8.6 Least Squares, Residuals, and Error Propagation

Least Squares Concepts for Survey Computations

Least squares is included in the official FS Survey Computations and Computer Applications area. The exam can test concepts and small computations without requiring a full professional network adjustment by hand. The essential idea is that redundant observations are adjusted so the weighted sum of squared residuals is minimized. Redundancy matters because without extra observations, there is no independent check on error.

An observation is a measured distance, angle, elevation difference, coordinate, or other quantity. A residual is the difference or correction associated with the adjusted result. In survey language, residuals help evaluate how well observations fit the model. A residual is not proof of a mistake by itself, but a large residual compared with expected precision can indicate a blunder, a weak model, or underestimated uncertainty.

Weights are commonly related to inverse variance. If one distance has a standard deviation of 0.01 ft and another has 0.03 ft, the first is more precise and should receive greater weight. Because variance is standard deviation squared, the weight ratio is not simply 3 to 1. It is related to 1/0.01 squared versus 1/0.03 squared, or 9 to 1 if all else is equal.

A simple weighted mean demonstrates the idea. If elevation difference observations are 10.02 ft with weight 4 and 9.98 ft with weight 1, the weighted mean is (4 x 10.02 + 1 x 9.98)/(4 + 1) = 10.012 ft. The more reliable observation pulls the result closer to itself. If equal weights were used, the mean would be 10.000 ft.

Error propagation estimates how uncertainty moves through formulas. If a computed distance depends on two coordinate differences, uncertainty in each coordinate affects the distance. If an area is computed from many coordinates, coordinate uncertainty affects area. The FS exam may ask for the conceptual effect or a simple propagation using given formulas. The correct approach is to combine independent random uncertainties statistically, not by casually adding maximum possible errors unless the problem specifies a worst-case method.

Least squares does not rescue bad data automatically. A wrong point ID, a busted prism height, or a swapped sign can distort an adjustment and still produce a numerical answer. Good practice includes preadjustment checks, residual review, network geometry evaluation, and documentation of excluded observations. In exam terms, choose answers that investigate large residuals rather than blindly accepting adjusted results.

A least-squares workflow at the FS level is:

1. Identify observations, unknowns, and constraints.
2. Confirm there is redundancy for adjustment and checking.
3. Assign weights from stated precision or variance information.
4. Compute or interpret adjusted values and residuals.
5. Compare residuals with expected observation precision.
6. Investigate blunders or model errors before final reporting.
7. Report adjusted values with uncertainty appropriate to the data.

This topic bridges computations and measurement science. The candidate does not need to memorize software menus. The candidate should understand why a weighted adjustment differs from an arithmetic average and why residuals, standard deviations, and checks matter in survey control.