|
Geoida
Help
|
Least Squares Adjustment - Tips and Suggestions
|
Question: Please provide some tips and suggestions
for locating and eliminating problems and refining a Least Squares
Network Adjustment
Answer:
-
- Errors, blunders, incorrect point numbers, entry errors
- If an adjustment will not converge within a few iterations but
rather increasingly diverges with unacceptable values for variance
factor etc, then probably there is at least one erroneous
measurement or blunder, or an observation with incorrectly-numbered
points. Examine the adjustment summary and scrutinise observations
close to where outlier tests have failed or where observations have
excessive residuals or corrections. To test their influence on the
network, suspect observations may be disabled or have their
a-priori standard deviations relaxed.
- Provided that provisional coordinates are reasonable, the
Relax Std Dev for tolerance errors (LS Options
button) option can be very helpful in locating outliers due to
these errors by concentrating most of the necessary adjustment into
affected observations for easier detection. Observations whose
standard deviations have been relaxed by this option may be located
by examining the summary for evidence of observations with large
residuals/corrections, large computed standard deviations or
standardised residuals, test failures, etc.
- Outliers and failed residual tests - Observations marked
as having failed the outlier test may not be in error themselves
but may have been influenced adversely by one or more adjacent
observations that are in error. Examine other surrounding
observations while applying the principles appropriate to the type
of test applied; for example where residuals are of excessive size,
the Baarda W-test and Pope's Tau-test assume that only one
observation contains a gross error so the observation with the
largest error or correction should be eliminated first. For an
excessive number of failed outlier tests it may be appropriate to
relax the confidence level for outlier detection to a more
reasonable value
- Poor quality observations - Combining observations of
poor or questionable quality with other observations of normal
precision may degrade the overall quality of the adjustment
- Minimally-constrained adjustment - It is good practice
to first check a network for internal consistency with a
minimally-constrained adjustment. Do this by limiting the
number of points or observations held fixed. For example, in a 2D
network hold only one point fixed and set one fixed azimuth for
orientation; in a 3D network with GNSS observations unfix all
points. By this means any undue strain in the network due to low
quality control points will be eliminated. Note that if the network
consists of more than one area with no connecting observations then
each area must have at least one constrained (fixed) point
- GNSS vector loop checks - Where GNSS vector observations
are included, turn on Check Loops and supply a suitable
Loop-check Tolerance PPM value prior to Compute. This will
check many of the vector loops and will list the details of any
loops that exceed the PPM tolerance and may help locate
observational or point numbering errors or entry blunders. Suspect
observations may be disabled to test their influence on the
network
- Consistently-signed DX, DY and DZ - Duplicated or
repeated GNSS vector observations between the same points must have
consistently-signed DX, DY and DZ values - different signs between
the observations will cause conflict
- Re-scale vector observations - GNSS vector observations
may require their variance-covariances to be rescaled if the values
appear too tight. Rescaling them all by the same amount may not
make any difference but by selectively rescaling according to their
reliability the adjustment may be improved
- Convergence Criteria - If the observations are of poor
quality in general, then relaxing the Convergence Criteria
in LS Options might help to get a workable convergence
- Disproportionate error ellipses - Disproportionate error
ellipses on certain points compared to most other adjusted points
suggests that there may be poor observations nearby; temporarily
holding different points 'Fixed' may help to isolate the problem
area
- Including/excluding observations - If the network
contains a range of observation types, including or excluding one
type at a time ( Observations Active checkbox) may help to
isolate where anomolous observations may be located. If both
terrestrial and GNSS observations are used together, changing the
dimension of the adjustment from 3D to 2D may be helpful
- Changing dimension - Where GNSS observations are used,
changing the dimension of the adjustment from 3D Terrestrial +
GPS to 3D Spatial may produce a better convergence
For further information please refer to the following
FAQs:
or to these topics in Geoida Help: