6.5 Geodetic Computations and Network Adjustments

1.5 Geodetic Computations & Network Adjustments

All physical measurements in surveying—whether angles, distances, zenith angles, or GNSS baseline vectors—contain unavoidable observational errors. In geodetic networks, redundant observations are purposefully collected to detect blunders, evaluate measurement precision, and ensure structural reliability. Network adjustment is the mathematical procedure used to resolve observational discrepancies (misclosures) and compute unique, statistically optimal coordinates for network control stations. A candidate surveyor must demonstrate complete fluency in empirical adjustment methods (Bowditch and Transit rules) as well as rigorous Least Squares Estimation (LSE).

---

1. Control Networks & Angular / Linear Misclosure Standards

Geodetic control networks are categorized into traverses, triangulation, trilateration, and hybrid GNSS vector networks.

Angular Misclosure in Closed Traverses

For a closed polygonal traverse with $n$ interior angles, the theoretical geometric sum of interior angles is $(n - 2) \times 180^\circ$. The angular misclosure ($w_\theta$) is the difference between observed sum and theoretical sum: $w_\theta = \sum \theta_{obs} - (n - 2) \times 180^\circ$

SURCON maximum allowable angular misclosure limits ($W_{max}$) are specified by traverse order:

• First-Order Traverse: $W_{max} = \pm 10'' \sqrt{n}$
• Second-Order Class A Traverse: $W_{max} = \pm 20'' \sqrt{n}$
• Third-Order Cadastral Traverse: $W_{max} = \pm 30'' \sqrt{n}$

Relative Fractional Linear Misclosure

After balancing angles, partial Easting changes (departures $\Delta E = L \sin \alpha$) and Northing changes (latitudes $\Delta N = L \cos \alpha$) are summed around the loop. The linear misclosure ($FL$) and fractional closure ratio ($1 : K_{rel}$) are computed as: $FL = \sqrt{(\sum \Delta E)^2 + (\sum \Delta N)^2}, \quad K_{rel} = \frac{\sum L}{FL}$

SURCON accuracy thresholds mandate $K_{rel} \ge 1 : 50,000$ for First-Order, $1 : 20,000$ for Second-Order Class A, and $1 : 10,000$ for Third-Order cadastral boundaries.

---

2. Empirical Adjustment Rules: Bowditch vs. Transit

Before digital computers enabled routine matrix algebra, empirical rules were used to distribute linear misclosures across traverse stations:

A. Bowditch (Compass) Rule

• Fundamental Assumption: Linear distance errors and angular measurement errors exert an equal influence on positional error. Distance error is assumed proportional to $\sqrt{L}$, and angular displacement error is proportional to $\sqrt{L}$.
• Mathematical Formula: Corrections are applied directly proportional to leg length ($L_i$) relative to total perimeter length ($\sum L$): $\delta E_i = -\left( \frac{L_i}{\sum L} \right) \cdot \sum \Delta E, \quad \delta N_i = -\left( \frac{L_i}{\sum L} \right) \cdot \sum \Delta N$
• Usage: Standard rule for general cadastral boundary traverses where distances are measured with tape/EDM and angles with theodolites.

B. Transit Rule

• Fundamental Assumption: Angular measurements are significantly more precise than distance measurements.
• Mathematical Formula: Corrections are distributed proportional to the absolute magnitude of individual departure or latitude segments: $\delta E_i = -\left( \frac{|\Delta E_i|}{\sum |\Delta E|} \right) \cdot \sum \Delta E, \quad \delta N_i = -\left( \frac{|\Delta N_i|}{\sum |\Delta N|} \right) \cdot \sum \Delta N$
• Usage: Preferred when grid directions are aligned parallel to primary axes or when transit-theodolite angular work exceeds distance accuracy.

---

3. Least Squares Estimation (LSE): Mathematical Derivation

Least Squares Estimation is the mathematically optimal adjustment method. It handles complex, multi-redundant networks without splitting loops and accounts for varying measurement precision via observational weights.

A. Principle of Least Squares

LSE minimizes the sum of weighted squared residuals ($\Phi$): $\Phi = \mathbf{v}^T \mathbf{P} \mathbf{v} = \sum_{i=1}^{n} w_i v_i^2 = \text{minimum}$ Where $\mathbf{v}$ is the $n \times 1$ residual vector, and $\mathbf{P}$ is the $n \times n$ observational weight matrix (inversely proportional to observational covariance matrix $\mathbf{\Sigma}_l$, so $\mathbf{P} = \sigma_0^2 \mathbf{\Sigma}_l^{-1}$).

B. Observation Equation Model (Parametric LSE)

Non-linear field observation equations (distance $L = \sqrt{(E_j - E_i)^2 + (N_j - N_i)^2}$, azimuth $\alpha = \arctan \frac{E_j - E_i}{N_j - N_i}$) are linearized using Taylor series expansion around approximate coordinates $(X_0)$: $\mathbf{v} = \mathbf{A} \mathbf{x} - \mathbf{l}$

• $\mathbf{A}$ ($n \times u$): Design Matrix (Jacobian matrix containing partial derivatives of observation functions with respect to unknown parameters).
• $\mathbf{x}$ ($u \times 1$): Vector of Unknown Parameter Corrections ($ \hat{x} = X - X_0$).
• $\mathbf{l}$ ($n \times 1$): Misclosure Vector (Observed value minus computed value from approximate coordinates: $\mathbf{l} = L_{obs} - f(X_0)$).

C. Normal Equations & Solution Vector

Minimizing $\mathbf{v}^T \mathbf{P} \mathbf{v}$ yields the fundamental Normal Equations: $(\mathbf{A}^T \mathbf{P} \mathbf{A}) \mathbf{x} = \mathbf{A}^T \mathbf{P} \mathbf{l}$ Setting $\mathbf{N} = \mathbf{A}^T \mathbf{P} \mathbf{A}$ (Normal Equations Matrix), the unique parameter solution vector is solved by matrix inversion: $\mathbf{\hat{x}} = (\mathbf{A}^T \mathbf{P} \mathbf{A})^{-1} \mathbf{A}^T \mathbf{P} \mathbf{l} = \mathbf{N}^{-1} \mathbf{u}$

Where $\mathbf{u} = \mathbf{A}^T \mathbf{P} \mathbf{l}$. Post-adjustment observational residuals are then updated as $\mathbf{v} = \mathbf{A} \mathbf{\hat{x}} - \mathbf{l}$.

---

4. Variance-Covariance Matrix & Error Ellipses

A primary advantage of LSE is that it provides full statistical error propagation for adjusted station coordinates via the Parameter Variance-Covariance Matrix ($ \mathbf{C}_x$):

$\mathbf{C}_x = \hat{\sigma}_0^2 (\mathbf{A}^T \mathbf{P} \mathbf{A})^{-1} = \begin{bmatrix} \sigma_E^2 & \sigma_{EN} \\ \sigma_{EN} & \sigma_N^2 \end{bmatrix}$ Where $\hat{\sigma}_0^2 = \frac{\mathbf{v}^T \mathbf{P} \mathbf{v}}{n - u}$ is the a posteriori variance of unit weight ($df = n - u$ degrees of freedom).

                  N (North)
                  ^        .---.
                  |       /     \  Semi-major axis (a_ellipse)
                  |      /   *   \------> Azimuth angle (theta)
                  |     (  Center ) 
                  |      \       /  Semi-minor axis (b_ellipse)
                  |       '---'
                  +-------------------> E (East)

2D Positional Error Ellipse Geometry:

To visualize 2D coordinate uncertainty at a station, an error ellipse is computed from the element covariance terms $(\sigma_E^2, \sigma_N^2, \sigma_{EN})$:

1. Semi-Major Axis ($a_{ellipse}$) & Semi-Minor Axis ($b_{ellipse}$): $$$$

\sigma_N^2}{2} \right)^2 + \sigma_{EN}^2 }$$a_{ellipse} = c_k \sqrt{\lambda_1}, \quad b_{ellipse} = c_k \sqrt{\lambda_2}$$ (Where $c_k = 1.0$ for $39.4%$ standard error ellipse, or $c_k = 2.447$ for $95%$ confidence region).

1. Orientation Azimuth ($\theta$): $\tan 2\theta = \frac{2 \sigma_{EN}}{\sigma_E^2 - \sigma_N^2}$

If $\sigma_{EN} = 0$ and $\sigma_E = \sigma_N$, the error ellipse reduces to a perfect circle (isotropic precision).