Activity Feed › Discussion Forums › GNSS & Geodesy › RTK Error Estimates in the Field – For Kent
RTK Error Estimates in the Field – For Kent
shawn-billings replied 9 years, 4 months ago 19 Members · 94 Replies
I did not mean to cast asperstions on your asparagus
I believe from what I have seen that gschrock and shawn billingsly are diligent surveyors . I merely suggest caution when looking to the data collector for warn fuzzies. Further I suggest that when an obvious saga appears amongst the mere mortals, that the mere mortals would do well to heed the advice of the saga.
I know. it is a pain in the ass to be told that what you thought was sufficient proof is in fact NOT. I hate it when that happens, as well. I suspect that because Mr. McMillan is a boundary surveyor, that he is used to having to PROVE his opinion is correct. This activity is in reality vastly different than surveying for engineering purposes.
This is like watching an R-rated French movie without subtitles. O.K., sometimes the female lead is naked, and I like that, but mostly I have no idea what the heck is going on. I do know that Kent is always right, so I don’t know what the heck you other guys are whining about.
Get over it.
Don
Generally, who is advocating button pushing in this thread?
I get a kick out of all this. It’s sick, it’s twisted, it’s me.
My biggest problem with the whole thing is this…
I use RTK properly, as do many others. I’ve trained MANY old farts to use RTK. It took time, it took patience, but they were thankful that I was able to bare with them and teach them proper procedures, answer their questions and address their concerns. They were willing to try, and I respected them for that; old dogs can learn new tricks.
We can go into this deep analysis constantly, or accept the facts that equipment manufacturers aren’t out to make a crappy equipment and that there are MILLIONS of people worldwide (America is not the leader in surveying, truth be told) using RTK properly and getting more than acceptable results for there scope of work.
I have used and continue to use RTK for boundary work, given good conditions and using proper procedures. I am a nerd AND a Land Surveyor. I don’t bother chasing mils, since I damn well know that it’s the evidence that defines the boundary, not the distance I measure. I could go into the fallacy of accuracy and tolerance much deeper, but that’s another topic we can all argue about later. I do the best I can using the best procedures that gives clients a professional service at a fair price. Grouping all RTK users into one barrel is ignorant. There are sh*tty surveyors everywhere; that we can all agree on.
> http://www.researchgate.net/publication/259451007_Local_Accuracies
>
> Here is the white paper with appropriate formulas.Okay, I’m afraid you’re not clear enough about this subject to even be offering an opinion. The paper you linked points out that local accuracy, i.e. the connections between adjacent points in a network, can be better than the accuracy of the network with respect to datum (meaning the CORS network as a practical matter in the US). This should not be news.
In Shawn’s case, what you are looking at are the uncertainties in the local ties between three points, none of which are expressed as network uncertainties. They are all local uncertainties and are mostly uncorrelated. So there the uncertainty of the resultant inverse between the points is simply that derived from random error propagation and it will ALWAYS be larger than the uncertainty of either point.
These definitions should help you understand the difference between network accuracy and local accuracy:
> For publication purposes, the network accuracy of a control point is a value that represents the uncertainty of its coordinates with respect to the geodetic datum at the 95 percent confidence level. Since the datum is considered to be best expressed by the Continuous Operating Reference Stations (CORS), which are held fixed during the adjustment.Local and Network accuracy values at CORS sites are considered to be infinitesimal (approach zero).
>The Local Accuracy of a control point is a value that represents the uncertainly of its coordinates relative to other directly connected, adjacent control points at the 95-percent confidence level.
Source:
Generally, who is advocating button pushing in this thread?
Well said. I agree 100%
I use RTK on a regular basis, but I am very, very aware of it’s limitations. I run static for 99% of my control work. The other 1% is for rough topo.
> This is like watching an R-rated French movie without subtitles. O.K., sometimes the female lead is naked, and I like that, but mostly I have no idea what the heck is going on…
I am with you this far.
Generally, who is advocating button pushing in this thread?
> 1. I’d replace “determining” with “estimating.”
I don’t have any problem with using the more correct term “estimating”.
> 2. The only way to test GPS processor estimates is to mix in measurements from a different technology — EDM, theodolite, taping, leveling — or compare the results to work done by others considered reliable (e.g. NGS). With experience, I think you can get a reasonable idea of a scalar to be applied to the particular processor involved, and then you can apply that scalar to other work that doesn’t include any of the non-GPS checks.
Yes, I agree entirely. That is what prompted the test method I suggested, which was establishing a test array of control points that were in some way representative of typical settings and surveying them at relative accuracies around +/-2mm (s.e.) or better by a combination of static GPS and convention methods so that the relative positions of the points were known with very small uncertainties. Then, using RTK to survey the same array at separations from a base that are typical of those expected in practice, and adjusting the RTK vectors by least squares to test the processor-estimated uncertainties of the vectors (which should mean determining the scalar to be applied to the variances to make the residuals consistent with the supposed uncertainties.
VERY true.
The point I was really trying to make is that the procedure that I indicated would test the accuracy indicated in the screen shots with a direct vector. I don’t believe the results will be bad. My interest is that screen shot data is correct. Have you tested this or just taking Javad’s word for it? For the record:
1. I do not believe you have to run something through a Networked Least Squares program to prove accuracy. Although it is definitely, the best way I know.
2. I use RTK regularly and determine positional certainty by multiple observations.
3. If you have an ALTA Survey, I could not believe not state that I have met the positional accuracy with single observation rtk shots.
4. I believe the main problem with bad surveys is bad procedures and surveyors and not the technology. Frankly even though this will put me in the old fart camp, I have actually observed great improvement in results since I started using GPS in the early 90s. Ground geometry and session lengths principles still affect the results. However presently, we can get reasonable results with less concern for either than when I started.
5. Why don’t we all just admit that a 2 sigma positional certainty can not be determined by a single RTK observation?
6. An accurate survey can be perform using RTK with the proper procedures. If I can retrace your survey within a reasonable accuracy, I will not condemn your procedure.
7. Comments being taken out of context is a major frustration with this site.
That’s a very fair question. I can only say that I’ve tested the distance relative accuracy on a couple of occasions and it seems very much in agreement with the residuals I see in the known baseline I’m inversing. But I’ve only tested it a few times as it is a fairly new feature.
Could you elaborate on your position on number 5. Why can we not determine a 2 sigma positional uncertainty from a single rtk position?
>the connections between adjacent points in a network, can be better than the accuracy of the network with respect to datum (meaning the CORS network as a practical matter in the US). This should not be news.
Was not your issue that the local tie inverses showed a confidence level lower than the uncertainties to the base? Is not the base the reference datum in this situation? You could use NAD83(2011) as a refrence frame, or you could use the corner of grandmas driveway, as long as that metadata is clearly stated. We could assume the base is a RTN CORS if it makes you feel better.
Its is not inappropriate to calculate local accuracy (relative accuracy) between two uncorrelated points with ties to the same reference datum.The way you answered me I suppose you have no issue with this anymore.
> Was not your issue that the local tie inverses showed a confidence level lower than the uncertainties to the base?
No. The observation was that the uncertainty indicated on the third screen as the relative uncertainty between the two points was obviously grossly incorrect because the absolutlely smallest the relative positional uncertainty could be was the value of the semi-major axis of the 95%-confidence error ellipse of the point with the larger semi-major axis. It turned out that the number displayed was actually the standard deviation of the distance component on the incorrect theory that this could test for compliance with the ALTA/ACSM relative positional precision spec, which it absolutely cannot.
>Is not the base the reference datum in this situation?
No. The base is absolutely not a datum. “Network accuracy” in the sense in which the term is used in the US refers to the uncertainty with respect to NAD83, which as a practical matter means with respect to the CORS network. Here’s an example that should make the usage plain:
Suppose that Shawn knew the NAD83 position of his base with an uncertainty of +/-0.1m at 95% confidence. That means that the base has a Network accuracy of 0.1m.
Then, he surveys two points ten feet away from the base with an uncertainty of 0.005m at 95% confidence. That is a measure of the Local accuracy of the base with respect to the two points.
Example of Relative Positional Precision – RTK
As an example of Relative Positional Precision, here are two points, L009 and L022, surveyed via RTK from a base, with the same relative uncertainties with respect to the base as Shawn’s example provided. The only difference is that the long axes of Shawn’s error ellipses were tilted a bit and, lacking the covariance matrix for each, I’ve settled for orienting them cardinal. It shouldn’t significantly effect the results.
[pre]
Relative Error Ellipses (FeetUS)
Confidence Region = 95%Stations Semi-Major Semi-Minor Azimuth of Vertical
From To Axis Axis Major Axis
BASE L009 0.047486 0.018848 0-00 0.000000
BASE L022 0.052382 0.030842 0-00 0.000000
L009 L022 0.070702 0.036145 0-00 0.000000
[/pre]As may be seen, the Relative Positional Precision of L009 with respect to L022 is actually 0.07 ft., not the figure of 0.0298 ft. that the Javad software was displaying, and is a value greater than the semi-major axis of either of the two points.
The RPP as defined by the ALTA/ACSM specification is the size of the semi-major axis of the relative error ellipse between two points and in this example, as the above table shows, the semi-major axis has a length of 0.07 ft.
This is why I think it is fair to describe the Javad display as mistakenly optimistic at best. That is, the RTK work was presented as if it easily passed the ALTA/ACSM RPP spec when in fact it was right at the razor’s edge of failing it.
For those who want to verify the computation, here’s the Star*Net input file I used:
[pre]
PH L022 32-26-25.34225 94-55-41.01833 333.487 0.0214 0.0126 !
PH L009 32-26-23.32835 94-55-32.90936 316.335 0.0194 0.0077 !
PH BASE 32-26-23.47099 94-55-36.41746 329.336 ! ! !D BASE-L022 437.317 *
D BASE-L009 301.002 *.REL BASE-L022, BASE-L009, L022-L009
[/pre]Texas Coordinate System of 1983, North Central Zone 4202 was the projection.
>the absolutlely smallest the relative positional uncertainty could be was the value of the semi-major axis of the 95%-confidence error ellipse of the point with the larger semi-major axis
Once again, what emperical evidence do you have that this is the proper statistical solution. There is no difference between relative position accuracy and local accuracy.
> The base is absolutely not a datum.
You glazed over the assumption that the base could be a RTN CORS which has zero network error per FGDC 1-A. What then?
Example of Relative Positional Precision – RTK
Good work as usual Kent :good:
Example of Relative Positional Precision – RTK
> Ditto. Excellent example, and with actual RTK data (finally…whew!)
Actually, that isn’t RTK data. Shawn never posted the actual RTK vectors and their processor-estimated covariances. My computation is based upon the shapes and sizes of the 95%-confidence error ellipses that were shown on the screenshot.
So, what that exercise consisted of was inputting the positions that Shawn’s screenshot displayed and assigning standard errors to the lat and long that were equal to 1/2.47 x semi-major and semi-minor axes of the 95%-confidence error ellipses for the two RTK points as displayed on the screenshots.
From examining the output I posted above, you can see that the 95%-confidence error ellipses derived for both points from that input are the same as those shown on the screenshots, but with the exception that the major axes aren’t tilted by about six degrees. That difference should have a negligible effect for the purposes of the exercise.
> You glazed over the assumption that the base could be a RTN CORS which has zero network error per FGDC 1-A. What then?
The fact that you’re wanting to dispute something that is obviously true just tells me that you are completely unaware of how random errors propagate. See my computation toward the bottom of this thread.
I am referring to the Network accuracy of the position of the point on the ground obtained with a single observation. I have the same opinion about a single static observation. My experience is there are too many variables that can affect the accuracy to truly determine the estimated accuracy from a single observation. The equipment does a much better job than even ten years ago. I still find differences in positions with multiple observations. Our equipment indicates that we are getting good data, but our results will be different than indicated based on multiple observations. Of course, my experience is in an area with less than optimum conditions due to tree coverage. We have to be very selective to get good results. I am sure my experience would be some what different in wide open spaces.
Log in to reply.