ERROR PROPAGATION OF A TOTAL STATION

By Shea Gleadle

Equations
Suppose two measured quantities x and y have standard errors (standard deviations) associated with those
measurements, σx and σy, we would report (x ± σx), and (y ± σy). From the measured quantities a new quantity, z, is
calculated as a function of x and y. Using error propagation formulas we can determine the standard deviation of our
derived value (σz).

Addition and Subtraction: z = x + y or z = x – y

For addition or subtraction, you simply add the square root of the standard errors, squared, together.

σz = √σx2 + σy2 + σn2…

Ex. The elevation difference between the optical center of the instrument and center of a reflector (VD) was measured
as 3.201 ± 0.003m, the HT was measured as 1.543 ± 0.005m, the HI was measured as 1.641 ± 0.002m and our setup
point (ELA) has an elevation of 100.323m and we are going to assume it is without error for now. We know that the
elevation of the unknown point ELB = ELA + VD + HI – HT. Therefor we can use the above equation to determine the
standard error (standard deviation) of the elevation of our unknown point, B.

σELB = √ (σELA2 + σVD2 + σHI 2 + σHT2)

Since we are assuming the elevation of A is without error, the standard error is 0 and therefor σELA2 reduces to 0. If we
wanted to consider the error of the determination of the elevation of our setup point, we would either estimate a value
for σELA or determine it through error propagation and substitute it into the above formula. For this example, we can
continue on with the formula;

σELB = √ (02 + 0.0032 + 0.0022 + 0.0052)

σELB = ± 0.006m

ELB = 100.323 + 3.201 + 1.641 – 1.543

ELB = 103.622 ± 0.006m

Our final answer is B has an elevation of 103.622m and we are sure of that elevation to within ± 0.006m at a confidence
interval of 68%. Another way of saying this would be that if we measured the elevation of B 100 times, statistics tells us
it would be likely that we would determine a value of 103.622 ± 0.006m 68 times. The other 32 times we would
determine a value outside of those limits, below 103.616m or above 103.628m. If we wanted to be more confident in
the value of the elevation range we are quoting, we would increase our confidence interval. To do this we must look at
the equation of a mean. It tells us the mean of a measured value (x̅) is precise to the Z score multiplied by the standard
deviation of the mean divided by the square root of the number of times we measured the mean (n).

x̅ +/- Z *σ / √n

For 95% confidence, the Z score is 1.959964 and for 99% confidence it is 2.575829. So, another way of representing the
elevation of B would be ELB = 103.622 ± 0.012m at E95%
Multiplication and Division: z = x * y or z = x/y

If z is proportional to a product of two or more measured quantities, then we must switch to looking at relative standard
errors. We will look at the equation;

σz / z = √((σx / x )2 + (σy / y)2 + (σn / n)2…)

The percent standard error of the unknown is equal to the square root of the percent standard error of the
measurements, squared & summed. We cannot directly use this equation, for example, to determine the standard error
of HD when HD = SD * cos(ZA) because we know the standard error of ZA, not cos(ZA). To use this equation we would
need to calculate σcos(ZA). However, we can just use partial derivates instead with the following equation;

σz = √((∂z/∂x)2 * σx2 + (∂z/∂y)2 * σy2 + (∂z/∂n)2 * σn2 …)

This equation is what the addition, subtraction, multiplication, power, ect error propagation equations are derived from.

Introduction
In the following paper and accompanying worksheet I will go over the random sources of error that affect the
measurements of a modern total station and how they influence the derived coordinates of a surveyed point. The
calculated precisions assume that the systematic errors have been accounted for and properly reduced with calibration
and field procedures.

Sources of Error
Random Error Systematic Error
• Pointing & Reading (σd) – Provided by manufacturer • Horizontal Collimation
• EDM Constant (σEc) – Provided by manufacturer • Tilting Axis
• EDM Scale (σEs) – Provided by manufacturer • Vertical Collimation
• Centering error @ Instrument over the point (σI) - Estimated • Compensator index
• Centering error @ Target over the point (σT) - Estimated • Circle Eccentricity
• Plumbness of Prism pole error (σP) - Calculated • Circle Graduation
• Centering error of the reflector optical center vs vertical axis • Collimation of EDM
of prism holder (σR) – Provided by manufacturer
*These can be reduced by F1 + F2 observations and
• Measurement of the height of instrument (σVi) - Estimated
regular calibration
• Measurement of the height of target (σVt) - Estimated
• Environmental conditions
• Un-even heating of the instrument / calibration
at a different temperature
• Vibrations/tripod instability
• Prism pole out of level
• EDM out of calibration
*These can be reduced equipment calibration & field
procedures.
Pointing & Reading error (σd)
Pointing and reading error is specified by the manufacturer in +/- arc seconds (σd) in ISO (17123-3) or DIN (18723)
standards. This is the angular standard deviation of a single direction of a f1+f2 measurement. To get the standard
deviation of an angle we will use the afore mentioned equation of error propagation for addition/subtraction because
an angle is the difference of two directions, where d2 is the direction to the foresight and d1 is the direction to a
backsight and σd2 & σd1 are their accompanying standard deviations. And because we may measure the direction/angle
more than once we must also consider that the standard deviation of the mean of a measurement is reduced by 1 / √n,
where n is the number of times the angle is measured.
Ag = d2 – d1
σAg = (√((σd2)2 + (σd1)2) / √n
σAg = (σd * √2) / √n
Ex. σAg = 3” * √2 (for a 3” instrument)
σAg = 4.24”
Ex. 4.24” / √3 (for 3 rounds of observations of an angle)
σAg = 2.45”

To get the standard deviation of the angle observed with a single face (f1 only), we will look back at the equation for the
standard deviation of a mean and consider that the manufacturer is providing us with σd in the terms that it includes
two pointings and readings (f1 and f2).

x̅ +/- Z *σ / √n

The mean is this case is the observed direction of f1 + f2. We will ignore the Z score for now and the σ is σd and n is
equal to 2 because we measured the mean twice, once in f1 and once in f2.
σd = σd1 / √2
σd1 = σd * √2
σAg1 = σd1 * √2 / √n
Ex. 4.24” * √2 (for a single face observation of an angle and a 3” instrument)
σAg = 6”
EDM Error (σEc + σEs)
The EDM error is specified by the manufacturer (ISO 17123-4) with two components, the constant, in +/- meters or mm
(σEc) and a scale component (σEs) that’s in PPM. This quoted accuracy is based off a single measured distance, such as in
tracking mode. However, it is common for the instrument to take 3-5 (depending on settings) distance observations and
use the mean of those measurements.
The σEc + σEs may be different for IR (infrared) vs RL (reflectorless) measurements, and it should be noted that RL EDM
measurements are only at their quoted accuracy when shooting a target perpendicular to the LOS (line of site) with
good reflectivity and a flat surface, free of obstructions. The accuracy degrades significantly past 30° with RL
measurements and should be avoided. To calculate the combined effect that these two components have on the
measured distance (σSD) we must use the following equation where nD is the number of distance measurements taken
by the instrument.
σSD = +/- (σEc + (σEs*SD)) / √ nD
Ex. σSD = (0.002 + ((2 / 1000000) * 500) / √5 (for a 2mm +/- 2 PPM instrument @ 500m measured with 5 distance measurements taken)
σSD = 0.0013m
Ex. σSD = (0.002 + ((2 / 1000000) * 500) / √1 (for a 2mm +/- 2 PPM instrument @ 500m measured in tracking mode)
σSD = 0.0030m

Centering error (horizontally) of the reflector glass over the point (σR)
Different reflectors are made to different quality standards. However, all reflectors contain some quantity of centering
error. This is the amount the optical center of the prism may vary when compared to the vertical axis of the prism
holder. These values can be found on most manufacturers’ websites. They generally range from 0.0003m to 0.005m.
Some prisms also contain an angular error (σRa), such as the Trimble Multitrack. The total reflector error (σR) is a
function of this angular error and the constant centering error (σRc), which can be calculated with the following formula;

σR = √ (σRc2 + ((tan(σRa/3600) * HD)2

This standard error is not reduced by 1 / √n because the amount or direction the optical center of the prism varies when
compared to the vertical axis of the prism holder will not change. This is not to be confused with beam deviation. When
a reflector is quoted as having an x” of beam deviation, that value affects the maximum range of the EDM, not so much
the precision of the distance measurement. The more the EDM beam diverges, the more the range is reduced.

Centering error (horizontally) over a point (σI + σT + σP)

The ability to perfectly center the instrument and target over the point horizontally affects the measured slope distance
(SD), derived horizontal distance (HD), the measured azimuth (AZ) and subsequently the derived horizontal position of
the point (N & E). These can be estimated based on how well the physical control points center can be defined, the
condition of the tribrach, whether a bipod was used and the height of instrument/target. This should not be confused as
the centering error provided by the manufacturer of a reflector based on the precision in which the reflector was
constructed (σR). It should be noted that when estimating this value, you should estimate it as a standard deviation for it
to properly fit into subsequent formula. How close could it be centered with 68% confidence.
An old, beat-up tribrach on a tripod that hasn’t been recently adjusted will have a larger centering error than a newer
tribrach in good condition. And a prism pole at a 2m HT (height of target) without a bipod is going to have a larger
standard deviation due to plumbness of pole (σP) versus a mini prism with a HT of 0.100m. This value can be
approximated with a function based on the bubble sensitivity of a poles level bubble (Ps) and the HT. The bubble
sensitivity is defined as the angular deviation of the prism pole with respect to the vertical axis that is present when a
level bubble moves 2 mm from the center. The idea is the surveyor wouldn’t allow the bubble to be out of center by
more than 2mm 68% of the time. At that distance, the edge of the bubble touches the sides of the black circle on the
viewing glass (Leica). If you are unsure of the sensitively of your poles level bubble you can assume it is 20’. 8’ bubbles
are available for higher precision work and some topo poles have 40’ bubbles.

σP = HT * tan(Ps)
 Ex. σP = 2.000 * tan(20’) (for a 2m rod)
σP = 0.012m
Ex. 0.10 * tan(20’) (for a mini prism at its lowest extension)
σP = 0.0006m

As you can see there is a significant difference based on your HT, this error can be greatly reduced using a bipod when
the HT is larger. If a bipod is used, I recommend reducing the affect Ps has on the derived error by 4x. This is because
now the surveyor is capable of precisely holding the bubble within the middle by approximately 0.5mm. This assumes
the pole is straight and the level bubble is calibrated. The equation to be used with a bipod is as follows:

σP = HT * tan(Ps / 4)
Ex. σP = 2.000 * tan(20’ / 4) (for a 2m topo rod)
σP = 0.003

This equation also assumes the pole tip is precisely centered on the point, which is easier done with forced centering
points, a mag nail or capped post for example.
If a tribrach and prism are used, then σP should be replaced with σT, which is the standard deviation of the target
centering over a point and is due to imperfect levelling, imperfections in perpendicularity of the optical plummet and
not being able to precisely make out the center of the point while viewing it through the optical plummet. This is more
difficult to calculate based on the variables of the equipment condition and viewability of the point. I recommend
estimating a value of 0.002m – 0.005m. If the point has a well-defined center and the tribrach is in good condition, I
would lean closer to the lower end of the range.
The perpendicularity of an optical plummet can be checked with tribraches that rotate by centering the point, rotating
the tribrach, and checking to see if the point is still centered. This is not possible with fixed tribraches. To check these,
you can center the tribrach, then install an instrument in good repair onto it and use the instrument’s eyepiece to verify
it agrees with the tribrach.
The centering error at the instrument (σI) has the same contributors as that of σT, however since the instrument has a
much more sensitive level bubble and generally the tribrach is stored in better conditions, I tend to use a smaller value
here. Somewhere around 0.001 - 0.002m if your setup point has a well-defined center.

Measurement of the height of instrument & target (σVi + σVt)

When we measure the HI or HT there are random errors associated with the measurements of those heights. These
values can be estimated as there is no accurate way to calculate them. The error associated with measuring the HI (σVi)
is generally smaller than that of measuring the HT (σVt) as the most instruments tend to have a more well-defined
vertical center compared to most reflector back plates. This is especially true when using the bottom mark that’s present
on newer total stations. This rule is reversed when using prism poles that have snap in vertical graduations, the Leica
GLS12 for example, or a mini prism with screw together extensions. I generally tend to use a value of 0.003m for σVi and
a value of 0.005m for σVt when using a prism on a tripod and 0.001-0.003m when using a prism pole (depending on the
model of pole and the condition of the tip). To reduce this error when using a tribrach you can use a height meter
specially made to fit into tribraches, the Leica GMH007 for example.
It should again be noted that the errors due to miscentering, vertical or horizontal are not directly reduced by 1/√n
because you are not remeasuring the HI, HT or recentering the instrument or target between each set.

Error in determining an azimuth (σAZ)

We know the azimuth to a point (AZ) can be determined by a function of the azimuth at the backsight (AZBS) and the
measured angle (Ag)

AZ = AZBS + Ag – 180

The standard deviation of that azimuth (σAZ) can be determined using error propagation on the function and adding in
the effect of the horizontal error due to centering as a ratio over the horizontal distance;

σAZ = √ ((σAZBS)2 + (σAg)2 + (σI / HD)2 + (σP / HD)2 + (σR / HD)2)

We are using the centering errors relative to the HD because we want a percent standard deviation to keep everything
unitless. The angular error is converted into radians for the same reason. We can convert the final answer back into
degrees, then arc seconds once we are done the calculation. Since we are assuming the control we are coming off of is
without error and are holding them as fixed (as we are calculating relative accuracy), then we can assume the derived
azimuth at the backsight is also without error. Therefor;

σAZ = √ ( (σAg)2 + (σI/HD)2 + (σP/HD)2+ (σR/HD)2)

(Continuing on from our above example with a HT = 1.500 and using a bipod)

Ex. σAZ = √(((3” * √2) * (π/648000)) / √1)2 + (0.002/499.981)2 + (0.0022/499.981)2 + (0.002/499.981)2)/(π/648000)

Ex. σAZ = 4.49”

Error in determining the horizontal distance (σHD)

The horizontal distance (HD) is the slope distance reduced with the vertical component removed. It is a function of the
SD and the zenith angle (ZA);
HD = SD * sin(ZA)

Using error propagation with partial derivatives, we can determine the standard deviation of the horizontal distance
(σHD) is equal to;

σHD = √((∂HD/∂SD)2 * σSD2 + (∂HD/∂ZA)2 * σZA2)

σZA is equal to the standard deviation of a single direction (σd) because a zenith angle is not the result of two pointings,
but a single direction relative to the vertical axis of the instrument as defined by gravity at that point. We could include
the standard deviation of the effect that mislevelling the instrument would have on the zenith angle, but most total
stations have a built-in levelling compensator that reduces this value to a negligible amount. Therefor;
 (∂HD/∂SD) = sin(ZA)
(∂HD/∂ZA) = SD * cos(ZA)

σHD = √((sin(ZA)2 * ((σEc + (σEs*SD))/√n)2 + (SD * cos(ZA))2 * (σd / √n)2)

Ex. √((sin(90.5)2 * (0.002 + ((2/1000000)*500))2 + (500 * cos(90.5))2 * (3” *(π/648,000) / √1)2)
(for a 2mm +/- 2 PPM, 3” instrument @ a SD = 500m and ZA = 90° 30’ 00” in tracking mode)

σHD = 0.0030m

The total standard deviation of the horizontal distance (σHDt) also includes the errors from centering the instrument
over the point (σI), target (σT) (or pole plumbness (σP)) and the centering error of the reflector (σR). Since all our units
are in meters, we do not need to use a percent standard deviation here. Thus, the final equation for the σHD is as
follows;

σHDt = √(σHD2 + σI2 + σP 2 or σT2 + σR2)

Ex. σHDt = √((0.003)2 + 0.0022 + (1.5 * tan(20/4”))2 + 0.0022)
(Continuing from above, using an instrument centering of 2mm, a pole and bipod at 1.5m HT and a standard Leica circular prism)

σHDt = 0.0047m

Error in determining the vertical distance (σVD)

The vertical distance (VD) is the slope distance with the horizontal component removed. It is a function of the SD and
the zenith angle (ZA).
VD = SD * cos(ZA)

Using error propagation, we can determine the standard deviation of the vertical distance (σVD) is equal to;

σVD = √((∂VD/∂SD)2 * σSD2 + (∂VD/∂ZA)2 * σZA2)

(∂VD/∂SD) = cos(ZA)
(∂VD/∂ZA) = -SD * sin(ZA)

σVD = √((cos(ZA)2 * ((σEc + (σEs*SD))/√n)2 + (-SD * sin(ZA))2 * (σd / √n)2)

Ex. √((cos(90.5)2 * (0.002 + ((2/1000000)*500))2 + (-500 * sin(90.5))2 * (3” *(π/648,000) / √1)2)
(for a 2mm +/- 2 PPM, 3” instrument @ a SD = 500m and ZA = 90° 30’ 00”)

σVD = 0.0073m
Error in determining the elevation of a point (σEL)
The standard deviation of the relative elevation of the point we are measuring to, is the square root of the sum of
squared errors of the measured vertical distance (σVD) and the standard deviation the measured HT (σVt) and HI (σVi).

σEL = √(σVD2 + σVt 2 + σVi 2)

Ex. σEL = √(0.00732 + 0.001 2 + 0.0032)
(Continuing from above, using an instrument vertical centering of 1mm and target vertical centering of 3mm)

σEL = +/- 0.0079m

E95%EL = 1.96 * (σEL) = +/- 0.0155m

Error in determining the horizontal coordinates of a point (σN + σE)

The standard deviation of the relative coordinates of the point we measure to are a function of the azimuth (AZ) and the
horizontal distance (HD)

∆N = HD * cos(AZ)

Using error propagation, we can determine the standard deviation of the relative northing of the measured point (σN) is
equal to;
σN = √((∂N/∂HD)2 * σHDT2 + (∂N/∂AZ)2 * σAZ2)
σN = √(cos(AZ)2 * σHDT2 + (-HD*sin(AZ))2 * σAZ2)
Ex. σN = √(cos(155)2 * (0.0047)2 + (-499.981*sin(155))2 * ((4.49”)2)
(Continuing from our above example and AZ = 155° 00’ 00”)

σN = +/- 0.0062m
E95%N = 1.96 * (σN) = +/- 0.0122m

∆E = HD * sin(AZ)

Using error propagation, we can determine the standard deviation of the relative easting of the measured point (σE) is
equal to;
σE = √((∂E/∂HD)2 * σHDT2 + (∂E/∂AZ)2 * σAZ2)
σE = √(sin(AZ)2 * σHDT2 + (HD*cos(AZ))2 * σAZ2)
Ex. σE = √(sin(155)2 * (0.0047)2 + (-499.981*cos(155))2 * ((4.49”)2)
(Continuing from our above example and AZ = 155° 00’ 00”)

σE = +/- 0.0101m
E95%E = 1.96 * (σE) = +/- 0.0198m
To get a better idea of the magnitude and direction of the possible error in the relative position of the coordinate we will
look at error ellipse at 95% confidence. This ellipse will tell us where the true location of the measured point will lie with
95% confidence. We can calculate the semi-minor axis as a function of twice the E95% HDt;

a = (2 *σHDt ) * 1.96
Ex. a = 2 * (0.004665) * 1.96
a = 0.018m

We can calculate the semi-major axis as a function of twice the tangent of E95% AZ multiplied by the HD;

b = (2 * tan(σAZ) * HD) * 1.96

Ex. b = (2 * tan(4.49”) * 499.981) * 1.96
b = 0.043

Since we are calculating the relative error ellipse we know the azimuth of the error ellipse will be perpendicular to the
azimuth to the measured point.
AzM = AZ +/-90°
AzM = 65° 00’ 00”

It should be noted that sometimes the a and b may be swapped, that is to say sometimes the error due to HD will be
larger than that of the error due to the angular measurement. In which case just note that the larger value is always
technically the semi-major. We must remember these numbers are the dimensions of the ellipse with the measured
point being at the center of the ellipse. So, for example the true location of the measured point is going to be within
0.043 m/2 or +/- 0.0215 m along the azimuth of 65°. If we want to know how large the ellipse would be with within 1
standard deviation of confidence, simply divide the a and b by 1.96.

Conclusion
The purpose of these formulae and the accompanying worksheet are to determine two things.
1. Statistically, how precise are you measuring/how precise are the coordinates you are deriving from your
measurements.
2. Given a required precision, what variables can you adjust to meet that precision.

For example, we are asked to “layout some building corners to within +/- 3mm.”
This is a common request from a client, and because it isn’t very specific, we will have to make a few assumptions.
A. The requested precision will be relative, and we will assume our control is without error. Usually, a project must
be relatively more precise than absolutely, as the structures and utilities within the scope of a project need to tie
in more precisely with each other than with structures and utilities outside of the project scope. Not to say
absolute precision isn’t important. But we can assume the control network has considered the absolute accuracy
required.
 B. It is rare that a confidence level accompanies a tolerance specified by a client, and there is a good chance they
won’t fully understand what a standard deviation is or how that relates to confidence intervals. Although it will
require more stringent controls, we should usually err on the side of safety and use 95% confidence depending
on what the layout is for. I’ve found it common that a client will ask for excessively precise tolerances without a
full understanding of what is required to achieve them. That being said, for this task I understand these building
corners are only for concrete formwork and probably don’t require 95% confidence of 3 mm. There will most
likely be deviations in the physical formwork of more than 3mm. We will work towards 1 standard deviation of
confidence for this example.
C. We can ignore elevations in this example as they were not asked for.
Continuing with our example above we know the semi-major axis of our 95% relative coordinate error ellipse is +/-
0.022m which means we are going to be sure that our measured location to within +/- 0.011 m. And since our error is
largest along the semi-major axis, we can ignore the semi-minor axis for now.
Since we have control setup around our worksite, we can ensure our max SD < 50 m. That alone gets our σSemi-major =
+/- 4 mm. Next, we are going to swap out our 1.5 m rod/bipod to a Leica GMP101 mini with a HT of 0.100 m which gets
us to +/- 3 mm. So, at this point we are within our clients’ specs for 68% of our measurements, at 95% confidence were
at +/- 6 mm. Let’s continue to see if it’s possible to get our 95% confidence down to 3 mm.
Because we are staking out, it is difficult and time consuming to make multiple sets to each interation of a stake out
point, so we will refrain from adjusting our n value. However, taking the instrument off tracking mode got us down to 5
mm at 95%. At this point it becomes very difficult to increase our precision. If we swap out our instrument for a +/- 1”
1mm +/- 1PPM instrument, we only get down to 4.5mm. To get that last little bit of precision we have two options. We
need to either center our instrument to within 1mm (This may be doable with extra care and ensuring the instrument
optical plummet is perpendicular to the vertical axis by centering over the control point and rotating the instrument and
ensuring there is no deviation) or using a prism with a lower σR. If we keep our GMP101 and take great care in centering
the instrument to +/- 1mm we get to +/- 3 mm at 95% confidence.