Triumph-LS Cluster Average and Relative Accuracy

Steve Douty

Well-Known Member
I like that for this process everything was done in the LS. No desktop software was needed except to develop the report for my client.
I agree that it is great that everything can be processed by the LS, and that we don't have to buy additional software.

However, I will continue to beat the same drum:
I often find that I would like to re-evaluate data in the office while the crew is in the field.
Please find a way to allow office evaluation and reprocessing without taking a unit out of production.

As I understand it the only way to process data in the office is to buy an LS for use in the office as a data processor. This is expensive. I would be willing to pay for software or online access to emulation.
 

John Thompson

Well-Known Member
Thanks for this post, Shawn. This system presents a wealth of information to the user, but understanding what it all means takes some study. I appreciate your helpful explanations. Here are a few questions I have so far:

1. Your third post mentions total duration and combined epoch count. I assume the total duration bug is being worked on (always displays 0s for the average), but where is the combined epoch count shown?

2. The box in the lower right of the base rover statistics screen shows CC: 17.5+1233.4. What does that mean?

3. In post #6 you show that Point 14 northing for the average is higher than either of the two points that created it. Did you get an explanation for why?

4. Can you give working definitions for HRMS, VRMS, 3DRMS, sigma a, sigma b, and sigma h? How do these relate to the shape of the scatter plot (flyspecks) on the collect action screen?

5. In the ALTA report, what do each of the of the columns mean?

6. In computing relative accuracy, how is the error ellipse of a connection calculated from the error ellipses of the end points? Are the ellipses added radially?

I remember when I was making my way up the steepest part of the J-Field learning curve (I'm still learning), one of the hardest things was remembering what all the numbers mean. There are several screens that have numbers with no label and no units, and it can be intimidating until you get used to using it. Threads like this (and the help button) are very helpful. Thanks.
 

Shawn Billings

Shawn Billings
5PLS
1. The combined epoch count can be seen in two places in the Points Screen: First, the Processed Point Information screen (the screen that allows you to select RTK or PPK) or under Additional Information under Edit Point.

2. That is the Confidence and Consistency.

3. I believe it is because the average position was not originally intended to be a true average. I believe it was putting the result in the most probable location, which would be the center of the region of overlap of the two error ellipses. Imagine two skinny ellipses, one oriented vertically, one horizontally, overlapping each other in such a way that the center is outside of the overlapping area. I believe we were determining the center of the overlapping area. This has been changed and is now a weighted mean.

4. The HRMS and VRMS are statistical estimates (not guarantees) of the likelihood that a coordinate is within a particular tolerance. We report a confidence level of 95%, which is also referred to as 2-sigma. The software estimates, based on the observed standard deviation of the observation and modeled accuracies of the receiver by the manufacturer, that a point has a 95% chance of being within whatever the HRMS and VRMS values are. The 3DRMS is a Pythagorean result of the two. Horizontally the error estimate is not a circle as HRMS would imply. It is an ellipse. The error ellipse has a major axis (a) and a minor axis (b). It also has a rotation (theta).

5. In the report, we see the vector length, the horizontal error estimate of the vector, the error estimate for the vertical difference of the vector, the 3D (Pythagorean of Hz and Vt), the length of the major semi-axis (from center to edge), the length of the minor semi-axis (center to edge). I'm not sure about the ratio. Perhaps @Vladimir Prasolov can answer that. The final column is the direction of the major axis of the ellipse.

6. Probably the easiest way that I can visualize an error ellipse is looking at a compass and tape measurement. The compass has an error and the tape has an error. Let's measure a line due North at 100'. How close are we at the end of that line to being due North, 100'? At a 100 feet, the tape may have a +/-0.10 foot we'll say. The compass has a potential error of +/-0.5° which at 100 feet is +/-0.9'. So we can estimate that our point that we've measured is somewhere within an ellipse that has a semi-major axis of 0.9' that is oriented E-W (90° perpendicular to the direction of the measurement) and has a semi-minor axis of 0.1'. If we made the measurement at an azimuth of 45°, then the semi-major axis azimuth would be 135° This assumes that the compass observation was using sound procedures (no magnetic interference) and that the tape measurement was made with a horizontal measurement and the tape wasn't bent. If the procedures aren't good, then our estimate is out the window. The same is true for GPS. If the RTK solution was a bad fix, the statistical estimate falls apart, or if the observation was under canopy, then the statistical estimates are less reliable, because part of the error estimate is based on the spread of the collected epochs and part is based on modeled accuracy of the receiver. If the model isn't based on the same conditions as the observation, then there is a loss of reliability. This is why we have comparisons in our adjustments for a priori and a posteriori error estimates. This compares the results of "How much error do I think I might have?" and "How much error does the adjustment suggest that I have?"

I'm still learning too. If anyone finds any explanation I've made to be incorrect or woefully incomplete (I'm trying to keep it somewhat simple) feel free to correct me.
 

Sean Joyce

Well-Known Member
Just finishing this survey. This is typical of many of the surveys I have done with the L.S.
I performed all the work solo with leaves on.
Attaching the Relative Accuracy Report of the corners I located and held, and from the as-staked coordinates of the corners I set.
Also some files to show the terrain and vegetation.
I am damn impressed yet again. There were some very tough spots to survey.
 

Attachments

  • Preliminary.pdf
    255 KB · Views: 384
  • AERIAL.JPG
    AERIAL.JPG
    1.8 MB · Views: 399
  • TERRAIN.JPG
    TERRAIN.JPG
    139.2 KB · Views: 395
  • Relative Accuracy11052017.pdf
    338.2 KB · Views: 369

John Thompson

Well-Known Member
Thanks, Shawn. It's impressive to get a detailed response like that, and even on a Saturday!

I'll number these according to my orignal questions and your answers.

1. OK, I thought you meant the combined epoch count was displayed on the base rover statistics screen. Might not be bad to put it there alongside duration.

4. Assuming the accuracy of the receiver is properly modelled, take the the case where the error ellipse is close to round and sigma a = sigma b. If HRMS and VRMS are reported at 2-sigma, then shouldn't HRMS be approximately twice sigma a and VRMS approximately twice sigma h? Looks like HRMS and VRMS run about half, rather than twice their respective sigma. Am I thinking this backwards?

6. You've given an great explanation of the error ellipse of a measured point. My question is how to compute the error ellipse of the connection between points, each of which has an error ellipse. So continuing your analogy, let's take the point 100 feet north of the base with the 0.1'x0.9' error ellipse oriented E-W and the second point with the azimuth of 45° from the base having the same error ellipse but theta of 135°. When we run the Relative Accuracy report of the connection between those two points, what is the size and orientation of the error ellipse of the connection?
 

Shawn Billings

Shawn Billings
5PLS
1. It's a bug. It should be there.

4. You are correct. Apparently HRMS and VRMS are at 1 sigma (68% confidence level). I apologize for the error. The sigma a, b and h are at 2 sigma (95% confidence level). The scalar from 68% to 95% for a 1D statistic is right at 2x.

6. The size of the error ellipse will be bigger and the orientation will change, probably close to the 135° direction. I believe the size would be the sum of the squares of the two axis in this case (since they are orthogonal to each other). So you'd have the NS axis at SQR(0.9^2 + 0.1^2) which would be 0.91' and for the EW axis you'd have SQR(0.1^2 + 0.9^2) which would also be 0.91'. So you'd end up with a circle for the error ellipse at the end of the second measurement. I'm afraid my understanding of least squares network adjustment is inadequate to go much further here.
 

Shawn Billings

Shawn Billings
5PLS
Every two or three years, I find a new subject to learn about that I don't really understand very well. To name a few, I was plagued with how to perform a three point resection several years ago. It took a while, but I finally figured out how it works and could work one out with a calculator. Then projections were on my list. I still need to learn more, but I now have a pretty good handle on the fundamentals. When I started consulting for Javad, I needed a better grasp of localizations which led me to learning about Helmert transformations. I now understand them pretty well. Least squares adjustments have been on my radar for years. I think it's about time to get in the deep end of the pool with that now. I sure do love this profession, which offers us so much depth to study and never get bored. I appreciate the opportunity I've had to work with Javad which has called for me to dig much deeper.
 

Cheyne McLean

New Member
Just finishing this survey. This is typical of many of the surveys I have done with the L.S.
I performed all the work solo with leaves on.
Attaching the Relative Accuracy Report of the corners I located and held, and from the as-staked coordinates of the corners I set.
Also some files to show the terrain and vegetation.
I am damn impressed yet again. There were some very tough spots to survey.

Sean, what are you using to generate the ALTA report?
 

Cheyne McLean

New Member
Cheyne;
On the home screen, Go to Cogo-Tools-Relative Accuracy, check the distance threshold to make sure it matches the RA you are trying to achieve and then enter the points.
you will be given an option to generate a report in the results display.
Thanks Sean, love this tool!
 
Shawn (or anybody):

Am I correct that all of the error statistics and the ALTA reports are for the LS only, without any estimate of plumbing and centering errors for the equipment (method) of centering it (prism pole, tripod with tribrach, etc.) on the ground target.? If so, it would be helpful if there were a place to enter those factors. Carlson's Survnet software allows you to do that and I believe that it is a significant (and often ignored) factor in qualifying (quantifying?) the results of our work.
 

Greg Flowe

Active Member
Shawn, Is clustering only done through cogo manually or is there a setting that we can turn on with parameters that will automatically average when we collect a group of points in our boundary profile? If not this would be a nice feature.
 
Top