GELE Adjustment Computations (Least Squares) — Error Propagation, Variance-Covariance and Error EllipsesDetailed Explanation
Detailed explanation of Error Propagation, Variance-Covariance and Error Ellipses for the GELE 2026. Full depth, full reasoning — exactly what you need when Professional Regulation Commission (PRC) — Board of Geodetic Engineering tests this chapter with applied or scenario-based questions in the GELE Adjustment Computations (Least Squares) subtest.
Exam context
For the Geodetic Engineer Licensure Examination, Professional Regulation Commission (PRC) — Board of Geodetic Engineering tests Adjustment Computations (Least Squares) under a "Core" label, with Error Propagation, Variance-Covariance and Error Ellipses in the 5th slot across 5 chapters. GELE candidates must clear the 70% weighted average, no sub-test below 50% cut on the 2026 paper, which draws about a meaningful share of Adjustment Computations (Least Squares) questions. Date to watch: September 2026.
Error Propagation, Variance-Covariance and Error Ellipses - Detailed Explanation
In geodetic surveying, every measured quantity carries an inherent uncertainty. When these measured quantities are combined — through arithmetic, geometric, or trigonometric operations — the resulting derived quantity inherits and accumulates those uncertainties. The discipline of Error Propagation provides the mathematical framework to track, quantify, and predict these accumulated uncertainties. In least-squares adjustment, the Variance-Covariance Matrix formally encapsulates not just the individual uncertainties of adjusted parameters (coordinates, distances, angles) but also how those uncertainties are correlated with each other. The Error Ellipse is the graphical and geometric expression of a 2-D positional uncertainty, directly derived from the variance-covariance matrix of adjusted coordinates. Together, these three tools — error propagation, the variance-covariance matrix, and the error ellipse — form the backbone of quality control in modern geodetic surveys and are frequently tested in the PRC Geodetic Engineer Licensure Examination. A clear understanding of these concepts is essential for practicing geodetic engineers working under PRS92 (Philippine Reference System of 1992), PPCS/UTM, and complying with the accuracy standards mandated by RA 8560 (Geodetic Engineering Act of the Philippines) and PD 1529 (Property Registration Decree).
Concepts
The Law of Propagation of Variances (Error Propagation)
The Law of Propagation of Variances answers a fundamental question in surveying: if I measure several quantities each with a known standard deviation, what is the standard deviation of a quantity computed from those measurements? The general law states that for a function y = f(x₁, x₂, …, xₙ) where x₁, x₂, …, xₙ are independent (uncorrelated) measurements, the variance of y is: σ²_y = (∂f/∂x₁)² σ²_x₁ + (∂f/∂x₂)² σ²_x₂ + … + (∂f/∂xₙ)² σ²_xₙ The partial derivatives ∂f/∂xᵢ are called the 'propagation coefficients' or the elements of the Jacobian matrix J. Each term (∂f/∂xᵢ)² σ²_xᵢ represents the contribution of measurement xᵢ to the total variance of y. In matrix form (the most general and powerful form), this becomes: Σ_y = J Σ_x Jᵀ where: - Σ_y is the variance-covariance matrix of the derived quantities - J is the Jacobian matrix of partial derivatives (m×n matrix for m functions of n variables) - Σ_x is the variance-covariance matrix of the original measurements - Jᵀ is the transpose of J The matrix form is critical because it correctly handles correlated observations by carrying off-diagonal covariance terms through the computation. KEY SPECIAL CASES: 1. Sum or Difference: y = x₁ ± x₂ σ_y = √(σ²_x₁ + σ²_x₂) [errors add in QUADRATURE, not linearly] 2. Linear Scaling: y = k·x σ_y = |k| · σ_x 3. Sum of n equal, independent observations: σ_sum = σ · √n [standard deviation of the SUM grows with √n] 4. Mean of n equal, independent observations: σ_mean = σ / √n [standard deviation of the MEAN shrinks with 1/√n] The contrast between cases 3 and 4 is a critical board-exam distinction. When you are summing n measurements (e.g., n taped segments), the error grows. When you are averaging n measurements to get one estimate, the error shrinks.
Examples
This illustrates the fundamental rule: independent errors add in quadrature (root-sum-of-squares), not linearly. The true propagated standard deviation (0.025 m) is less than the naive linear sum (0.035 m). In cadastral surveys under PD 1529, this distinction matters when evaluating closure tolerances.
Scenario
A cadastral survey measures two adjacent lot boundaries in Quezon City (PRS92). Segment AB = 45.230 m (σ₁ = 0.015 m) and segment BC = 62.415 m (σ₂ = 0.020 m). Find the standard deviation of the total distance AC = AB + BC.
Solution
Given: σ₁ = 0.015 m, σ₂ = 0.020 m Function: y = x₁ + x₂ Partial derivatives: ∂y/∂x₁ = 1, ∂y/∂x₂ = 1 Applying the law of propagation: σ²_AC = (∂y/∂x₁)² σ²₁ + (∂y/∂x₂)² σ²₂ σ²_AC = (1)²(0.015)² + (1)²(0.020)² σ²_AC = 0.000225 + 0.000400 σ²_AC = 0.000625 m² σ_AC = √0.000625 = 0.025 m Note: 0.015 + 0.020 = 0.035 m ≠ 0.025 m. The linear sum overestimates the error.
The partial derivatives act as amplification factors. The width W = 40 amplifies the error in L, and the length L = 80 amplifies the error in W. The larger dimension (L = 80) amplifies the error in W more, so that term dominates (2.560 vs 1.440). This type of problem appears in cadastral engineering assessments.
Scenario
A rectangular lot in a PPCS Zone III survey has measured dimensions L = 80.00 m (σ_L = 0.03 m) and W = 40.00 m (σ_W = 0.02 m). Compute the standard deviation of the computed area A = L × W.
Solution
Given: L = 80.00 m, σ_L = 0.03 m; W = 40.00 m, σ_W = 0.02 m Function: A = L × W Partial derivatives: ∂A/∂L = W = 40.00 m ∂A/∂W = L = 80.00 m Applying the law: σ²_A = (∂A/∂L)² σ²_L + (∂A/∂W)² σ²_W σ²_A = (40.00)²(0.03)² + (80.00)²(0.02)² σ²_A = (1600)(0.0009) + (6400)(0.0004) σ²_A = 1.440 + 2.560 σ²_A = 4.000 m⁴ σ_A = √4.000 = 2.000 m² Relative error check: σ_A/A = 2.000/3200 = 0.000625, or about 1 part in 1600.
This contrast is a classic board-exam trap. Summing (taping in segments) accumulates error with √n. Averaging (measuring the whole distance multiple times) improves precision by 1/√n. Both use the same formula basis but apply to completely different physical scenarios.
Scenario
A First-Order geodetic traverse leg in the Philippines is measured by taping six equal segments, each 50 m long with σ = 0.008 m per segment. (a) Find σ of the total 300-m distance. (b) If instead, the 300-m distance was measured six times independently and averaged, find σ of the mean.
Solution
Part (a) — Sum of 6 segments: n = 6, σ_segment = 0.008 m σ_total = σ√n = 0.008 × √6 = 0.008 × 2.449 = 0.0196 m ≈ 0.020 m Part (b) — Mean of 6 independent measurements: n = 6, σ_single = 0.008 m (per reading of the full 300 m) σ_mean = σ/√n = 0.008/√6 = 0.008/2.449 = 0.0033 m Comparison: Summing 6 segments: σ = 0.020 m (error grows — you accumulate 6 separate taping errors) Averaging 6 full-length readings: σ = 0.003 m (error shrinks — redundancy improves precision)
Applications
- Computing the standard deviation of derived cadastral areas from measured boundary distances and angles
- Propagating GPS baseline errors through traverse computations to assess positional accuracy at unknown stations
- Evaluating traverse closure tolerances in relation to the number of courses and individual leg precisions
- Computing the uncertainty of a computed height difference from multiple leveling sections
- Assessing the precision of computed horizontal distances derived from slope distances and zenith angles
- Quality control of survey results submitted under RA 8560 and PD 1529 — verifying that propagated errors meet accuracy specifications
Misconceptions
- WRONG: Adding standard deviations linearly (σ_total = σ₁ + σ₂). CORRECT: Add variances, then take square root: σ_total = √(σ₁² + σ₂²)
- WRONG: Using σ/√n for the error of a SUM of n segments. CORRECT: σ/√n is the error of the MEAN; σ√n is the error of the SUM
- WRONG: Forgetting that the propagation coefficient for a product A=L×W is NOT 1 for each variable — it is W for ∂A/∂L and L for ∂A/∂W
- WRONG: Ignoring covariance terms when inputs are correlated. When Σ_x has non-zero off-diagonal terms, the full matrix form Σ_y = J Σ_x Jᵀ must be used
- WRONG: Using the same propagation formula when errors are systematic (not random). The law of propagation applies to RANDOM errors only
Related Concepts
- Standard Deviation and Variance (Measures of Precision)
- Least-Squares Adjustment — Normal Equations
- Weighted Mean and Weighted Observations
- Variance-Covariance Matrix of Adjusted Parameters
- Error Ellipse and Positional Uncertainty
- Jacobian Matrix in Non-Linear Least Squares
- Closure Error and Tolerance in Traverse Surveys
Common Exam Questions
Example
Two horizontal distances with σ₁ = 0.025 m and σ₂ = 0.018 m are added. What is σ of the sum? Answer: √(0.025² + 0.018²) = √(0.000625 + 0.000324) = √0.000949 = 0.0308 m
Approach
Identify the function, compute partial derivatives (=1 for each term in a sum), apply σ_y = √(σ₁² + σ₂² + …). Convert to consistent units before computing.
Question Type
Direct Computation — Sum/Difference
Example
L=100 m (σ=0.04 m), W=50 m (σ=0.02 m). σ_A = √[(50×0.04)² + (100×0.02)²] = √[4 + 4] = √8 = 2.83 m²
Approach
For A = L × W: σ_A = √[(W·σ_L)² + (L·σ_W)²]. Identify L, W, σ_L, σ_W from the problem, substitute, compute.
Question Type
Area Computation from Measured Dimensions
Example
A distance is measured 9 times (σ=0.015 m each). σ_mean = 0.015/√9 = 0.015/3 = 0.005 m. σ_sum (if all 9 were added) = 0.015×√9 = 0.045 m.
Approach
For the SUM use σ√n; for the MEAN use σ/√n. Read the problem carefully to determine which quantity is being asked for.
Question Type
Sum vs Mean — n equal observations
Example
y = x₁² + 3x₂, x₁=5 m (σ=0.1), x₂=10 m (σ=0.2). ∂y/∂x₁=2x₁=10; ∂y/∂x₂=3. σ_y=√[(10×0.1)²+(3×0.2)²]=√[1.00+0.36]=√1.36=1.166
Approach
Write out f(x₁, x₂, …), compute each ∂f/∂xᵢ algebraically, substitute the measured values of xᵢ, then apply the propagation formula.
Question Type
General Function — Partial Derivatives Required
Key Points To Remember
- Errors propagate as variances (squares), NOT as standard deviations directly — always work with variances first, then take the square root at the end
- Independent errors add in quadrature: σ_total = √(σ₁² + σ₂² + … + σₙ²), NOT σ₁ + σ₂ + …
- The propagation coefficient for each measurement is the partial derivative ∂f/∂xᵢ, which must be evaluated at the measured values
- For a sum of n EQUAL and independent measurements: σ_sum = σ√n (grows); σ_mean = σ/√n (shrinks)
- Linear scaling: σ_y = |k|·σ_x — the constant multiplier scales the standard deviation directly
- The matrix form Σ_y = J Σ_x Jᵀ is the most general form and handles correlated inputs
- When inputs are correlated, the off-diagonal covariance terms in Σ_x must NOT be ignored
- Board exams frequently test whether you use σ√n (sum) vs σ/√n (mean) — do not confuse them
Variance-Covariance Matrix of Adjusted Parameters
In a least-squares adjustment using the observation equation model (or parametric model), we have a system of observation equations of the form: Ax̂ = L + v where: - A is the design matrix (coefficient matrix, n×m, n observations, m unknowns) - x̂ is the vector of adjusted (estimated) unknown parameters - L is the vector of observations - v is the vector of residuals - P is the weight matrix (W = σ₀² · Σ_L⁻¹) The least-squares solution is: x̂ = (AᵀPA)⁻¹ AᵀPL The VARIANCE-COVARIANCE MATRIX of the adjusted parameters x̂ is: Σ_x̂ = σ̂₀² · (AᵀPA)⁻¹ where σ̂₀² is the a posteriori variance of unit weight (reference variance), computed as: σ̂₀² = vᵀPv / (n - m) with: - vᵀPv = weighted sum of squared residuals - n = number of observations - m = number of unknowns (parameters) - (n - m) = degrees of freedom (redundancy) The matrix N = AᵀPA is called the NORMAL EQUATION MATRIX (or coefficient matrix of the normal equations). Its inverse N⁻¹ = (AᵀPA)⁻¹ is called the COFACTOR MATRIX of the parameters, Q_x̂. So: Σ_x̂ = σ̂₀² · Q_x̂ = σ̂₀² · (AᵀPA)⁻¹ For a 2-D position adjustment (e.g., horizontal control), x̂ = [x, y]ᵀ and the 2×2 variance-covariance matrix is: Σ_x̂ = | σ²_x σ_xy | | σ_xy σ²_y | where: - σ²_x = variance of the adjusted x-coordinate (diagonal element) - σ²_y = variance of the adjusted y-coordinate (diagonal element) - σ_xy = covariance between x and y (off-diagonal element) The diagonal elements give the individual coordinate standard deviations: σ_x = √(Σ_x̂[1,1]), σ_y = √(Σ_x̂[2,2]) The off-diagonal covariance σ_xy reflects the correlation between x and y uncertainties. This correlation arises from the geometry of the observations (e.g., all observations approaching from one direction create correlated x and y errors). The CORRELATION COEFFICIENT is: ρ_xy = σ_xy / (σ_x · σ_y), where -1 ≤ ρ ≤ +1 A ρ = 0 means x and y errors are independent; ρ = ±1 means perfectly correlated (degenerate geometry).
Examples
The adjusted y-coordinate is slightly less precise (7.6 mm) than the x-coordinate (6 mm). The moderate correlation (ρ = 0.316) means the x and y errors are somewhat correlated — common when observations come from limited azimuth diversity. These values would be used to construct the error ellipse at the adjusted point.
Scenario
In a simple horizontal control adjustment, the cofactor matrix of the adjusted 2-D position of a new PPCS Zone I control point is Q_x̂ = [[0.25, 0.10],[0.10, 0.40]] and σ̂₀ = 0.012 m. Compute the variance-covariance matrix and the standard deviations of the adjusted coordinates.
Solution
Given: Q_x̂ = [[0.25, 0.10],[0.10, 0.40]], σ̂₀ = 0.012 m → σ̂₀² = (0.012)² = 0.000144 m² Σ_x̂ = σ̂₀² · Q_x̂ = 0.000144 × [[0.25, 0.10],[0.10, 0.40]] = [[0.000036, 0.0000144],[0.0000144, 0.0000576]] m² Standard deviations: σ_x = √0.000036 = 0.006 m = 6 mm σ_y = √0.0000576 = 0.0076 m = 7.6 mm Covariance: σ_xy = 0.0000144 m² Correlation coefficient: ρ_xy = σ_xy/(σ_x · σ_y) = 0.0000144/(0.006 × 0.0076) = 0.0000144/0.0000456 = 0.316
The reference variance σ̂₀² = 1.0 is the ideal result when a priori weights are perfectly assigned. A value significantly different from 1.0 triggers quality control checks. In practice, σ̂₀² between 0.5 and 2.0 is often acceptable, but this should be confirmed by a formal chi-square hypothesis test at a chosen significance level (typically α = 0.05).
Scenario
After a least-squares adjustment of a leveling network (8 observations, 5 unknowns), the weighted sum of squared residuals is vᵀPv = 18.6 (in mm² with appropriate units). Compute σ̂₀².
Solution
Degrees of freedom: r = n - m = 8 - 5 = 3 A posteriori variance of unit weight: σ̂₀² = vᵀPv / r = 18.6 / 3 = 6.2 Note: σ̂₀ = √6.2 = 2.49 Interpretation: Since σ̂₀² = 6.2 >> 1.0, the residuals are significantly larger than expected from the a priori standard deviations assigned to the observations. This suggests the observations were not as precise as assumed, or there may be gross errors in the data. A chi-square test (χ² test) would be applied to assess statistical significance.
Applications
- Computing standard deviations of adjusted coordinates for geodetic control points in the PRS92 network
- Assessing the quality of horizontal and vertical control established under RA 8560 specifications
- Input data for error ellipse computation to visualize positional uncertainty at survey control points
- Performing statistical hypothesis testing (chi-square test) on the goodness of fit of an adjustment
- Propagating adjusted coordinate uncertainties into derived quantities (bearings, distances, areas) using Σ_y = J Σ_x Jᵀ
- Designing survey networks with adequate redundancy (degrees of freedom) to achieve target accuracy standards
Misconceptions
- WRONG: Thinking the diagonal of the cofactor matrix Q directly gives variances in physical units. Q must be scaled by σ̂₀² to get Σ_x̂ in physical units (m²).
- WRONG: Ignoring the off-diagonal covariance terms when computing derived quantities. When positions are correlated, σ_xy affects the uncertainty of derived bearings, distances, and areas.
- WRONG: Expecting σ̂₀² to always equal 1.0. It equals 1.0 only when the a priori weights were correctly assigned and the model is correct.
- WRONG: Using n (total observations) as degrees of freedom. Degrees of freedom = n - m (number of observations minus number of unknowns).
- WRONG: Confusing the cofactor matrix Q_x̂ with the variance-covariance matrix Σ_x̂. They differ by the scale factor σ̂₀².
Related Concepts
- Normal Equations in Least-Squares Adjustment
- A Posteriori Variance of Unit Weight (Reference Variance)
- Degrees of Freedom and Redundancy
- Chi-Square Test for Goodness of Fit
- Error Ellipse Parameters
- Weighted Least Squares and Weight Matrix
- Cofactor Matrix and Cofactor of Unit Weight
Common Exam Questions
Example
5 observations, 2 unknowns, Σv² = 12.5 mm². σ̂₀² = 12.5/(5-2) = 12.5/3 = 4.17 mm²; σ̂₀ = 2.04 mm
Approach
Identify n (observations) and m (unknowns). Compute r = n - m. Apply σ̂₀² = vᵀPv / r. If weights are equal (P = I), then vᵀPv = Σv².
Question Type
Compute σ̂₀² from given residuals and degrees of freedom
Example
Q_x̂ = [[0.5, 0.2],[0.2, 0.8]], σ̂₀ = 0.010 m. Σ_x̂ = 0.0001 × [[0.5, 0.2],[0.2, 0.8]] = [[5×10⁻⁵, 2×10⁻⁵],[2×10⁻⁵, 8×10⁻⁵]]. σ_x = √(5×10⁻⁵) = 7.07 mm; σ_y = √(8×10⁻⁵) = 8.94 mm
Approach
Simply multiply each element of Q_x̂ by σ̂₀². Then extract diagonal elements and take √ for individual standard deviations.
Question Type
Compute Σ_x̂ from Q_x̂ and σ̂₀
Example
If ρ_xy ≈ 0, the x and y errors are essentially independent (good azimuth diversity in observations). If ρ_xy ≈ ±1, the network geometry is degenerate in one direction.
Approach
State the meaning of each component: diagonal = individual variances, off-diagonal = covariances. Compute correlation coefficient ρ = σ_xy/(σ_x·σ_y). Comment on geometry quality.
Question Type
Interpret the variance-covariance matrix
Key Points To Remember
- Σ_x̂ = σ̂₀² · (AᵀPA)⁻¹ — the variance-covariance matrix of adjusted parameters scales the cofactor matrix by the reference variance
- Diagonal elements of Σ_x̂ are the variances of individual parameters (σ²_x, σ²_y); take √ to get standard deviations
- Off-diagonal elements are covariances (σ_xy) — they indicate correlation between parameters
- σ̂₀² = vᵀPv / (n-m) — the reference variance is computed from the weighted residuals divided by degrees of freedom
- Degrees of freedom = n (observations) - m (unknowns); must be positive for the problem to be over-determined
- If σ̂₀² ≈ 1.0, the a priori weights were correctly assigned; σ̂₀² >> 1 suggests observations are worse than expected
- The cofactor matrix Q_x̂ = (AᵀPA)⁻¹ is dimensionless; multiplying by σ̂₀² gives physical units (m² for coordinates)
- The variance-covariance matrix is always symmetric and positive semi-definite
Error Ellipse — Construction, Interpretation, and Confidence Levels
The Error Ellipse is the 2-D generalization of a standard deviation interval. While a 1-D uncertainty is represented by ±σ on a number line, a 2-D positional uncertainty is represented by an ellipse centered at the adjusted position. The ellipse's size, shape, and orientation are all determined by the 2×2 variance-covariance submatrix of the adjusted position: Σ_P = | σ²_x σ_xy | | σ_xy σ²_y | COMPUTING THE ERROR ELLIPSE PARAMETERS: The semi-axes of the error ellipse are found from the eigenvalues of Σ_P. The eigenvalues λ₁ and λ₂ (λ₁ ≥ λ₂) satisfy the characteristic equation: det(Σ_P - λI) = 0 (σ²_x - λ)(σ²_y - λ) - σ²_xy = 0 λ² - (σ²_x + σ²_y)λ + (σ²_x·σ²_y - σ²_xy) = 0 Solving the quadratic: λ₁,₂ = ½[(σ²_x + σ²_y) ± √((σ²_x - σ²_y)² + 4σ²_xy)] The semi-axes of the STANDARD ERROR ELLIPSE are: a = √λ₁ (semi-major axis) b = √λ₂ (semi-minor axis) The ORIENTATION ANGLE θ (azimuth from the x-axis to the major axis) is: tan(2θ) = 2σ_xy / (σ²_x - σ²_y) NOTE: θ is the angle between the x-axis (usually Easting) and the major axis of the ellipse, measured counterclockwise. This defines the azimuth of maximum uncertainty. CONFIDENCE LEVELS: The STANDARD ERROR ELLIPSE (a = √λ₁, b = √λ₂) corresponds to approximately 39.35% probability — meaning there is only a ~39% chance that the true position lies within it. This is because in 2-D, the probability contained within 1σ in each direction is less than in 1-D. For the 95% CONFIDENCE ELLIPSE, the semi-axes are scaled by c₀.₉₅: - a₀.₉₅ = √(χ²₀.₉₅,₂) · √λ₁ = 2.448 · a [using χ²₀.₉₅,₂ = 5.991, so √5.991 ≈ 2.448] - b₀.₉₅ = 2.448 · b For the 90% CONFIDENCE ELLIPSE: - Scale factor = √(χ²₀.₉₀,₂) = √4.605 ≈ 2.146 For the 99% CONFIDENCE ELLIPSE: - Scale factor = √(χ²₀.₉₉,₂) = √9.210 ≈ 3.035 INTERPRETING THE ERROR ELLIPSE: 1. CIRCULAR ELLIPSE (a ≈ b): Uniform precision in all directions. Indicates well-conditioned, geometrically diverse observation network. This is ideal. 2. ELONGATED ELLIPSE: Strongly directional uncertainty. The major axis points in the direction of weakest constraint. For example, if observations come mostly from the east and west, the east-west direction is well-constrained but north-south is poorly constrained, producing a north-south elongated ellipse. 3. LARGE ELLIPSE: Overall poor precision at the point — few redundant observations, or poor geometry. 4. SMALL ELLIPSE: High precision — many redundant observations with good geometric diversity. In the Philippine context, horizontal control points established under RA 8560 are classified into accuracy orders (First, Second, Third Order), and the error ellipse dimensions must satisfy specific tolerance requirements for each order.
Examples
The standard error ellipse (a = 41.4 mm, b = 28.1 mm) is moderately elongated — the major axis is about 1.5× the minor axis, indicating somewhat weaker constraint in the major axis direction. The 95% confidence ellipse (≈ 101 mm × 69 mm) is the one most commonly cited in accuracy specifications and survey reports.
Scenario
The 2×2 variance-covariance matrix of a new horizontal control point adjusted in the PRS92/PPCS Zone IV network is: Σ_P = [[0.0009 0.0003],[0.0003 0.0016]] m² Compute: (a) the standard error ellipse semi-axes, (b) the orientation angle, and (c) the 95% confidence ellipse semi-axes.
Solution
Given: σ²_x = 0.0009 m², σ²_y = 0.0016 m², σ_xy = 0.0003 m² Part (a) — Eigenvalues: λ = ½[(σ²_x + σ²_y) ± √((σ²_x - σ²_y)² + 4σ²_xy)] Sum = 0.0009 + 0.0016 = 0.0025 Discriminant = √[(0.0009 - 0.0016)² + 4(0.0003)²] = √[(−0.0007)² + 4(9×10⁻⁸)] = √[4.9×10⁻⁷ + 3.6×10⁻⁷] = √[8.5×10⁻⁷] = 9.220×10⁻⁴ λ₁ = ½(0.0025 + 0.0009220) = ½(0.0034220) = 0.001711 m² λ₂ = ½(0.0025 - 0.0009220) = ½(0.0015780) = 0.000789 m² Semi-axes: a = √0.001711 = 0.04136 m ≈ 41.4 mm b = √0.000789 = 0.02809 m ≈ 28.1 mm Part (b) — Orientation: tan(2θ) = 2(0.0003)/(0.0009 - 0.0016) = 0.0006/(-0.0007) = -0.857 2θ = arctan(-0.857) = -40.6° → θ = -20.3° (or 159.7° measured CCW from x-axis) Part (c) — 95% Confidence Ellipse: Scale factor = √5.991 = 2.448 a₀.₉₅ = 2.448 × 0.04136 = 0.1013 m ≈ 101 mm b₀.₉₅ = 2.448 × 0.02809 = 0.0688 m ≈ 68.8 mm
Equal diagonal elements with zero covariance always produce a circular error ellipse. This means the survey geometry provided equal strength in all horizontal directions — ideal for establishing new control points. In practice, perfect isotropy is rarely achieved, but modern GPS-based control surveys in the PRS92 network approach this ideal.
Scenario
A horizontal control adjustment yields σ_x = 0.012 m, σ_y = 0.012 m, and σ_xy = 0 (covariance = zero). Describe and compute the error ellipse.
Solution
Since σ²_x = σ²_y = (0.012)² = 0.000144 m² and σ_xy = 0: Eigenvalues: λ₁ = ½[(0.000144 + 0.000144) + √((0 )² + 0)] = 0.000144 m² λ₂ = 0.000144 m² Semi-axes: a = b = √0.000144 = 0.012 m = 12 mm Since a = b, the ellipse is a CIRCLE with radius 12 mm. Orientation: Undefined (all directions equally constrained). This is a perfectly isotropic uncertainty — the point is equally well-determined in all horizontal directions. This is the ideal result for a well-designed geodetic network with observations from diverse azimuths.
Applications
- Visual quality assessment of adjusted horizontal control point positions in the PRS92/PPCS network
- Identifying directions of weak geometric strength in a survey network, guiding additional observation planning
- Compliance verification with NAMRIA horizontal accuracy standards for geodetic control under RA 8560
- Communication of positional uncertainty to cadastral survey clients and land registration authorities under PD 1529
- Network design and pre-analysis — computing predicted error ellipses before fieldwork to optimize observation plans
- Relative error ellipses between two adjacent control points to assess the precision of relative positioning
Misconceptions
- WRONG: Thinking the standard error ellipse represents 68% or 95% probability. The standard error ellipse in 2-D represents only ~39% probability.
- WRONG: Confusing the 95% scale factor of 2.448 (for 2-D ellipse) with the 1-D 95% factor of 1.96. The chi-square distribution with 2 degrees of freedom is used for 2-D ellipses.
- WRONG: Thinking a circular error ellipse means the survey has no error. It means the error is equal in all directions — there is still uncertainty, just isotropic.
- WRONG: Computing semi-axes as σ_x and σ_y directly. These are the coordinate standard deviations along the axis system, not the ellipse semi-axes (unless σ_xy = 0).
- WRONG: Interpreting the major axis direction as the direction of most data — it is the direction of LEAST constraint (maximum uncertainty).
Related Concepts
- Eigenvalues and Eigenvectors of a Matrix
- Chi-Square Distribution and Hypothesis Testing
- Variance-Covariance Matrix of Adjusted Parameters
- Circular Error Probable (CEP) in GPS positioning
- Relative Error Ellipse Between Two Adjusted Points
- Network Geometry and Strength of Figure
- Confidence Intervals and Probability in 2-D
Common Exam Questions
Example
Σ_P = [[4, 2],[2, 9]] (in mm²). λ₁,₂ = ½[(13) ± √(25 + 16)] = ½[13 ± √41] = ½[13 ± 6.40]. λ₁ = 9.70 mm², λ₂ = 3.30 mm². a = √9.70 = 3.11 mm; b = √3.30 = 1.82 mm.
Approach
Apply λ₁,₂ = ½[(σ²_x + σ²_y) ± √((σ²_x - σ²_y)² + 4σ²_xy)]. Then a = √λ₁, b = √λ₂. Check: a ≥ b always.
Question Type
Compute eigenvalues and error ellipse semi-axes from a given Σ_P
Example
Standard ellipse: a = 5 mm, b = 3 mm. 95% ellipse: a₀.₉₅ = 2.448 × 5 = 12.24 mm; b₀.₉₅ = 2.448 × 3 = 7.34 mm.
Approach
Multiply both semi-axes by √(χ²₀.₉₅,₂) = √5.991 ≈ 2.448. Memorize this scale factor for the board exam.
Question Type
Scale standard ellipse to 95% confidence
Example
If a = 25 mm (oriented N-S) and b = 5 mm, the survey is much weaker in the N-S direction. This could be caused by all observations coming from the east and west with no observations from the north or south.
Approach
Compare a and b. If a >> b, weak geometry in major axis direction. If a ≈ b, good isotropic geometry. State which direction is weakest.
Question Type
Interpret ellipse shape and relate to survey geometry
Example
σ²_x = 9, σ²_y = 4, σ_xy = 2. tan(2θ) = 2(2)/(9-4) = 4/5 = 0.800. 2θ = 38.66°. θ = 19.33° (major axis at 19.33° CCW from x-axis)
Approach
Use tan(2θ) = 2σ_xy/(σ²_x - σ²_y). Solve for 2θ, then θ. Handle the quadrant carefully based on signs of numerator and denominator.
Question Type
Compute orientation angle of error ellipse
Key Points To Remember
- Error ellipse semi-axes come from the EIGENVALUES of the 2×2 variance-covariance matrix: a = √λ₁, b = √λ₂
- The standard error ellipse corresponds to ~39% probability in 2-D (NOT 68% as in 1-D)
- For 95% confidence ellipse: multiply semi-axes by √(χ²₀.₉₅,₂) = √5.991 ≈ 2.448
- Orientation angle: tan(2θ) = 2σ_xy/(σ²_x - σ²_y) — the major axis points in the direction of maximum uncertainty
- A circular error ellipse (a ≈ b) indicates uniform precision and good network geometry
- An elongated error ellipse indicates weak geometry in the direction of the major axis — the network is poorly constrained in that direction
- The error ellipse is centered at the adjusted position (coordinates from least-squares solution)
- Point Position Uncertainty (PPU) is sometimes expressed as the radius of a circle containing the error ellipse — used in specifications
Practice Problems
The three independent segment errors combine in quadrature, yielding σ_D = 0.016 m. This is significantly less than the naive linear sum of 0.026 m. The relative precision of approximately 1:4600 would be checked against the accuracy requirements for cadastral surveys under PD 1529 and RA 8560 (Third-Order: 1:5000; Second-Order: 1:20,000). This survey just meets Third-Order requirements.
Problem
PROBLEM 1 — Error Propagation (Sum): In a boundary survey under PD 1529, a property line is measured in three consecutive segments: d₁ = 23.456 m (σ₁ = 0.008 m), d₂ = 31.789 m (σ₂ = 0.012 m), d₃ = 18.234 m (σ₃ = 0.006 m). Calculate the standard deviation of the total boundary length D = d₁ + d₂ + d₃.
Solution
Function: D = d₁ + d₂ + d₃ Partial derivatives: ∂D/∂d₁ = ∂D/∂d₂ = ∂D/∂d₃ = 1 Applying the law of propagation of variances: σ²_D = (1)²(0.008)² + (1)²(0.012)² + (1)²(0.006)² σ²_D = 0.000064 + 0.000144 + 0.000036 σ²_D = 0.000244 m² σ_D = √0.000244 = 0.01562 m ≈ 0.016 m Total boundary length: D = 23.456 + 31.789 + 18.234 = 73.479 m Relative precision: σ_D/D = 0.016/73.479 = 1/4,593 ≈ 1:4600 Comparison — if errors were added linearly (WRONG): σ_wrong = 0.008 + 0.012 + 0.006 = 0.026 m [overestimates the error by 67%]
Note that the W²·σ²_L term (9.000) and L²·σ²_W term (12.960) are both significant, but the width uncertainty (σ_W) contribution through the larger length (L = 120 m) dominates. To improve compliance, the surveyor should focus on reducing σ_W first. This problem demonstrates practical compliance assessment against survey accuracy standards, a critical skill tested in the PRC board exam.
Problem
PROBLEM 2 — Error Propagation (Area Computation): A rectangular agricultural lot in Pampanga (PPCS Zone III) has measured dimensions L = 120.00 m (σ_L = 0.04 m) and W = 75.00 m (σ_W = 0.03 m). (a) Compute the area A and its standard deviation σ_A. (b) Express the relative error in the area. (c) If the government standard requires σ_A/A ≤ 1/2000, does this survey comply?
Solution
Given: L = 120.00 m, σ_L = 0.04 m; W = 75.00 m, σ_W = 0.03 m Part (a): A = L × W = 120.00 × 75.00 = 9,000.00 m² Partial derivatives: ∂A/∂L = W = 75.00 m ∂A/∂W = L = 120.00 m Propagation: σ²_A = (75.00)²(0.04)² + (120.00)²(0.03)² σ²_A = (5625)(0.0016) + (14400)(0.0009) σ²_A = 9.000 + 12.960 σ²_A = 21.960 m⁴ σ_A = √21.960 = 4.686 m² Part (b): Relative error = σ_A/A = 4.686/9000 = 0.000521 = 1/1921 ≈ 1:1900 Part (c): Required: σ_A/A ≤ 1/2000 = 0.000500 Actual: σ_A/A = 0.000521 > 0.000500 The survey does NOT comply — the relative area error (1:1921) exceeds the 1:2000 requirement. The surveyor must remeasure with tighter tolerances (e.g., σ_L ≤ 0.035 m and σ_W ≤ 0.025 m).
This contrast illustrates a critical conceptual distinction. In Case (a), you are summing 8 independent error sources — the total error accumulates. In Case (b), you are averaging 8 estimates of the same quantity — redundancy improves precision. The formula σ√n applies to sums (accumulation), while σ/√n applies to means (averaging). Confusing these is one of the most common board-exam errors.
Problem
PROBLEM 3 — Sum vs. Mean Contrast: A 200-m baseline on Palawan is (a) measured in 8 equal taped segments of 25 m each, each with σ = 0.005 m, OR (b) measured as the full 200 m in 8 independent trials and averaged. Compute σ for the final baseline distance in each case.
Solution
Case (a) — Sum of 8 segments (physical taping in sections): n = 8 segments, σ_per_segment = 0.005 m σ_total = σ√n = 0.005 × √8 = 0.005 × 2.828 = 0.01414 m ≈ 0.014 m Physical meaning: Each taped segment introduces an independent random error of 0.005 m. When 8 segments are taped end-to-end, these errors accumulate. Case (b) — Mean of 8 independent full-length measurements: n = 8 measurements, σ_single = 0.005 m (assumed same precision per full-length measurement) σ_mean = σ/√n = 0.005/√8 = 0.005/2.828 = 0.001768 m ≈ 0.0018 m Physical meaning: Each of the 8 independent full-length measurements is slightly different due to random errors. Averaging reduces the random error by 1/√8. Comparison: Case (a): σ = 0.014 m (error GROWS — taping in segments) Case (b): σ = 0.0018 m (error SHRINKS — averaging redundant measurements) Ratio: 0.014/0.0018 ≈ 7.8× more precise to average than to tape in segments
The complete error ellipse analysis provides: the magnitude of positional uncertainty in the worst direction (a = 21 mm standard), the best direction (b = 14 mm), the direction of worst constraint (NNW), and the 95% confidence limits (≈51 mm × 33 mm). This information guides both quality reporting and further survey planning. The moderate elongation (1.53:1) is acceptable for many applications but may indicate insufficient observations from the NNW direction.
Problem
PROBLEM 4 — Error Ellipse Computation: After a horizontal control adjustment of a new Second-Order PRS92 control monument, the 2×2 variance-covariance matrix of the adjusted E, N coordinates is: Σ_P = [[0.000225 0.000090],[0.000090 0.000400]] m² Compute: (a) the standard error ellipse semi-axes a and b, (b) the orientation angle θ, and (c) the 95% confidence ellipse semi-axes. (d) Comment on the survey geometry.
Solution
Given: σ²_E = 0.000225 m², σ²_N = 0.000400 m², σ_EN = 0.000090 m² Part (a) — Eigenvalues: λ₁,₂ = ½[(σ²_E + σ²_N) ± √((σ²_E - σ²_N)² + 4σ²_EN)] Sum = 0.000225 + 0.000400 = 0.000625 Discriminant term: (σ²_E - σ²_N) = 0.000225 - 0.000400 = -0.000175 (σ²_E - σ²_N)² = (0.000175)² = 3.0625×10⁻⁸ 4σ²_EN = 4(0.000090)² = 4(8.1×10⁻⁹) = 3.24×10⁻⁸ Discriminant = √(3.0625×10⁻⁸ + 3.24×10⁻⁸) = √(6.3025×10⁻⁸) = 2.5105×10⁻⁴ λ₁ = ½(0.000625 + 0.00025105) = ½(0.00087605) = 4.3803×10⁻⁴ m² λ₂ = ½(0.000625 - 0.00025105) = ½(0.00037395) = 1.8698×10⁻⁴ m² Semi-axes: a = √(4.3803×10⁻⁴) = 0.02093 m = 20.93 mm b = √(1.8698×10⁻⁴) = 0.01368 m = 13.68 mm Part (b) — Orientation: tan(2θ) = 2σ_EN/(σ²_E - σ²_N) = 2(0.000090)/(-0.000175) = 0.000180/(-0.000175) = -1.0286 2θ = arctan(-1.0286) = -45.8° (in 2nd quadrant since denominator < 0, numerator > 0) 2θ = 180° - 45.8° = 134.2° θ = 67.1° (major axis at 67.1° CCW from E-axis, i.e., oriented more toward N) Part (c) — 95% Confidence Ellipse: Scale factor = √5.991 = 2.448 a₀.₉₅ = 2.448 × 20.93 = 51.24 mm b₀.₉₅ = 2.448 × 13.68 = 33.49 mm Part (d) — Geometry Comment: Elongation ratio: a/b = 20.93/13.68 = 1.53 — moderately elongated. The major axis (direction of maximum uncertainty) is at ≈67° from East (roughly NNW direction). This suggests the survey network had fewer or weaker observations from the NNW-SSE direction. For Second-Order classification, compare 95% semi-major axis (51 mm) against the applicable NAMRIA specification.
The covariance term σ_EN = 0.0002 m² contributes +0.000199 m² to the total variance, which increases σ_d from 25.5 mm (ignoring covariance) to 29.1 mm (correct result). This 14% underestimation when covariance is ignored demonstrates why the full matrix form Σ_y = J Σ_x Jᵀ must be used when inputs are correlated. The Jacobian for a distance function evaluated at the actual coordinate values is a standard board-exam computation.
Problem
PROBLEM 5 — Matrix Form Propagation: A derived quantity is computed from two adjusted coordinates: d = √[(E₂-E₁)² + (N₂-N₁)²]. Given adjusted values E₁ = 100.000 m, N₁ = 200.000 m, E₂ = 150.000 m, N₂ = 250.000 m (all units in m), and the variance-covariance matrix of point 2 only (point 1 assumed fixed): Σ_{P₂} = [[0.0004 0.0002],[0.0002 0.0009]] m² Compute the standard deviation of the derived distance d.
Solution
Step 1 — Compute the distance: ΔE = E₂ - E₁ = 150.000 - 100.000 = 50.000 m ΔN = N₂ - N₁ = 250.000 - 200.000 = 50.000 m d = √(50² + 50²) = √(2500 + 2500) = √5000 = 70.711 m Step 2 — Jacobian (partial derivatives): d = √(ΔE² + ΔN²) ∂d/∂E₂ = ΔE/d = 50.000/70.711 = 0.70711 ∂d/∂N₂ = ΔN/d = 50.000/70.711 = 0.70711 Jacobian row vector: J = [0.70711 0.70711] Step 3 — Apply Σ_d = J · Σ_{P₂} · Jᵀ: First compute J · Σ_{P₂}: J · Σ = [0.70711 0.70711] × [[0.0004 0.0002],[0.0002 0.0009]] = [0.70711(0.0004) + 0.70711(0.0002), 0.70711(0.0002) + 0.70711(0.0009)] = [0.000283 + 0.000141, 0.000141 + 0.000636] = [0.000424, 0.000777] Then multiply by Jᵀ: Σ_d = [0.000424 0.000777] × [0.70711; 0.70711] = 0.000424(0.70711) + 0.000777(0.70711) = 0.000300 + 0.000549 = 0.000849 m² σ_d = √0.000849 = 0.02914 m ≈ 0.029 m = 29 mm Note: If we had naively computed ignoring covariance: σ²_d_naive = (0.70711)²(0.0004) + (0.70711)²(0.0009) = 0.5(0.0004 + 0.0009) = 0.000650 m² σ_d_naive = 0.0255 m [underestimates by ~12%]
Exam Preparation Tips
- MASTER the three key formulas by heart: (1) σ_y² = Σ(∂f/∂xᵢ)²σᵢ² for error propagation; (2) Σ_x̂ = σ̂₀² · (AᵀPA)⁻¹ for the covariance matrix; (3) λ₁,₂ = ½[(σ²_x + σ²_y) ± √((σ²_x - σ²_y)² + 4σ²_xy)] for eigenvalues. These three formulas cover 80% of board exam questions in this topic.
- NEVER add standard deviations linearly for independent errors — always add variances (squares) first, then take the square root. This is the single most common error in board exam answers for error propagation problems.
- MEMORIZE the contrast: σ√n for the SUM of n equal measurements (error grows); σ/√n for the MEAN of n equal measurements (error shrinks). Write this pair on your formula sheet and review it the morning of the exam.
- MEMORIZE the 95% confidence ellipse scale factor: √5.991 ≈ 2.448. This is derived from the chi-square distribution with 2 degrees of freedom. Also know that the standard error ellipse ≈ 39% probability (NOT 68%).
- For area propagation problems (A = L × W), immediately write the two partial derivatives: ∂A/∂L = W and ∂A/∂W = L. The larger dimension amplifies the error of the other measurement more.
- For the variance-covariance matrix interpretation: diagonal elements = variances (σ²); off-diagonal = covariances (σ_xy). Standard deviations are square roots of diagonal elements only.
- When computing error ellipses, always check that your eigenvalues are both positive (since Σ must be positive definite) and that λ₁ ≥ λ₂ (semi-major axis ≥ semi-minor axis).
- For the orientation angle, use tan(2θ) = 2σ_xy/(σ²_x - σ²_y) and carefully handle the quadrant. When σ²_x = σ²_y, the denominator is zero — the ellipse is circular and θ is undefined.
- PRACTICE the full workflow for error ellipse problems: Σ_P → eigenvalues → semi-axes → orientation → scale for confidence level. Each step should take less than 2 minutes with practice.
- Connect to Philippine practice: Know that NAMRIA classifies horizontal control into accuracy orders under RA 8560, and that error ellipse dimensions are used to verify compliance. First-Order horizontal control requires 1:100,000 accuracy; Second-Order: 1:50,000; Third-Order: 1:10,000. These ratios relate directly to propagated coordinate uncertainties.
- For board exam multiple-choice, wrong answer choices often correspond to: (1) linear addition of errors instead of quadrature, (2) using σ/√n instead of σ√n for a sum, or (3) using coordinate σ_x directly as the ellipse semi-axis instead of the eigenvalue √λ. Recognize these traps.
- REVIEW the physical meaning of each quantity: σ̂₀² ≈ 1.0 means good model fit; elongated ellipse means weak geometry in major axis direction; large σ_xy relative to σ_x·σ_y means the observations constrain position better in one direction than the other. Conceptual interpretation questions are common in the essay/problem sections.
In summary
Error Propagation, the Variance-Covariance Matrix, and Error Ellipses form an integrated framework for quantifying and communicating positional uncertainty in geodetic surveys. The Law of Propagation of Variances provides the mathematical bridge between the precision of individual measurements and the precision of quantities derived from them — a bridge built from partial derivatives and the fundamental principle that independent errors add in quadrature, not linearly. The Variance-Covariance Matrix of adjusted parameters, computed as Σ_x̂ = σ̂₀² (AᵀPA)⁻¹, is the formal product of a least-squares adjustment that encodes both the individual uncertainties of adjusted coordinates and their inter-correlations. The Error Ellipse translates this abstract matrix into a concrete geometric representation of 2-D positional uncertainty, with semi-axes derived from eigenvalues and orientation from the principal directions of the covariance matrix. For the PRC Geodetic Engineer Licensure Examination, these concepts are tested both computationally (applying the propagation formula, computing eigenvalues, scaling for confidence levels) and conceptually (interpreting elongated ellipses, understanding the Sum vs. Mean distinction, recognizing when covariances cannot be ignored). Mastery of the three key formulas — the propagation law, the covariance matrix formula, and the eigenvalue equations — combined with a firm grasp of their physical meaning in terms of survey geometry and accuracy standards under RA 8560 and PD 1529, will equip any geodetic engineering reviewee to handle this topic confidently on examination day and in professional practice. Remember the two most common pitfalls: never add standard deviations linearly (always add variances), and never confuse σ√n (error of a sum) with σ/√n (error of a mean). With these anchored firmly in mind and the worked examples practiced thoroughly, this chapter's board exam questions become highly manageable.
Ready to practise for the GELE 2026?
Super Tutor's AI review plan adapts to your weak areas and builds a weekly practice schedule around your target GELE exam date.