Indiacon CivilSubscribe

Textbook › Surveying › Chapter 1 › Lesson 1.2

1.2 Errors and their adjustment

Surveying Chapter 1: Fundamentals Under review — not yet three-checked Author: Indiacon

Learning objectives

After this lesson you should be able to:

  • Distinguish gross, systematic and random errors, and say which of them is removed by an independent check, which is reduced by a correction, and which is reduced by repeating a measurement.
  • Compute the arithmetic mean, the residuals, the sample standard deviation s = √[Σv²/(n−1)] and the standard deviation of the mean s/√n for a set of equally careful observations.
  • Form a weighted mean with weights w = 1/σ², and adjust the three angles of a plane triangle, when the observations are equally reliable, so that they sum to 180°.
  • Propagate standard deviations, to first order, for a sum and for a product, and state the assumptions of that law.

Theory

No measurement is exact. A recorded length, angle or height always differs from the value that would be obtained in an ideal experiment, and the art of surveying is not to pretend otherwise, but to keep that difference within the tolerance of the work (Lesson 1.1) and to show, by checks, that it has been kept there. This lesson is about the kinds of discrepancy that appear, how a set of repeated observations is reduced to a single adopted value, and how the uncertainty of that value is stated. Adjustment of a full traverse is Lesson 4.2; the present page is the arithmetic and the principles that later adjustment uses.

Errors are usefully put in three groups, without ranking the groups. A gross error, or mistake, is a blunder: reading a tape as 31.2 m when it was 13.2 m, booking the wrong station, or leaving a target off-centre. It is not a small scatter; it is a wrong observation, and the remedy is an independent check, not an average. A systematic error follows a rule: a tape that is too short, a line of sight that is not horizontal, temperature that has not been applied, or a compass that is drawn by a nearby fence. Repeating the same measurement with the same instrument, in the same conditions, will not cancel it; the remedy is a correction, a better procedure, or a method that reverses the effect (face-left and face-right, for example). A random error is the remaining scatter when mistakes have been rejected and systematic effects have been reduced as far as the work requires. It changes sign and size from one observation to the next, and it is this scatter that the mean and the standard deviation describe. The three must not be mixed: averaging five bookings of the same wrong station does not correct a mistake, and quoting a small standard deviation does not cancel an uncorrected tape standardisation.

Errors and their adjustmentmeanv = x - mean; s = sqrt(Σv²/(n-1))standard deviation of the mean: s/√nA+B+C= 180°equal weights: share themisclosure equallyweighted mean: w = 1/σ², x_w = Σ(w x)/ΣwIndiacon original, not to scale
Figure 1.2: Repeated length readings about their mean, and a plane triangle adjusted so that the three angles sum to 180°. Indiacon original, not to scale.

For n equally careful, independent observations of the same quantity, the least-squares (and the ordinary most-probable) value is the arithmetic mean, x̄ = (Σ x_i) / n. The residual of each observation is v_i = x_i − x̄, and the residuals sum to zero. The sample standard deviation is s = √[Σ v_i² / (n − 1)]. The divisor is n − 1, not n, because the residuals are taken from the mean of the same sample; that is the ordinary experimental standard deviation of measurement theory, as used in JCGM 100:2008 (the GUM). The standard deviation of the mean is s/√n: four times as many equally good observations halve the uncertainty of the adopted value, not the uncertainty of a single observation. If the observations are not equally careful, each is given a weight. When the weight is taken as inversely proportional to the variance, w_i = 1/σ_i², the weighted mean is x_w = Σ(w_i x_i) / Σ w_i, and the standard deviation of that mean is 1 / √(Σ w_i). A length whose random variance is taken as proportional to its length is sometimes given a weight inversely proportional to the length; that is the same rule under a different assumption, and the assumption must be stated. For a normal distribution of random error, half of the observations are expected to lie within about 0.6745 s of the mean (the 0.75-quantile of a standard normal variable is approximately 0.6745). That interval is sometimes called the probable error. It is a property of the normal curve, not a code limit, and it is not used in the worked arithmetic of this lesson.

A computed quantity has an uncertainty that follows from the uncertainties of the measurements that enter it. To first order, for uncorrelated inputs, if z = f(x, y, …), then σ_z² = (∂f/∂x)² σ_x² + (∂f/∂y)² σ_y² + ⋯. For a sum z = a + b this is σ_z = √(σ_a² + σ_b²). For a product, such as an area P = L B, it is σ_P = √[(B σ_L)² + (L σ_B)²]. The law assumes that the errors are random, that they are not materially correlated, and that they are small enough for a linear approximation to be enough. It does not remove a systematic error that has been left in L or in B. A simple, exact adjustment of the same kind is the plane triangle: the three angles should sum to 180°. If they are equally reliable, the misclosure δ = A + B + C − 180° is shared equally, and each angle is corrected by −δ/3. That is the least-squares result for three equal weights and one condition. Unequal weights, a spherical excess, and the adjustment of a traverse are later lessons and are not computed here.

Derivation

  1. The arithmetic mean minimises Σ v_i² for a set of equally weighted observations. Differentiating Σ (x_i − m)² with respect to m and setting the derivative to zero gives m = (Σ x_i)/n. The residuals from that mean sum to zero.
  2. The sample variance uses the divisor n − 1 because one degree of freedom has already been used to form the mean. Then s = √[Σ v_i² / (n − 1)], and the standard deviation of the mean, for uncorrelated observations of common s, is s/√n.
  3. If each observation has a known standard deviation σ_i and the weight is w_i = 1/σ_i², the quantity to be minimised is Σ w_i (x_i − m)². Setting the derivative to zero gives x_w = Σ(w_i x_i)/Σ w_i. The variance of x_w is 1/Σ w_i, so its standard deviation is 1/√(Σ w_i).
  4. A first-order Taylor expansion of z = f(x, y, …) gives dz = (∂f/∂x) dx + (∂f/∂y) dy + ⋯. Squaring, taking the expectation, and dropping correlation terms produces the propagation law used above. For P = L B the partial derivatives are B and L.
  5. For a plane triangle, the single condition is A + B + C = 180°. If the three observations have equal weight, the least-squares corrections are equal, each −δ/3, with δ = A + B + C − 180°.
Formulas, symbols and SI units
  • Arithmetic mean, equal weights: x̄ = (Σ x_i) / n
  • Residual: v_i = x_i − x̄, and Σ v_i = 0
  • Sample standard deviation: s = √[Σ v_i² / (n − 1)]
  • Standard deviation of the mean: s_x̄ = s / √n
  • Weight of an observation, variance known: w_i = 1 / σ_i²
  • Weighted mean: x_w = Σ(w_i x_i) / Σ w_i; standard deviation of x_w = 1 / √(Σ w_i) when w_i = 1/σ_i²
  • Propagation, uncorrelated, first order: σ_z² = (∂f/∂x)² σ_x² + (∂f/∂y)² σ_y² + ⋯
  • Sum: σ_{a+b} = √(σ_a² + σ_b²). Product P = L B: σ_P = √[(B σ_L)² + (L σ_B)²]
  • Plane triangle, equal weights: each angle is corrected by −δ/3, with δ = A + B + C − 180°
Assumptions and limits
  • The mean and the standard deviation describe random scatter after gross errors have been rejected and systematic effects have been corrected, or reduced, as far as the work requires. They do not replace an independent check.
  • The formula s = √[Σ v²/(n−1)] is for n independent observations of the same quantity, reduced to their own mean. It is not a standard deviation taken from a pre-assigned σ.
  • Weights w = 1/σ² assume that the stated σ_i are comparable and that the observations are uncorrelated. A weight proportional to 1/L is a different assumption (variance proportional to length) and must be stated if it is used.
  • The first-order propagation law assumes uncorrelated, relatively small random errors. It is not a correction for a systematic error left in an input.
  • The equal-share triangle adjustment is for a plane triangle (Lesson 1.1) and equal reliability. It is not a spherical or ellipsoidal closure, and it is not a traverse adjustment.
  • The factor 0.6745 is a labelled normal-distribution property and is not used in the examples.
  • No Indian Standard clause is cited. Official BIS surveying-error texts were not on the box at writing. JCGM 100:2008 is named as the uncertainty guide; section numbers of that guide are not cited.

Codes and standards

Indian Standards are cited by number and clause only; no clause text, tables or figures are reproduced. Other standards are listed for comparison, not ranked.

  • JCGM 100:2008, Evaluation of measurement data — Guide to the expression of uncertainty in measurement (GUM), BIPM / JCGM — named as the international uncertainty guide. Section numbers are not cited, because a clause-level reading of that text was not opened on the box at writing.
  • BIPM SI Brochure (9th edition, 2019) — SI units of length and angle.
  • No Indian Standard clause is used. Official BIS texts on surveying errors were not on the box at writing.
  • Background textbooks, not ranked, editions as in the site references register: S. K. Duggal, Surveying, Vol. 1, McGraw Hill Education (India), 4th edition, 2013; B. C. Punmia, A. K. Jain, A. K. Jain, Surveying, Vol. 1, Laxmi Publications, 17th edition, 2016; C. D. Ghilani, Elementary Surveying: An Introduction to Geomatics, Pearson, 16th edition, 2022 (published 2021).

Worked examples

Example 1 (illustrative): mean, residuals and standard deviation of a length

Problem. A distance is measured five times, equally carefully, with the following results: 126.52, 126.48, 126.55, 126.50 and 126.45 m. Find the arithmetic mean, the residuals, the sample standard deviation, and the standard deviation of the mean.

  1. Sum of the observations = 126.52 + 126.48 + 126.55 + 126.50 + 126.45 = 632.50 m. Mean x̄ = 632.50 / 5 = 126.50 m.
  2. Residuals v = x − x̄: +0.02, −0.02, +0.05, 0.00 and −0.05 m. Their sum is zero, as it must be.
  3. Σ v² = (0.02)² + (0.02)² + (0.05)² + 0 + (0.05)² = 0.0004 + 0.0004 + 0.0025 + 0.0025 = 0.0058 m².
  4. Sample standard deviation s = √[0.0058 / (5 − 1)] = √0.00145 = 0.03808 m, or 0.0381 m to four decimal places in metres.
  5. Standard deviation of the mean s_x̄ = s / √5 = 0.03808 / 2.23607 = 0.01703 m, or 0.0170 m. This is the uncertainty of the adopted 126.50 m, not of a single reading, and it assumes that the five observations are independent and equally careful, with no uncorrected systematic error.

Answer. Mean 126.50 m; residuals +0.02, −0.02, +0.05, 0.00, −0.05 m; s = 0.0381 m; s_x̄ = 0.0170 m.

Example 2 (illustrative): weighted mean of two length measurements, and a plane-triangle adjustment

Problem. (a) Two crews measure the same line. Crew A reports 150.20 m with a standard deviation of 0.04 m; crew B reports 150.32 m with a standard deviation of 0.08 m. Taking weights w = 1/σ², find the weighted mean and the standard deviation of that mean. (b) The three observed angles of a plane triangle are A = 62° 14′ 20″, B = 47° 36′ 30″ and C = 70° 10′ 40″. The observations are equally reliable. Adjust the angles so that they sum to 180°.

  1. Weights: w_A = 1 / (0.04)² = 625; w_B = 1 / (0.08)² = 156.25. The second measurement is given a quarter of the weight of the first, because its stated variance is four times as large.
  2. Weighted mean x_w = (625 × 150.20 + 156.25 × 150.32) / (625 + 156.25) = (93 875 + 23 487.5) / 781.25 = 117 362.5 / 781.25 = 150.224 m.
  3. Standard deviation of the weighted mean = 1 / √781.25 = 0.03578 m, or 0.0358 m. It is a little smaller than the better of the two input standard deviations, as it should be when a second, weaker observation is included with its proper weight.
  4. Convert each angle to seconds: A = 62 × 3 600 + 14 × 60 + 20 = 224 060″; B = 47 × 3 600 + 36 × 60 + 30 = 171 390″; C = 70 × 3 600 + 10 × 60 + 40 = 252 640″. The sum is 648 090″. 180° = 648 000″, so the misclosure is δ = +90″, or +1′ 30″.
  5. Equal-weight correction to each angle = −90″ / 3 = −30″. Adjusted angles: A = 62° 13′ 50″, B = 47° 36′ 00″, C = 70° 10′ 10″. The sum is 648 000″, which is 180° 00′ 00″, the plane-triangle condition.

Answer. x_w = 150.224 m, with standard deviation 0.0358 m. Adjusted angles 62° 13′ 50″, 47° 36′ 00″ and 70° 10′ 10″, summing to 180°.

Step-by-step procedure

  1. Inspect the observations for gross errors. A value that cannot be explained as scatter or as a known systematic effect is checked, or rejected, before any mean is formed.
  2. Apply known systematic corrections (tape standardisation, temperature, slope, dislevelment) before reducing the set. Do not bury an uncorrected systematic effect in a standard deviation.
  3. If the observations are equally careful, compute the arithmetic mean, the residuals (check that they sum to zero), s = √[Σv²/(n−1)], and s/√n as the uncertainty of the mean.
  4. If they are not equally careful, assign weights. When σ_i are known, use w_i = 1/σ_i², then the weighted mean and 1/√(Σw). State any other weight rule (for example w ∝ 1/L) explicitly.
  5. For a computed quantity, write the function, form the first-order propagation, and report the result with its propagated standard deviation. For a plane triangle of equal reliability, share the misclosure equally.
  6. Record the checks (close of a loop, second measurement, face-left and face-right) with the adopted value. An adopted mean without a check is not a finished result.
  7. Do not treat a small s as a licence to ignore the purpose-tolerance comparison of Lesson 1.1.

Site notes

  • A booking error is a gross error. If two chainmen disagree by a whole metre, measure again; do not average 17.2 m with 18.2 m and call the result 17.7 m.
  • A tape used in the afternoon sun without a temperature correction is a systematic error. Five repetitions in the same heat will agree with one another and still be wrong.
  • When one crew has a much looser stated σ than another, give it less weight; do not throw the weaker observation away, and do not give it equal weight “to be fair”.
  • A triangle that is not closed to 180° by more than the expected random scatter, after equal-share adjustment would still leave a large residual, should be re-observed, not forced.
  • Heights, angles and lengths have different systematic effects. A standard deviation computed from length residuals does not describe the uncertainty of a level run.
  • Write the units of s and of s/√n. A figure “±0.02” without a unit, and without saying whether it is s or s/√n, cannot be used later.
Common mistakes
  • Averaging in a blunder, or rejecting a reading only because it is the furthest from the mean without a check.
  • Using √[Σv²/n] for the sample standard deviation of a set that has already been reduced to its own mean, instead of √[Σv²/(n−1)].
  • Reporting s as if it were the uncertainty of the adopted mean, or reporting s/√n as if it were the scatter of a single observation.
  • Giving two observations equal weight when their standard deviations differ, or taking weights as 1/σ instead of 1/σ².
  • Sharing a triangle misclosure equally when the three angles were not equally reliable, or treating a 180° close as a geodetic (spherical) close.
  • Propagating only one of the two lengths into the uncertainty of an area, or using the propagation law to “correct” a systematic error.
  • Inventing an Indian Standard clause number for an error formula, or ranking one textbook’s treatment of probable error against another.
Quick check

Reject mistakes; correct systematic effects; then reduce random scatter. Equal weights: x̄ = (Σx)/n, s = √[Σv²/(n−1)], s_x̄ = s/√n. Unequal: w = 1/σ², x_w = Σ(wx)/Σw. Plane triangle, equal weights: share the misclosure equally. For a function, use first-order propagation of uncorrelated standard deviations.

Summary

  • Gross errors are removed by checks, systematic errors by corrections or by a reversing procedure, and random errors are described by the mean and the standard deviation.
  • For equally careful observations the adopted value is the arithmetic mean, s = √[Σv²/(n−1)], and the uncertainty of the mean is s/√n.
  • When standard deviations differ, the weighted mean with w = 1/σ² is the consistent combination, and its standard deviation is 1/√(Σw).
  • A plane triangle of equal reliability is adjusted by sharing the 180° misclosure equally. Computed quantities use first-order propagation of uncorrelated standard deviations.
  • The GUM (JCGM 100:2008) is named as the international uncertainty guide. No Indian Standard clause is cited in this lesson.

Practice questions

Tap a question to see the answer.

1The residuals of four equally careful observations are +0.03, −0.01, −0.02 and 0.00 m. What is s?

Σv² = 0.0009 + 0.0001 + 0.0004 + 0 = 0.0014 m²; s = √[0.0014 / 3] = √0.0004667 = 0.0216 m.

2For those four observations, what is the standard deviation of the mean?

s/√4 = 0.0216 / 2 = 0.0108 m.

3Two measurements of a length have σ = 0.06 m and σ = 0.03 m. What is the ratio of their weights, taking w = 1/σ²?

w₂ / w₁ = (0.06 / 0.03)² = 4. The tighter observation has four times the weight of the looser one.

4A rectangle is 80.00 m by 40.00 m, with standard deviations 0.05 m and 0.04 m respectively, uncorrelated. What is the area, and what is its propagated standard deviation?

P = 80.00 × 40.00 = 3 200 m². σ_P = √[(40.00 × 0.05)² + (80.00 × 0.04)²] = √(4.00 + 10.24) = √14.24 = 3.774 m².

5Three equally reliable plane-triangle angles sum to 180° 01′ 30″. What correction is applied to each angle?

Misclosure δ = +1′ 30″ = +90″. Each correction is −30″, or −0′ 30″.

6Why will repeating a measurement with the same tape, at the same temperature, not remove a standardisation error in that tape?

Because the standardisation error is systematic: it has the same sign and, for the same length, the same size in every repetition. A correction, or a re-standardised tape, is required.

References

Standard texts, listed in no order of merit. Our explanations, examples and diagrams are original; nothing is copied from these books.

  • S. K. Duggal, Surveying, Vol. 1, McGraw Hill Education (India), 4th edition, 2013.
  • B. C. Punmia, A. K. Jain, A. K. Jain, Surveying, Vol. 1, Laxmi Publications, 17th edition, 2016.
  • C. D. Ghilani, Elementary Surveying: An Introduction to Geomatics, Pearson, 16th edition, 2022 (published 2021).
  • JCGM 100:2008 (GUM), named as the international uncertainty guide; section numbers not cited. BIPM, The International System of Units (SI Brochure), 9th edition, 2019.
Suggest a correction

If something on this page is wrong, tell the editor. Your message is sent privately and is not published.

Sent privately to the Indiacon editor; not published.