct
But this formula will also work for E
50
and E
95
as well.
By changing the variable of E , we can calculate any error value we wish.
54
Error in a Sum
E
sum
is the square root of the sum of each of
the individual measurements squared
It is used when there are several
measurements with differing standard errors
2 2 2 2
n 3 2 1 sum E ... E E E E + + + + =
Error in a series and error in a sum are basically the same.
If the variable E is the same for each of the measurements, then the result is the
Error of a Series formula.
If they arent the same value, then you use the Errors of a Sum formula.
55
Exercise for Errors in a Sum
Assume a typical single point occupation. The
instrument is occupying one point, with tripods
occupying the backsight and foresight.
How many sources of random error are there in this
scenario?
Hint: First look at errors that would affect distance, then errors that would affect the
angle.
56
Exercise for Errors in a Sum
There are three tribrachs, each with its own
centering error that affects angle and distance
Each of the two distance measurements have errors
The angle turned by the instrument has several
sources of error, including poor leveling and parallax
The combination of all of the possible random errors exceeds the amount of error
we normally associate with a single measurement
For someone to say this is a half-second gun, or The EDM is accurate to 2mm
ignores all of the other possible error sources
57
Error in Redundant
Measurements
If a measurement is repeated multiple
times, the accuracy increases, even if
the measurements have the same value
n
E
E . meas . red =
If sigma= one (1), and n=1,then one over the square root of 1 = 1
If sigma= one (1), and n=2, then one over the square root of 2 = 0.707
If sigma= one (1), and n=2, then one over the square root of 4 = 0.5
What is error in 1000 distance using error value determined before
=0.015 (10
)
=0.047 = 0.05
Error in redundant measurement is used when a value is measured more
than one time
What is error value when 1000 distance is measured 4 times.
=0.047 4 = 0.024 = 0.02
58
Sample of Redundancy
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
0 20 14 12 10 9 8 8 7 7 6 0 40 28 23 20 18 16 15 14 13 13
1 22 16 13 11 10 9 8 8 7 7 1 42 30 24 21 19 17 16 15 14 13
2 24 17 14 12 11 10 9 8 8 8 2 44 31 25 22 20 18 17 16 15 14
3 26 18 15 13 12 11 10 9 9 8 3 46 33 27 23 21 19 17 16 15 15
4 28 20 16 14 13 11 11 10 9 9 4 48 34 28 24 21 20 18 17 16 15
5 30 21 17 15 13 12 11 11 10 9 5 50 35 29 25 22 20 19 18 17 16
6 32 23 18 16 14 13 12 11 11 10 6 52 37 30 26 23 21 20 18 17 16
7 34 24 20 17 15 14 13 12 11 11 7 54 38 31 27 24 22 20 19 18 17
8 36 25 21 18 16 15 14 13 12 11 8 56 40 32 28 25 23 21 20 19 18
9 38 27 22 19 17 16 14 13 13 12 9 58 41 33 29 26 24 22 21 19 18
10 40 28 23 20 18 16 15 14 13 13 10 60 42 35 30 27 24 23 21 20 19
Shots
Base
Li ne
(KM)
Expected accuracy of a given number of unique observations at a given baseline
length, at the 95% confidence interval and stated in mm.
Shots
Base
Li ne
(KM)
Horizontal Vertical
This is the published accuracy of the Trimble R7 receiver
To apply the Errors of Redundant Measurement formula,
Use the number in column 1 of Horizontal shots as E
95
, In the 10 KM Baseline, that
would be 40mm
Divide that number by the square root of the number of shots (example, 4) and you
should get the same value as the chart.
40 divided by the square root of 4 equals 20. That agrees with the chart.
So four measurements are twice as accurate as a single measurement!
59
Eternal Battle of Good Vs. Evil
With Errors of a Sum (or Series), each
additional variable increases the total
error of the network
With Errors of Redundant
Measurement, each redundant
measurement decreases the error of
the network.
This may be the single most important statement in this entire
course.
As networks become more complex, there is is a greater chance of error.
Also, a blunder can hide in a complex network, by having the error spread
out to more points. At the beginning we had the example of a level network
with 0.10 closure per mile (0.25 in three miles). A single three mile level
run cant isolate a bust. But three one mile loops will show whether you
have poor measurements or poor control.
Always think of redundancy when planning a network.
60
Sum vs. Redundancy
Therefore, as the network becomes
more complicated, accuracy can be
maintained by increasing the number of
redundant measurements
.
Redundancy can mean:
1. Turning more sets of angles with a Total Station. This is very easy with servo
instruments turning rounds in auto mode.
2. Traverses with cross-ties and double stubbing
3. Longer occupations using GPS
4. Multiple occupations of GPS points using different configurations
5. Level runs that use several loops, instead of a single long run between two known
points.
61
Error Ellipses
Used to described the accuracy of a
measured survey point.
Error Ellipse is defined by the dimensions of
the semi-major and semi-minor axis and the
orientation of the semi-major axis
Assuming standard errors, the
measurements have a 39.4% chance of
falling within the Error Ellipse
E
95
= 2.447
See Sample Test Question A-4 1991 LS for error ellipse example
Part A and B (this is in the Caltrans Exam Prep Video and Workbook)
Draw the example problem above. After completing the calculations, draw a
typical error ellipse at point C.
Now draw line B-C. If Point C were to be measured from B, how would
the error ellipses overlap?
Which form of redundancy would be more accurate,
Repetitions of angle BAC?, or
Measuring BAC and ABC?
.
.
.
\
C
P
S
D
\
G
R
W

1
1
7
\
d
r
a
w
i
n
g
s
.
d
g
n


5
/
2
7
/
2
0
0
5

1
:
2
7
:
3
6

P
M
62
Coordinate Standard
Deviations and Error Ellipses
Coordinate Standard Deviations and Error Ellipses:
Point Northing Easting N SDev E SDev
12 583,511.320 2,068,582.469 0.021 0.017
Northing Standard Deviation
{
}
Easting Standard Deviation
This is why the standard errors have only a 39.4% chance of falling within the error
ellipse.
The standard deviations arent oriented the same as the ellipse.
63
Positional Accuracy vs.
Precision Ratio
Or, How good is one error ellipse
compared to all those others?
Older surveys use closure as a measure of accuracy. Newer adjustments dont. How
do you compare the two?
Or to put it another way, How close together can two error elipses be and still have
an accurate survey?
See Attached Document
Positional Accuracy vs. Precision Ratio

Traditional compass rule adjustments were analyzed using precision ratios.
The length of a traverse is divided by the error in closure. The result is the precision ratio.

The standard for a control traverse run to second order accuracy is 1:20,000.
The standard for a landnet traverse run to third order accuracy is 1:10,000.
(See Chapter 5 Surveys Manual)

0.01 ft x 10,000 =100.00

Therefore, any single distance measured to an accuracy of less that 0.01 per 100 ft cannot
meet the 1:10,000 ratio. This is one reason why all landnet points are double-tied.

RTK.

RTK only measures baselines between the base station and the rover. Each measurement
to a monument is independent of measurements to other monuments. The vector between
two unknown stations is never measured. Each is independently measured to a known
base station. This is one reason why RTK can only be used for surveys of third order or
less.


Positional Accuracy

Least square adjustments dont publish precision ratios. Instead, each point is given a
position and an error ellipse, defining the most likely position of the point. A position can
also be defined as the circle in which the true position has a 95% chance of being located.
(E
95
). The question then becomes How do you determine the precision ratio of a
measurement that doesnt have a traverse closure?

The simple way to check for precision ration is to divide the distance between two points
by the sum of the standard errors of the two points.

Errors in a Sum
The Standard Error of the sum of two quantities is equal to the square root of the sum of
the squares of the standard errors of the individual quantities .The concept can be
extended to the sum of any number of quantities that are not correlated.
-Moffitt/ Bouchard

To determine the precision ratio between to monuments:

The ratio between the length of the line and the sum of the errors of the two point.


Y =(Distance)(A +B)
Where A is the positional accuracy at the first station and B is the
positional accuracy at the second station.

1:Y is the resultant precision ratio, where Y shall be greater than or
equal to 10,000 to achieve third order accuracy.

Assume that you locate two monuments using RTK that are approximately 140 m apart.
Each has a positional accuracy of 10mm. What is the precision ratio of the measured
distance between the two monuments?

Y =(140.00)(0.010) +(0.010)
Y =(140.00)0.0002
Y =140.00 0.014
Y =10,000 and 1:Y =1:10,000

OR
Given two RTK monuments at E
95
of 10mm,
The minimum distance between the two monuments that would
achieve a 1: 10,000 ratio would be 140 meters (460 ft.)


Monuments found at distances less than 140 meters (460 ft) apart must be tied using
conventional total station methods to achieve third order standards. Monuments
between 140 and 200 meters apart should be checked for positional error before
being accepted.

The 140 meter standard applies when each monument has been occupied according to
standards (occupied twice for minimum of 15 epochs) and are within a properly boxed
control net. See Surveys Manual Chapter 6

Exercise #2
An EDM with an accuracy () of 2 mm 2.0 ppm is used to measure a distance of 40
meters. The instrument and foresight are on tribrachs with an accuracy of 1.5 mm.
Using the Errors of a Sum formula, calculate the total measurement error. Then calculate
the shortest distance that such a setup could measure a 1: 10,000 precision ratio (land
net), and a 1: 50,000 ratio (project control)


64
Introduction to Adjustments
Adjustment - A process designed to remove
inconsistencies in measured or computed
quantities by applying derived corrections to
compensate for random, or accidental errors,
such errors not being subject to systematic
corrections.
Definitions of Surveying and
Associated Terms,
1989 Reprint
65
Introduction to Adjustments
Common Adjustment methods:
Compass Rule
Transit Rule
Crandall's Rule
Rotation and Scale (Grant Line Adjustment)
Least Squares Adjustment
Compass rule assumes that both angles and distances are measured with equal
precision. The most common way of adjusting metes and bounds descriptions.
The Compass Rule can only solve a traverse, not redundant measurements.
Transit Rule assumes angles are more accurate than distances, but the formula
results in different corrections depending on the orientation of a figure (if you have
a closed traverse, and then rotate it 45 degrees, the adjustment for each leg will
change)
Crandalls rule again assumes angles superior than distances, but is more
complicated than Transit Rule
Rotation and scale holds interior angles as fixed, and adjusts distances. This is the
same as the BLM Grant line Adjustment.
Least Squares simultaneously adjusts the angular and linear measurements to make
the sum of the squares of the residuals a minimum.
If there are no redundant measurements, the results are the same as a Compass Rule.
66
Weighted Adjustments
Weight - The relative reliability (or worth) of
a quantity as compared with other values of
the same quantity.
Definitions of Surveying and
Associated Terms,
1989 Reprint
67
Weighted Adjustments
The concept of weighting measurements to
account for different error sources, etc. is
fundamental to a least squares adjustment.
Weighting can be based on error sources, if
the error of each measurement is different, or
the quantity of readings that make up a
reading, if the error sources are equal.
68
Weighted Adjustments
Formulas:
W (1 E
2
) (Error Sources)
C (1 W) (Correction)
W n (repeated measurements of
the same value)
W (1 n) (a series of
measurements)
Symbol means proportional
Weights are inversely proportional to the residuals. The closer a measurement is
to the mean, the more heavily weighted it should be.
Therefore, corrections are inversely proportional to the weights. The farther a
measurement is from the mean, the more it will be corrected.
Weights are proportional to redundancy. The more times a value is repeated, the
stronger the weight.
Weights are inversely proportional to measurements of a series. A level run of 4
turns is stronger than a run using 8 turns. (All other factors being even)
69
Weighted Adjustments
A
B C
A = 432436, 2x
B = 471234, 4x
C = 892220, 8x
Perform a weighted
adjustment based on the
above data
70
ANGLE No. Meas Mean Value Rel. Corr. Corrections Adjusted Value
A 2 43 24 36
4
/
4
or
4
/
7
4
/
7
X 30 = 17 43 24 53
B 4 47 12 34
2
/
4
or
2
/
7
2
/
7
X 30 = 09 47 12 43
C 8 89 22 20
1
/
4
or
1
/
7
1
/
7
X 30 = 04 89 22 24
TOTALS 17959 30
7
/
4
or
7
/
7
= 30 180 00 00
The relative correction for the three angles are 1 : 2 : 4, the inverse proportion to
the number of turned angles. This is the first set of relative corrections.
The sum of the relative corrections is 1 + 2 + 4 = 7 , This is used as the
denominator for the second set of corrections. The sum of the second set of
relative corrections shall always equal 1. The second set is used for corrections.
The correction to angle C should be one fourth the correction to angle A, and one
half the correction of angle B. This ration is the relative correction factors between
the measurements. This is the first correction factor.
The sum of the relative factors results in the total correction factor for the figure.
The total figure correction factor is then used to correct the measured angles.
71
Weighted Adjustments
BM A
Elev. = 100.0
BM B
Elev. = 102.0
BM C
Elev. = 104.0
+6.2, 10 mi.
+7.8, 2 mi.
+10.0, 4 mi.
BM NEW
This exercise doesnt have a published solution. Instructors may include it as an
exercise, or save time by skipping it.
See Moffitt for a good example of solving this type of problem.
72
Introduction to Least Squares
Adjustment
Simple Examples
73
What Least Squares Is ...
A rigorous statistical adjustment of survey
data based on the laws of probability and
statistics
Provides simultaneous adjustment of all
measurements
Measurements can be individually weighted
to account for different error sources and
values
Minimal adjustment of field measurements
Compass rule adjustment is based on proportional adjustment of data
Simultaneous adjustment of all measurements is the most important
benefit of least squares. In multiple traverses, a compass adjustment must
solve each traverse in order, and hold the results as fixed for the next
traverse. Least squares can solve the entire network simultaneously
Each measurement can have its own error estimate or you can globally
set the error estimate or a combination of the two
Maintains the integrity of the field measurements, least squares tries to
minimize the amount of adjustment to each measurement
74
A Least Squares adjustment distributes random errors
according to the principle that the Most Probable Solution
is the one that minimizes the sums of the squares of the
residuals.
This method works to keep the amount of adjustment to
the observations and, ultimately the movement of the
coordinates to a minimum.
What is Least Squares?
Think of ways that other adjustment methods can skew data.
The Compass Rule adjusts angles based on the length of the legs. But short sights
are less accurate than long ones, so why adjust the long sight more?
A least squares adjustment can take weighted means, redundancy, and strength of
figure to adjust a network.
75
What Least Squares Isnt ...
A way to correct a weak strength of figure
A cure for sloppy surveying - Garbage in /
Garbage out
The only adjustment available to the land
surveyor
>Any survey can be manipulated to pass a least squares adjustment by
freeing up data or changing error estimates
>All adjustments must be reviewed prior to moving on to next step
>A traverse that runs 3 miles along a straight highway is inherently weak
>If you occupy the wrong monument, and dont perform a check shot, least
squares wont help you
>A survey with no redundancy will have the same results whether
adjustment is compass rule or least squares
76
Least Squares
Least Squares Should Be Used for
The Adjustment Of: Collected By:
Conventional Traverse
Control Networks
GPS Networks
Level Networks
Resections
Theodolite & Chain
Total Stations
GPS Receivers
Levels
EDMs
77
Least Squares
What happens?
Iterative Process
Each iteration applies adjustments to
observations, working for best solution
Adjustments become smaller with each
successive iteration
A B
C
D
E
F
G
Observed
1st Iteration
2nd Iteration
.
78
1 Creates a calculated observation for each field
observation by inversing between approximate
coordinates.
2 Calculates a "best fit" solution of observations and
compares them to field observations to compute
residuals.
3 Updates approximate coordinate values.
4 Calculates the amount of movement between the
coordinate positions prior to iteration and after
iteration.
5 Repeats steps 1 - 4 until coordinate movement is no
greater than selected threshold.
Least Squares
The Iterative Process
79
1 Errors
2 Coordinates
3 Observations
4 Weights
Least Squares
Four component that need to be addressed
prior to performing least squares adjustment
80
Errors
Blunder - Must be removed
Systematic - Must be Corrected
Random - No action needed
81
Coordinates
Because the Least Squares process begins by
calculating inversed observations approximate
coordinate values are needed.
1 Dimensional Network (Level Network) - Only
1 Point.
2 Dimensional Network - All Points Need
Northing and Easting.
3 Dimensional Network - All Points Need
Northing, Easting, and Elevation. (Except for
adjustments of GPS baselines.)
82
Weights
Each Observation Requires an Associated Weight
Weight = Influence of the Observation on Final
Solution
Larger Weight - Larger Influence
Weight = 1/
2
= Standard Deviation of the Observation
The Smaller the Standard Deviation the Greater the
Weight
= 0.8 Weight =
1
/
0.8
2 = 1.56
= 2.2 Weight =
1
/
2.2
2 = 0.21
More
Influence
Less
Influence
83
Observational Group
Least Desirable Method
Example: All Angles Weighted at the Accuracy of
the Total Station
Each Observation Individually Weighted
Best Method
Standard Deviation of Field Observations Used as
the Weight of the Mean Observation
Methods of Establishing Weights
Good for combining
Observations from
different classes of
instruments.
Good for projects
where standard
deviation is calculated
for each observation.
84
Least Squares Adjustment Is a Two Part Process
1 - Unconstrained Adjustment
Analyze the Observations, Observations
Weights, and the Network
2 - Constrained Adjustment
Place Coordinate Values on All Points in the
Network
Least Squares
If you remember nothing else about least squares today,
remember this!
85
Also Called
Minimally Constrained Adjustment
Free Adjustment
Used to Evaluate
Observations
Observation Weights
Relationship of All Observations
Only fix the minimum required points
Unconstrained Adjustment
86
Flow Chart
Field Observations
Setup
Observation
Standard Deviation
Field Data
Needs
Editing?
Edit Field Data
Remove Blunders
Correct Systematic
Errors
Perform
Unconstrained Least
Squares Adjustment
No
Analyze
Adjustment
Statistics
Statistics
Indicate
Problems
Modify Input
Data
Constrain Fixed
Control Points
No
Perform
Constrained Least
Squares Adjustment
Print out
Unconstrained
Adjustment Statistics
Print out Final
Coordinate Values
for All Points in
Adjustment
Yes
Yes
Least Square
Adjustment
Software
Decision Step
Performed by
User
Start
Finish
87
Analyze the Statistical Results
There are 4 main statistical areas that need to be looked
at:
1. Standard deviation of unit weight
2. Observation residuals
3. Coordinate standard deviations and error ellipses
4. Relative errors
A 5th statistic that is sometimes available that should be
looked at:
Chi-square Test
88
Also Called
Standard Error of Unit Weight
Error Total
Network Reference Factor
The Closer This Value Is to 1.0 the Better
The Acceptable Range Is ? to ?
> 1.0 - Observations Are Not As Good As Weighted
< 1.0 - Observations Are Better Than Weighted
Standard Deviation of
Unit Weight
89
Observation Residuals
Amount of adjustment applied to observation to
obtain best fit
Used to analyze each observation
Usually flags excessive adjustments (Outliers)
(Star
*
net flags observations adjusted more
than 3 times the observations weight)
Large residuals may indicate blunders
This is the residual that is being minimized
90
Observation Residuals
Site Observation Residual S Dev. Flag
10-11-12 214

33 17.2 1.7 1.2

11-12-13 174

16 43.8 7.2 1.9

*
12-13-14 337

26 08.6 2.1 1.3

4.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 -0.5 -1.0 -1.5 -2.0 -2.5 -3.0 -3.5 -4.0
Outlier
0
91
Coordinate Standard
Deviations and Error Ellipses
Coordinate standard deviations represent the
accuracy of the coordinates
Error ellipses are a graphical representation of the
standard deviations
The better the network the rounder the error
ellipses
High standard deviations can be found in networks with
a good standard deviation of unit weight and well
weighted observations due to effects of the network
geometry
92
Relative Errors
Predicted amount of error that
can be expected to occur
between points when an
observation is made in the
network.
93
Chi-square Test
noun: (ki'skwr) a statistic that is a sum of terms
each of which is a quotient obtained by dividing the
square of the difference between the observed and
theoretical values of a quantity by the theoretical
value
In other words: A statistical analysis of the statistics.
The chi-squared () test.
Consider flipping a coin and getting 7 heads and 3 tails. Now continue until you
have 70 heads and 30 tails. The first sample is reasonably close to your expectation
of a 50-50 split. The second sample is very far off. But the 7:3 ratio is the same.
The test should reveal problems such as this.
(7-5) 5 = 0.8 (70-50) 50 = 8.0
The closer the solution is to zero, the better the network.
Statisticians really only feel comfortable using chi-squared for enumeration data
(such as counting left-handed red-headed people), rather than measurement data
(surveying).
All statistical programs will generate a chi-squared test, but be careful when to
accept it as proof of a good network.
94
Least Squares Examples
Arithmetic Mean
Straight Line Best Fit
95
Least Squares Examples
Straight Line Best Fit
Explain scenario (must be straight line thru points)
This is an example of determining a best fit alignment for a prescriptive
easement.
In a boundary problem, it might help you reject a monument, but best fit
is never to be used as a boundary solution
96
Straight Line Best Fit
Perpendicular offsets:
1 = (0,0)
2 = (100,100)
3 = (200, 400)
This example - Perpendicular offset = 141.421
1: r = 0, r sq. = 0
2: r = 0, r sq. = 0
3: r = 141.421, r sq. = 20,000
Sum r sq. = 20,000
97
Straight Line Best Fit
1: r = 63.246, r
2
= 4,000
2: r = 0, r
2
= 0
3: r = 0, r
2
= 0
Sum r
2
= 4000
98
Straight Line Best Fit
1: r = 0, r
2
= 0
2: r = 44.721, r
2
= 2000
3: r = 0, r
2
= 0
Sum r
2
= 2000
99
Straight Line Best Fit
1, 2 & 3: r = 22, r
2
= 484
Sum r
2
= 3*484 = 1452
This has the lowest Sum r
2
therefore is best result so far
Actual best result is a skewed line that runs 19.9 feet SE of point 1 to 8.4
feet SE of point 3.
100
Least Squares Rules
Redundancy of survey data strengthens
adjustment
Error Sources must be determined correctly
Each adjustment consists of two parts:
Minimally Constrained Adjustment
Fully Constrained Adjustment
Redundancy is a good thing!!
Explain the necessity of two adjustments
A closed traverse that is minimally constrained (one point and bearing held)
should result in a tight closure. If it doesnt, that means that your
measurements were poor.
If you have a good minimally constrained adjustment, then you run a fully
constrain the adjustment (hold all found control monuments as fixed).
If the results are poor, then you know that it is the control that is weak, not
your measurements.
Then you go back to the minimally constrained adjustment, and start adding
one control monument each run, until you can isolated the poor control.
101
Star*Net Adjustment Software
A Tour of the Software Package
Star*Net
1
3
2
6
4
5
102
Sample Network Adjustment
A Simple 2D Network Adjustment
Star*Net
1
3
2
6
4
5
Printout from this adjustment in in appendix
Run adjustment and review printout (unconstrained & constrained)
add mistake to input data and run adjustment
explain how least squares will point to potential mistake (if only one
mistake!)
If inputting data by hand, input one page then run adjustment and
check for errors, input second page and check for errors, etc.
If time permits, do adjustment with GPS vectors
Show the results of traverse (linear precision) in this adjustment
104
Sample Network Adjustments
A 3D Grid Adjustment using GPS and
Conventional Data
0012
224.299
North Rock
0017
209.3
AZDO
0013
205.450
BM-9331
0051
201.018
0052
192.051
SW Bridge
0053
203.046
0018
204.86
0015
188.195
0016
186.655
Star*Net Star*Net
105
Beyond Control Surveys
Other Uses for Least Squares
Adjustments / Analysis
Thinking outside of the box!
106
Questions & Discussion
Statistics Glossary

Error the difference between a measured or computed result and the true value. In
mathematics, errors can be systematic or random. See Residual.

Systematic Errors an error that is not determined by chance but is introduced by
an inaccuracy (as of observation or measurement) inherent in the system. If they are
cumulative, such as temperature corrections for a steel tape, applying correction factors
can compensate for the effects. If they are variable, such as error caused by a poorly
adjusted tribrach, they can be controlled by proper field procedures or calibrations.

Random Errors Often called accidental errors. They are unpredictable errors
that remain after mistakes and systematic errors have been eliminated. They are usually
compensating, and follow the laws of probability. Present in all survey measurements.


Residual () The difference between a measured value and the most probable value,
which is usually the mean. Residuals are similar to errors except that residuals can be
calculated and errors cant, because a true value is never known. All adjustment
calculations therefore use residuals. The symbol is used because residuals are
sometimes referred to as variations.

Variance () The variance is a measure of the range of a set of measurements. It is a
function of the sum of the residuals. Its square root is the standard deviation. The greater
the range of measurements, the larger the standard deviation.

Standard Deviation () A measurement of the precision of a set of measurements. Also
referred to as standard error. In a normal distribution curve, the area within one standard
deviation is 68.27% of the total.

Standard Deviation of the Mean (m) is a measure of accuracy. The mean is the
average of a data set. The Standard Deviation of the Mean is the expected accuracy of the
calculated mean. It gives more weight to a mean calculated from many measurements
than one derived from only a few. Also called Standard Error of the Mean.

Notes on Statistical words: In the above glossary, many concepts are called by different
names. Many surveying texts use the terms Standard Error and Standard Error of the
Mean because the authors feel that the technical definition of error is closer to the way it
is used in surveying than the terms Standard Deviation and Standard Deviation of the
Mean. This presentation uses the term Deviation because it is more common in standard
statistical texts. Either way is acceptable as long as both instructor and students agree that
they are identical terms.