Activity Feed › Discussion Forums › Strictly Surveying › Least Squares Traverse Method
Least Squares Traverse Method
Posted by Texian on March 9, 2018 at 10:45 pmWould anyone be willing to share field traverse procedures to use for least squares adjustment? All the companies I’ve worked for used the Compass/Transit method for adjustment, but I would like to begin familiarizing myself with least squares. There may not be any difference in how you run your traverse, but I’m curious to know.
Much thanks in advance
luke-j-crawford replied 6 years ago 12 Members · 15 Replies- 15 Replies
The LS answer will usually be quite similar to the Compass Rule answer if all you have is basic traverse data. Where LS shines is if you have more data to combine. Redundancy adds a lot to accuracy.
You can mix GNSS data with total station data (or compass and chain data, for that matter) around the traverse, and assign realistic standard errors to each type of measurement to get a best estimate based on all the data. If you can see between non-adjacent points in the traverse, measuring extra angles and distances will add strength to the solution, and this is something not easy to accommodate with other methods.
When you get done, it will give you measures of fit that are more indicative of quality than a simple closure value.
.It might be useful to know which software you plan to use. If the answer is “none” I can’t help you.
What Bill said, cross ties, extra angles to as many sights as you can, it all enhances least squares. Doing a pt to pt traverse will usually result in numbers very close to a classical compass solution. You can spend time weighting elements and change the solutions, but that can be iffy.
It is very similar to the compass adjust, as they guys mentioned above. It is way better way to adjust side shots when you tie from multiple occupations or multiple backsights. Also, help keeps the same point number going for the same point. As was always my frustration with compass adjusts that I was doing.
Off the bat, it’s important to note that least squares “adjustment”‘is a term of art. Least squares is a statistical technique and the term “adjustment” doesn’t have a proper statistical meaning.
Think of it as “least squares estimation” since the observations aren’t adjusted to get the solution, rather the the observations (along with estimations of the errors inherent in the equipment making the observations) are used to estimate the coordinate values. Any “adjusted” measurements are then “backed in” from the estimated coordinate values.
More observations + accurate equipment error estimates = better least squares solutions.
The value of least squares, especially for smallish conventional control networks, is less in the solution qua solution and more in the fact that you get a solution along with rigorous, meaningful, statistical statements about the quality of the solution
Run a loop with 6-8 legs. Setup in the middle and resect from two non-consecutive points.
(1) Process the loop without resection with Compass.
(2) Process the loop without resection with Least Squares.
(3) Process the loop with the resection with Least Squares.
(4) Check the resection with solution 1.
Overlay the 3 solutions. 1 & 2 should be very similar. 3 should shine. 4 should have a bit of error in the resection.
Run a loop with 6-8 legs. Setup in the middle and resect from two non-consecutive points.
(1) Process the loop without resection with Compass.
(2) Process the loop without resection with Least Squares.
(3) Process the loop with the resection with Least Squares.
(4) Check the resection with solution 1.
Overlay the 3 solutions. 1 & 2 should be very similar. 3 should shine. 4 should have a bit of error in the resection.
In response to Norman Oklahoma, right now I process my raw data through TBC and import into C3D. My understanding is that Star*Net may be a better solution for least squares, but I’m not familiar with it.
This is very helpful. Thank you.
…My understanding is that Star*Net may be a better solution for least squares, but I’m not familiar with it.
StarNet is what I use. I’m a big fan and supporter.
If the least squares program in C3D is the same as it was under Land Development Desktop (and I expect that it is) you will find it compares to Star*Net as a two-bladed penknife compares to a Leatherman.
The advantage of StarNet is that it is easy to mix and match different brands of data.
If you have all Trimble then TBC is fine.
Be sure to set your centering errors and angle and distance standard errors to realistic values.
Least squares is a way to use all of your measurements to achieve the best solution. Also you can use your equipment to their best advantage. You can close your survey with a single static vector instead of a 15 turn traverse, for example. Think in terms of networks instead of closed loops.
My experience with least squares in Civil3d is that it crashes. For me the traverse adjustment also no longer works. I’ve been doing it on my HP48 lately.
A +1 for Star*Net. Easy to use once you get the hang of its formatting, and quite versatile IMO. Least squares is pretty much all I use nowadays as redundant measurements, cross-ties, etc. can all be utilized to improve the final results.
The only superior evidence is that which you haven’t yet found.We use Carlson SurvNET, great program, great tech support. I’ve not used Star*Net so I cant comment but SurvNET will provide a full ALTA accuracy report as well as a traditional loop closure report and then draw the relative error, traverse and side shots into Carlson Survey which is very nice for visuals.
For field procedures, it helps in processing if you keep your data running in one raw file until you process or you’ll have to append the files and clean out the clutter of repeated data.
Log in to reply.