Theory of Errors in Observations

Theory of Errors in Observations
Introduction
No matter how carefully made, however, observations (measurements) are never exact and will always contain error. Surveyors (geomatic engineers) should understand the different kinds of errors, their sources and expected magnitudes under varying conditions, and their manner of propagation.
Elementary Surveying, 11e 2006 Prentice-Hall, Inc..
1/45

Direct and Indirect Observations
Direct Observations:

Applying a tape to a line, Fitting a protractor to an angle, or Turning an angle with a total station instrument. The distance across the river
Indirect Observations:
2/45

Errors in Observations
By definition an error is
E=X X
where E : the error in an observation X : the observed value X : its true value
3/45

It can be unconditionally stated that

1. 2. 3. 4. no observation is exact, every observation contains error, the true value of an observation is never known, and the exact error present is always unknown.
4/45

For example:
When a distance is observed with a scale divided into tenths of an centimeter (cm), the distance can be read only to hundredths. If a better scale graduated in hundredth of a cm, and read under magnification, the same distance might be estimated to thousandths of a cm. so on.
5/45

Obviously, accuracy of an observations depends on

the scales division size, reliability of equipment used, and human limitations in estimating closer than about onetenth of a scale division.
As better equipment is developed, observations one closely approach their true values. But they can never be exact. Note that observation such as counts are not considered here (e.g. one car, two men).
6/45

Mistakes
These are observer blunders and are usually caused by

misunderstanding the problem, carelessness, fatique (being tiered), missed communication, or poor judgment.
7/45

Mistakes
Examples:
Transportation of numbers (recording 73.96 correct value is 79.36) Reading an angle counterclockwise but indicating it is a clockwise angle in the field notes. Sighting the wrong target.
Large mistake (instead of 862.38 recording 682.38) must be detected by careful and systematic checking of all work. It is very difficult to detect small mistakes.
8/45

Sources of Errors in Making Observations
Natural Errors:
Errors that caused by variation in:
wind, temperature, humidity, atmospheric pressure, gravity, etc.
Instrumental Errors:
Errors that result from any imperfection in the construction or adjustment of instruments.
Personal Errors:
Arise from limitations of the human senses of sight and touch.
9/45

Types of Errors
Errors in observations are of two types:

Systematic errors, and Random errors. Result from factors that comprise the measuring system. For example: The use of a 100 ft steel tape that has been calibrated and found to be 0.02 ft too long. The error is readily eliminated by applying a correction.
Systematic Errors (biases):

10/45

Precision and Accuracy
Discrepancy:
A discrepancy is the difference between two observed values of the same quantity. A small discrepancy probably no mistakes and random errors. Precision refers to the degree of refinement or consistency of a group of observations. High precision: If multiple observations are made of the same quantity and small discrepancy results.
11/45
Precision:

Precision and Accuracy
Accuracy:
Accuracy denotes the absolute nearness of observed quantities to their true values. The difference between precision and accuracy:
(a) Precise but not accurate (b) Neither Precise nor accurate (c) Both Precise and accurate
12/45

Eliminating Mistakes and Systematic Errors
In the field, mistakes can be minimized by experienced observers who alertly perform their observations using standardized repetitive procedures. Mistakes can be corrected only if discovered. Comparing several observations of the same quantity is one of the best way to identify mistakes. For example: Assume that five observations of a line are recorded as:

567.91, 576.95, 567.88, 567.90, and 567.93 576.95 disagrees with the others. There is a mistake.
13/45

Eliminating Mistakes and Systematic Errors
Systematic errors can be calculated and proper corrections applied to the observations. Procedures for making these corrections are described on the following chapters.
14/45

Probability
Events that happen randomly or by chance are governed by mathematical principals referred to as probability. Probability = (number of times a result should occur) / (total number of possibilities). If a result occur in m ways and fail to occur in n ways, then the probability of its occurrence is m/(m+n). The probability is a fraction between 0 and 1

0 indicates impossibility. 1 denotes absolute certainty.

15/45

Probability
For the remainder of this chapter, the term error will be used to mean only random error. It will be assumed that all mistakes and systematic errors have been eliminated before random errors are considered.
16/45

Most Probable Value

The true value of any quantity is never known. Its most probable value can be calculated, however, only if redundant observations have been made. Redundant Observations: Measurement in excess of the minimum needed to determine a quantity. For example: Measuring a line length directly and independently number of times using the same equipment and procedures. The first measurement establishes a value for the quantity and all additional measurements are redundant.
17/45

Most Probable Value
The most probable value is simply the arithmetic mean.
M =
M
i =1
M1 + M 2 + L + M n = n
where Mi= i th observation n = the total number of observation
18/45

Residuals
A residual is simply the difference between any observed value of a quantity and its most probable value.
i = M Mi
or
i = Mi M
where Mi= i th observation i = the residual in observation Mi

Elementary Surveying, 11e 2006 Prentice-Hall, Inc.. 19/45

Occurrence of Random Errors
Assume these 100 observations are free from mistakes and systematic errors. The data have been rearranged in column 1 from smallest to largest value. Frequency of the values is tabulated in column 2. Dispersion = Largest smallest = 30.8 19.5 = 11.3 sec The most probable value for the observed angle is calculated. Residuals for all observed values are computed using (column 3)
20/45

Histogram:
A bar graph showing the size of the observations (residuals) versus their frequency of occurrence. Histogram gives an immediate visual impression of the distribution pattern of the observations (residuals). To obtain a histogram with an appropriate number of bars for portraying the distribution of residuals adequately, the interval of residuals (class interval) represented by each bar was chosen as 0.7. This produced 17 bars on the histogram.
21/45

22/45

In practice, normal distribution curves that approximate histograms and frequency polygons are preferred to represent the distribution of error.
Normal Distribution Curve


The histogram for a set of observations shows the probability of occurrence of an error of a given size graphically by bar areas. For example, 14 of the 100 residuals (errors) are between -0.35 and +0.35. This represent the 14 percent of the errors. The probability of occurrence of an error being between -0.35 and +0.35 is 14% or 0.14.
24/45

Likewise, the area between ordinates constructed at any two abscissas of a normal distribution curve represents the percent probability that an error of that size exist. The total area beneath a normal distribution curve is 1.
Normal Distribution Curve


If the same observations of the previous example had been take using better equipment and more caution, smaller errors would be expected and normal distribution cure would be taller and narrower showing that a greater percentage of values have smaller errors and, fewer observations contain big ones. The observations in Figure (a) are taken more precisely than the observations in Figure (b).
Normal Distribution Curve a) Increase precision b) Decrease precision
26/45

It is seen that relative precisions of groups of observations become readily apparent by comparing their normal distribution curves.

A shorter and wider curve Less precision A taller and narrower curve More precision
Normal distribution curve for a set of observations can be computed using parameters derived from the residuals.
27/45

General Laws of Probability
From the analysis of the data on the preceding sections and curves some general laws of probability can be stated:
Small residuals (errors) occur more often than large ones; that is, they are more probable. Large errors happen infrequently and are therefore less probable; for normally distributed errors, usually large ones may be mistakes rather than random errors. Positive and negative errors of the same size happen with equal frequency, that is, they are equally probable.
28/45

Measure of Precision
It is seen that
More precision: A taller and narrower curve Less precision: A shorter and wider curve
To express precision of group of observation Standard Deviation can also be used.

=
2 i =1 i n
n 1
29/45

Measure of Precision
It is seen that

A small High precision A large Low precision The closer spacing between inflection points More precision The larger spacing between inflection points Less precision
In other words,
30/45

Interpretation of
Area = 68.3%
establishes the limits within which observations are expected to fall 68.3% of the time.
31/45

Interpretation of
For the data of Table 3.1:

= 2.19 sec In examining the residuals in the table 68% of the values are between -2.19 and +2.19.
5.4 2 + 4.9 2 + L + (5.9) 2 = = 2.19 100 1 n 1
2 i =1 i
32/45

The 50, 90, and 95 Percent Errors
50% Error:
E50 = 0.6745
90% Error:
E90 = 1.6449 = 3.60 sec
95% Error:
E95 = 1.9599 = 4.29 sec
For example: E90 establishes limits within which the observations should fall 90% of the time. In other words, an observation has 90% chance of being in these limits.

The 50, 90, and 95 Percent Errors
Within a group of observations, any value whose residuals exceeds 3 is considered to be a mistake. In this case either a new observation must be taken or that value must be disregard.
34/45

Error Propagation
The process of evaluating errors in quantities computed from observed values which contain errors is called error propagation. Error of a Sum:
Assume the sum of independent observed observations a, b, c, is Z. Z= a + b + c + The propagated error in the sum of quantities, each of which contains a different random error:
2 Esum = Ea + Eb2 + Ec2 + L
35/45

Error Propagation
Error of a Series:
The total error in the sum of all observed series of similar quantities with about the same amount of error is called the error of the series The propagated error in the series:
Eseries = E 2 + E 2 + E 2 + L = nE 2 = E n
where n : number of observations E : an error in each individual observations
36/45

Error Propagation
Error in a Product:
The equation for propagated error in a product AB, where Ea and Eb are the respective errors in A and B, is
2 Eprod = A2 Eb2 + B 2 Ea
37/45

Error Propagation
Error of the Mean:

The most probable value Arithmetic Mean Mean is computed from individual observed values, Observed value contains an error Mean value contain an error. E Error of the mean: Em = series
E n E Em = = n n

Error Propagation
Error of the Mean:
The error of the mean at any percentage probability can be determined and applied to all criteria that have been developed. For example: The standard deviation of the mean
Eseries Em = n
( E68 ) = m =
(n 1) n
n(n 1)
39/45

Error Propagation
Error of the Mean:
The 90 and 95% errors of the mean are
E90 1.6449 = = ( E90 ) m = n n E95 1.9599 ( E95 ) m = = = n n

1.6449 n 1.9599 n
(n 1)
2
= 1.6449
n(n 1)
2
(n 1)
= 1.9599
n(n 1)
40/45

Error Propagation
Error of the Mean:
These equations show that the error of the mean varies inversely as the square root of the number of repetitions. Thus to double the accuracy (to reduce the error by onehalf) four times as many observations must be made.
41/45

Conditional Adjustment of Observations

True value of any observed quantity is never known. In some types of problems, however, the sum of several observations must equal a fixed value.
For example: The sum of the three angles in a plane has to total 180.
In practice, therefore, the observed angles are adjusted to make them add to the required amount.
42/45

Weights of Observations
It is evident that some observations are more precise than others because of

Better equipment, Improved techniques, and Superior field conditions.
Therefore, it is desirable to assign relative weights to individual observations. High precision Small standard deviations or variance In the adjustment of the observations:
Higher the precision the larger the relative weight

43/45

Relative weights are inversely proportional to variances:
Wa
2 a
where Wa= the weight of an observation a. a2= a variance (square of the standard deviation)
44/45

If the individual observations have varying weights, the weighted mean can be computed as
MW
WM = W
where WM = The sum of the individual weights times their corresponding observations. W = The sum of the individual weights.
45/45

Theory of Errors in Observations

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Theory of Errors in Observations

Загружено:

Авторское право:

Доступные форматы

Theory of Errors in Observations

Elementary Surveying, 11e 2006 Prentice-Hall, Inc..