Measurement Theory, Errors and Accuracy

Introduction

Surveying is an observational science. Distances, directions, elevations and coordinates are not obtained as perfectly exact quantities; they are measured with instruments operated by people under physical and environmental conditions. The professional surveyor therefore does more than read a graduated scale. The surveyor must decide whether an observation is sufficiently reliable for its intended purpose, identify possible sources of error, detect mistakes, estimate uncertainty, and communicate results with an appropriate number of digits.

Measurement theory provides the foundation for these decisions. In construction surveying, an error of a few millimetres may be insignificant when locating rough excavation limits but unacceptable when setting out a steel connection or monitoring settlement. Accuracy is consequently related to purpose, specification and risk rather than to numerical precision alone.

This chapter develops the concepts required to evaluate survey observations: units and significant figures, accuracy and precision, tolerances, error classification, uncertainty, error propagation, calibration, checks, redundancy and elementary statistical analysis.

Learning objectives

By the end of this chapter, the learner should be able to:

1. Distinguish accuracy, precision, resolution, sensitivity and tolerance.
2. Apply significant figures and appropriate rounding to survey results.
3. Classify systematic, random and gross errors.
4. Estimate mean, residuals, variance, standard deviation and standard error for repeated observations.
5. Explain uncertainty and combine independent sources of uncertainty.
6. Propagate uncertainty through simple addition, subtraction, multiplication and derived survey quantities.
7. Design field checks and redundant observations that reveal mistakes and improve reliability.
8. Explain the purpose of calibration and recognize limitations of measurement models.
9. Judge whether observations satisfy a stated surveying tolerance.

1. Measurement as a model-based process

A measurement is an estimate of a quantity. Let the measurand be the ideal or intended quantity X, and let an observation be x. The observation can be represented conceptually as

where e is the measurement error. In practice, X is usually unknown, so error cannot normally be calculated directly. Instead, the surveyor uses standards, repeated observations, closures, calibration certificates and statistical models to assess likely error.

A complete measurement process includes:

• defining the quantity and its reference system;
• selecting an instrument and method;
• controlling conditions such as temperature, atmospheric pressure and line of sight;
• observing and recording readings;
• checking observations independently;
• reducing observations mathematically; and
• reporting a result with an uncertainty or tolerance appropriate to the work.

The distinction between a physical quantity and its recorded representation is important. A total station reading may be displayed to 1″, but this does not mean that the direction is accurate to 1″. Display resolution is only one component of measurement capability.

2. Accuracy, precision and related terms

Accuracy describes closeness of a measured value to the accepted or reference value. Precision describes the closeness of repeated measurements to one another. A set of observations can be precise but inaccurate if a systematic bias affects every observation. Conversely, observations may have a small average error but poor precision if they fluctuate widely.

Resolution is the smallest increment that an instrument can display or distinguish. Sensitivity is the change in instrument indication produced by a change in the measurand. Repeatability is agreement under substantially identical conditions, while reproducibility concerns agreement when conditions such as instrument, observer, location or time change.

A useful conceptual classification is:

Tolerance is a design or contractual requirement; uncertainty is an assessment of the measurement process. A result may lie within tolerance while still having large uncertainty, which creates risk. Sound practice requires the uncertainty to be substantially smaller than the tolerance whenever feasible.

3. Units, significant figures and rounding

Survey observations should be recorded in coherent units. In SI practice, length is commonly expressed in metres or millimetres, angle in degrees, minutes and seconds or radians, and area in square metres. Conversions must preserve dimensional consistency. For example, 25 mm = 0.025 m, while 25 mm² = 0.000025 m².

Significant figures are the digits that convey meaningful measurement information. Leading zeros are not significant; zeros between non-zero digits are significant; trailing zeros may be significant when a decimal point or stated precision makes their meaning clear. Thus, 0.00420 m has three significant figures, whereas 4.20 m also has three.

General rules are:

• retain guard digits during intermediate calculations;
• for addition and subtraction, round according to decimal place;
• for multiplication and division, round according to the least number of significant figures;
• do not report more precision than the observation or uncertainty supports;
• avoid premature rounding in coordinate and adjustment calculations.

If a distance is measured as 36.482 m with an estimated uncertainty of approximately 0.003 m, reporting 36.482 m is defensible; reporting 36.482000 m falsely implies greater knowledge. A result such as 36.482 ± 0.003 m communicates both an estimate and its uncertainty.

4. Classification of errors

4.1 Gross errors

Gross errors are mistakes, not inherent limitations of the measuring process. Examples include reading the wrong staff graduation, entering 18.62 instead of 16.82, occupying the wrong station, sighting the wrong prism or using an incorrect prism constant. Gross errors may be very large and are often detected by independent checks, implausible residuals or disagreement between field notes and electronic files.

Gross errors must not be treated as random observations and averaged away. The appropriate response is investigation, correction from original evidence where possible, or rejection and re-observation.

4.2 Systematic errors

Systematic errors have a consistent or predictable pattern. Sources include an incorrectly calibrated tape, collimation error, atmospheric refraction, prism-constant error, incorrect instrument height, scale-factor error and temperature effects on a measuring tape. A systematic error may be represented by a bias b, so that

where r_i is the random component of observation i. Repetition alone does not remove b. It must be eliminated, modelled, corrected or reduced through procedure.

For a tape whose actual length is L_a and nominal length is L_n, the scale correction for a measured distance D_m, ignoring other effects, is

where D_c is the corrected distance. The formula assumes the tape scale error is known and uniform and that temperature, pull, sag and alignment effects are either negligible or corrected separately.

4.3 Random errors

Random errors vary in sign and magnitude in an unpredictable manner. They arise from reading limitations, small pointing variations, instrument noise and changing atmospheric conditions. Repeated observations can reduce their effect on an average, but cannot guarantee that any individual observation is free from error.

When random errors are approximately independent and have zero mean, the arithmetic mean of n observations is

The mean is an estimate of the population mean under the stated assumptions. It is not automatically the true value, particularly when systematic error remains.

5. Basic statistical treatment of repeated observations

Suppose five distance observations, in metres, are 42.318, 42.321, 42.319, 42.320 and 42.322.

The mean is

The residual for observation i is

The residuals are −0.002, +0.001, −0.001, 0.000 and +0.002 m. Their sum is zero, apart from rounding.

The sample variance is

and the sample standard deviation is

For these observations, the sum of squared residuals is 0.000010 m². Therefore,

and

The standard error of the mean is

so here sₓ̄ = 0.00158 / √5 = 0.000707 m, or approximately 0.7 mm. This reduction applies to the random component under independence and comparable observation conditions; it does not remove an uncorrected bias.

For small samples, a confidence interval for a population mean may use the Student t distribution rather than a normal critical value. For example, an approximate two-sided interval is

where α is the significance level and t is the appropriate t critical value. The interval interpretation depends on the statistical model and does not automatically include systematic effects.

6. Uncertainty and tolerances

Uncertainty is not the same as error. Error is the difference between a measured value and the unknown true value; uncertainty expresses the dispersion or doubt associated with the reported result. Uncertainty may be evaluated from repeated observations (Type A evaluation) or from other information such as calibration certificates, instrument specifications, previous data and engineering judgement (Type B evaluation).

A standard uncertainty is commonly denoted u. If independent uncertainty components are represented by u₁, u₂, …, the combined standard uncertainty is obtained by the root-sum-square rule:

This rule assumes the components are expressed in the same units, are suitably represented by standard uncertainties, and are uncorrelated. If components are correlated, covariance terms are required and must not be ignored.

An expanded uncertainty is often reported as

where k is a coverage factor selected for the desired coverage probability and statistical assumptions. A statement such as 125.430 ± 0.006 m with k = 2 should identify the meaning of the uncertainty and the coverage basis.

A tolerance is assessed against a requirement. If a point must be within ±10 mm and the estimated error or uncertainty is significant relative to 10 mm, a simple pass/fail decision may require a guard band or decision rule. Contract documents and applicable standards should govern acceptance decisions.

7. Propagation of uncertainty

A measured result is often calculated from several input quantities. Let y = f(x₁,x₂,…,xₙ). For small uncertainties, first-order propagation gives

for uncorrelated inputs. More generally, covariance terms are included:

The partial derivatives are sensitivity coefficients. The equations are appropriate when the function can be adequately linearized over the uncertainty range.

Important special cases are:

For addition or subtraction, y = a ± b,

when a and b are independent.

For multiplication by an exact constant, y = c a,

For a product or quotient, relative uncertainties combine approximately as

for independent inputs.

As an example, let a horizontal distance be computed from a measured slope distance s = 50.000 ± 0.005 m and zenith angle Z = 60.000° ± 0.010°, using H = s sin Z. Convert the angular uncertainty to radians: u_Z = 0.010 × π/180 = 0.0001745 rad. The sensitivity coefficients are ∂H/∂s = sin Z and ∂H/∂Z = s cos Z. At 60°, these are 0.8660 and 25.000 m/rad. Thus,

The horizontal distance is H = 50.000 sin 60° = 43.301 m, with an estimated standard uncertainty of approximately 6 mm. This calculation assumes independent input uncertainties, a valid angle convention, and negligible additional effects such as instrument centring and atmospheric corrections.

8. Calibration and instrument control

Calibration compares an instrument or measuring system with a reference of known or traceable value. It may identify scale error, zero error, collimation error, angular index error, EDM scale error or prism-constant error. Calibration does not permanently make an instrument correct; it provides evidence about its behaviour at a particular time and under specified conditions.

A practical calibration system includes:

1. pre-use inspection and battery or compensator checks;
2. periodic calibration by a competent facility or established baseline;
3. recording of certificate number, date, reference standard and uncertainty;
4. adjustment only by authorized personnel;
5. removal from service when performance is outside specification; and
6. verification after transport, shock, repair or suspected malfunction.

Calibration traceability means that a measurement result can be related through an unbroken documented chain to an appropriate reference standard. The chain must also include uncertainties; a certificate without an uncertainty statement is incomplete for rigorous uncertainty evaluation.

9. Checks, redundancy and closure

A check is an independent comparison designed to reveal error. Examples include measuring a line from both ends, closing a traverse, checking a benchmark loop, observing a point from two setups, comparing direct and inverse coordinate computations, and independently entering a critical dimension.

Redundancy exists when more observations are collected than are strictly necessary to determine unknowns. If two independent observations determine a quantity that requires only one observation, their disagreement provides a check. Redundancy also supports adjustment and precision estimation.

For a levelling loop that begins and ends on points of known elevation, the misclosure is

where BS denotes backsight readings, FS foresight readings and H point elevations. For a closed loop with equal start and end elevations, the final term is zero. A positive or negative misclosure indicates inconsistency, but its cause must be investigated before distributing corrections.

A common proportional distribution, used only when permitted by the specification and when error is assumed to accumulate approximately with distance, is

where c_i is the correction assigned to segment i, d_i is its length and D is total length. The corrections sum to −f_h. This is a practical rule, not a universal law; a least-squares adjustment is preferable when observation uncertainties and correlations are known.

10. Adjustment and least squares

When redundant observations are available, adjustment estimates the most defensible values subject to observation precision and mathematical conditions. In least squares, corrections are selected to minimize a weighted sum of squared residuals:

where v is the vector of residuals and P is the weight matrix. Larger weight is assigned to observations considered more precise. The method assumes an adequate functional model, correctly identified observations and reasonable stochastic assumptions. Least squares does not detect every gross error automatically; blunders can distort an adjustment and must be investigated using field checks, residuals and data snooping methods.

For simple independent observations, weight is commonly proportional to inverse variance:

where σᵢ is the standard deviation of observation i. A high-precision observation should therefore influence the adjusted result more than a low-precision observation.

11. Applied case: setting out a foundation grid

Background

The following is an illustrative construction-surveying case. A contractor must set out grid intersection G for a reinforced-concrete foundation. The specification permits a horizontal position error of ±8 mm. The control network contains two known stations, and G is observed from two total-station setups.

Problem

The first setup gives coordinates G₁ = (125.436 m, 84.218 m), while the second gives G₂ = (125.429 m, 84.224 m). The differences are −7 mm in the x-coordinate and +6 mm in the y-coordinate. The independent position discrepancy is therefore

or 9.2 mm. This exceeds the 8 mm tolerance if interpreted as a direct positional discrepancy.

Analysis

The disagreement is not automatically evidence that one observation is wrong. Possible causes include poor centring, incorrect instrument or prism height, an uncorrected prism constant, unstable control, atmospheric effects or a transcription error. The discrepancy is larger than expected from the stated field method, so simply averaging the two coordinates would conceal a problem.

The mean coordinates would be Ḡ = (125.4325 m, 84.221 m), but this average is not an acceptable solution until the disagreement is explained. Averaging can reduce random error; it cannot validate inconsistent observations or remove a common bias.

Approach and solution

The survey team checks the field book, confirms station and target identifiers, measures instrument and prism heights again, performs a face-left/face-right observation, and observes an independent check line to a third control point. The prism constant is found to have been entered incorrectly in the second setup. After correcting the constant and re-observing, the two coordinate solutions differ by 3 mm. The corrected point is then compared with a tape or offset check to the formwork reference, and the result satisfies the project tolerance.

Outcome and learning points

The original discrepancy was not treated as a random fluctuation. Redundancy exposed it; documentation and independent checks identified the cause; correction and re-observation produced a defensible result. The case demonstrates that a numerical average is not a substitute for quality control. In construction, the cost of a short re-observation is usually far less than the cost of mislocated reinforcement, rework or structural nonconformance.

12. Professional practice, limitations and contemporary tools

Digital levels, robotic total stations, GNSS receivers, laser scanners and building-information-model workflows can improve productivity and provide automatic checks. They do not eliminate measurement uncertainty. Automated observation may repeat a wrong target, incorrect coordinate system, wrong antenna height or erroneous prism constant very efficiently.

Artificial intelligence and automated anomaly detection can assist by flagging outliers, unusual residuals or inconsistent metadata. Such tools should support—not replace—professional judgement. A surveyor remains responsible for coordinate reference systems, field verification, instrument status, data provenance and acceptance decisions.

Good practice includes:

• define the required accuracy before observing;
• use calibrated equipment appropriate to the tolerance;
• maintain clear field notes and immutable raw data;
• observe independently where consequences are high;
• balance sights and control environmental influences in levelling;
• retain units, metadata, corrections and uncertainty assumptions;
• investigate outliers before deletion;
• separate correction of systematic effects from statistical averaging; and
• report limitations honestly.

Ethically, surveyors must not alter observations to force compliance, omit unfavorable checks, or report unsupported precision. Contractual specifications, national standards, calibration requirements and workplace procedures should be consulted because tolerance classes and acceptance rules vary by project and jurisdiction.

Chapter Summary

Measurement in surveying is an estimate obtained through an instrument, observer, method and reference system. Accuracy concerns closeness to an accepted value, while precision concerns repeatability. Resolution and displayed digits do not establish accuracy, and a tolerance is a requirement rather than an uncertainty.

Errors may be gross, systematic or random. Gross errors require investigation and correction; systematic errors require calibration, modelling or procedural control; random errors can be characterized and partly reduced through repetition. Means, residuals, variance, standard deviation and standard error provide basic statistical tools, but averaging does not remove bias.

Uncertainty expresses doubt associated with a result. Independent uncertainty components combine by root-sum-square, while correlated components require covariance terms. Uncertainty propagates through calculated quantities using sensitivity coefficients and first-order approximations. Calibration, traceability, checks, redundancy, closures and weighted adjustment provide practical control of survey quality.

The professional surveyor evaluates observations against their intended use, documents assumptions and limitations, and never substitutes excessive decimal places or unexamined automation for sound measurement practice.

Key Terms

Accuracy: Closeness of a measurement to an accepted or reference value.

Precision: Closeness of repeated measurements to one another.

Resolution: Smallest increment an instrument can display or distinguish.

Tolerance: Maximum permitted deviation from a specified requirement.

Uncertainty: Quantified doubt associated with a measurement result.

Error: Difference between an observed value and the unknown true value.

Gross error: A mistake such as a wrong reading, entry or station occupation.

Systematic error: Predictable or consistent error that requires correction or modelling.

Random error: Variable error whose sign and magnitude are not predictable.

Bias: Persistent signed tendency in observations.

Calibration: Comparison of an instrument with a reference standard.

Traceability: Documented relationship to an appropriate reference standard.

Residual: Difference between an observation and an estimated or adjusted value.

Variance: Mean squared dispersion measure based on a statistical sample.

Standard deviation: Square root of variance, expressing dispersion in the measurement units.

Standard error: Estimated standard deviation of a sample mean.

Redundancy: More observations than are minimally required to determine unknowns.

Misclosure: Discrepancy remaining when a survey is closed on known or identical conditions.

Weight: Relative influence assigned to an observation, commonly related to inverse variance.

Propagation of uncertainty: Calculation of output uncertainty from uncertain input quantities.

Review Questions

Short Answer Questions

1. Distinguish accuracy from precision and give one surveying example of each.
2. Why does repeating a measurement not necessarily remove a systematic error?
3. Define resolution, tolerance and uncertainty.
4. What is the purpose of an independent field check?
5. State two conditions required for the root-sum-square uncertainty rule.

Long Answer Questions

1. Explain gross, systematic and random errors, including their sources and appropriate control measures.
2. Describe how repeated observations are analyzed using the mean, residuals, sample standard deviation and standard error.
3. Explain calibration and measurement traceability in construction surveying.
4. Discuss the role of redundancy, closure and least-squares adjustment in survey quality control.
5. Explain why significant figures and uncertainty must be considered when reporting survey results.

Analytical / Application Questions

1. Five independent distance observations are 75.214, 75.219, 75.216, 75.213 and 75.218 m. Calculate the mean, sample standard deviation and standard error of the mean. State whether reporting the result to the nearest micrometre would be justified.
2. A construction point has an allowable positional tolerance of ±6 mm. Two independent observations differ by 5 mm in the east coordinate and 4 mm in the north coordinate. Determine the positional discrepancy and recommend whether the point should be accepted without further investigation.
3. A horizontal distance is calculated from a slope distance and vertical angle. Identify at least four uncertainty sources, state which may be systematic or random, and propose a field procedure that would improve confidence in the result.

Multiple Choice Questions with Answers

Question 1

Which statement best distinguishes accuracy from precision?

A. Accuracy is closeness to a reference value, while precision is agreement among repeated values.

B. Accuracy is instrument resolution, while precision is the contractual tolerance.

C. Accuracy concerns units, while precision concerns calibration dates.

D. Accuracy and precision always have identical meanings.

Correct Answer: A. Accuracy is closeness to a reference value, while precision is agreement among repeated values.

Question 2

Which type of error is most directly associated with entering the wrong station number into a field controller?

A. Random error

B. Systematic error

C. Gross error

D. Resolution error

Correct Answer: C. Gross error

Question 3

Five observations have a mean of 42.320 m and a sample standard deviation of 0.00158 m. What is the approximate standard error of the mean?

A. 0.00032 m

B. 0.00071 m

C. 0.00158 m

D. 0.00790 m

Correct Answer: B. 0.00071 m

Question 4

For independent uncertainty components of 3 mm and 4 mm, what is the combined standard uncertainty?

A. 1 mm

B. 5 mm

C. 7 mm

D. 12 mm

Correct Answer: B. 5 mm

Question 5

What does calibration primarily provide?

A. A guarantee that no future error can occur

B. A comparison with a reference standard and evidence of instrument behaviour

C. Automatic removal of all random errors

D. Permission to report unlimited decimal places

Correct Answer: B. A comparison with a reference standard and evidence of instrument behaviour

Question 6

For independent repeated observations, the standard error of the mean is calculated as:

A. s × √n

B. s + n

C. s / √n

D. s² / n

Correct Answer: C. s / √n

Question 7

A tape is consistently 4 mm too long than its nominal length. This is primarily an example of:

A. Systematic error

B. Gross error

C. Random error

D. Sampling error only

Correct Answer: A. Systematic error

Question 8

Two coordinate solutions differ by 7 mm in one coordinate and 6 mm in the other. What is their positional discrepancy approximately?

A. 1 mm

B. 6.5 mm

C. 9.2 mm

D. 13 mm

Correct Answer: C. 9.2 mm

Question 9

Why is redundancy valuable in surveying?

A. It guarantees that every observation is accurate.

B. It provides independent checks and supports estimation or adjustment.

C. It eliminates the need for calibration.

D. It permits gross errors to be averaged without investigation.

Correct Answer: B. It provides independent checks and supports estimation or adjustment.

Question 10

In weighted least squares, an observation with smaller variance generally receives:

A. A smaller weight

B. A larger weight

C. Zero weight

D. A negative weight

Correct Answer: B. A larger weight