Yesterday we were working on a small job about 12 miles from the office (as the crow flies). I had a Triumph2 broadcasting corrections over TCP from our office internet WiFi. At the site I observed four points by RTK for five minutes each. I then setup a total station on one of those points and turned in the other three. All were within about 300 feet of each other. I then did a Helmert transformation of the total station coordinates to the RTK coordinates, allowing the software to determine the best rotation and translation to transform the total station system to the RTK system. I forced the scale to remain at 1. Points L001, L002, L003 and L004 were RTK derived. I was setup on T003 (L003) and turned T012, T002 and T004. The differences between the two sets of coordinates are as follows:
Point# Northing Easting Elevation
Delta-Y Delta-X Delta-Z
----------------------------------------------------------------------------
L001 872.791 405.084 419.534
T012 872.767 405.061 419.575
-0.024 -0.023 0.040
L002 904.213 702.294 410.933
T002 904.227 702.298 410.933
0.014 0.004 0.000
L003 899.143 482.485 415.170
T003 899.147 482.487 415.266
0.004 0.002 0.095
L004 170.840 530.628 424.130
T004 170.846 530.644 424.023
0.006 0.017 -0.106
I was very pleased and surprised to see this sort of accuracy from a single baseline solution at 12 miles. Five minutes is a fairly long observation time, but in the grand scheme of things four points at five minutes each is 20 minutes of observation time.
Shawn,
Please pack up that Javad LS of yours, and send it to my mailing address. I'll get by with my old base, or something.
Thank you.
Nate
> At the site I observed four points by RTK for five minutes each.
Just out of curiosity:
(a) What did the post-processed solutions look like?
and
(b) What were the RTK controller's estimates of the uncertainties of the positions it generated?
(a) No post-processed results. I do have raw data for each of the four points, but I doubt that I will attempt to process it.
(b) 95% Semi Major; Semi Minor; Height
L001: 0.035;0.031;0.081
L002: 0.056;0.044;0.118
L003: 0.058;0.046;0.095
L004: 0.050;0.036;0.060
> (b) 95% Semi Major; Semi Minor; Height
> L001: 0.035;0.031;0.081
> L002: 0.056;0.044;0.118
> L003: 0.058;0.046;0.095
> L004: 0.050;0.036;0.060
That's interesting that the vertical uncertainty estimates generated by the RTK controller were unrealistic in the direction of optimism (based upon your total station measurements) and the horizontal uncertainty estimates appear to be unrealistic in the direction of pessimism (although some systematic scale error in the RTK vectors wouldn't be shown by the total station checks and may yet be present).
The horizontal is obviously a best fit transformation. This hides much of the horizontal error. The vertical is held fixed at one point making those differences all appear as error. Very predictable...
>This hides much of the horizontal error.
I'll concede that this test doesn't address the absolute position of the points relative to the base, but the relative position of the points is most certainly reflected in this test. The transformation doesn't hide anything. The total station geometry is still in tact and the RTK geometry is in tact (nothing has been distorted - I even forced the scale factor to remain at 1). Helmert simply provides the best translation and rotation to mate the two together.
[sarcasm]I suppose I could have left the total station coordinates rotated to a compass and translated to 5000/5000, but that wouldn't really have much meaning.
[/sarcasm]
This result may have been predictable to you, but independent points surveyed by RTK from a base 12 miles away indicating this level of precision was not really what I was expecting.
The vertical is based on holding a single point, I could have averaged and come up with a best fit translation (not a best fit plane) and these values would probably improve. +/-0.10 foot vertical in 12 miles is pretty impressive as far as I'm concerned though.
Regarding absolute accuracy, I wrote a paper for Javad that details a retracement of a control network in my home town. I used the same methodology, except that the occupation times were only about 2 minutes and the longest baseline was about 7 miles.
Here's a link:
> The horizontal is obviously a best fit transformation. This hides much of the horizontal error. The vertical is held fixed at one point making those differences all appear as error. Very predictable...
Well, when I read that Shawn had used a best fit transformation, I assumed that applied to the verticals as well. With a standard error of about 0.10 ft. or more in vertical, I'd ordinarily expect a horizontal standard error on the order of about 0.03 ft. or more (i.e. about 0.06 ft. 95%-confidence).
I intentionally did not include verticals. The program I was using in the field would have forced a tilted plane to fit the verticals, which I did not want, so I only constrained to one for the vertical translation and allowed the other three to float. Averaging them all would have only changed the translation by about 0.01 usft.
> I intentionally did not include verticals. The program I was using in the field would have forced a tilted plane to fit the verticals, which I did not want, so I only constrained to one for the vertical translation and allowed the other three to float. Averaging them all would have only changed the translation by about 0.01 usft.
The more familiar approach would have been to have used the uncertainties in the four different positions to weight them in the transformation. That way, you relax the positions that have higher uncertainties and conform better to those with lower uncertainties. Naturally, if all of the uncertainties are the same, all of their weights are equal.
I suspect that Shawn used the application he had available in the field to make the transformation. I avoid doing calcs in the field, so I probably would have rolled the total station data along with the RTK gfile into a single Star*Net adjustment, holding the base position fixed. Star*Net would handle the transformation for me.
> I suspect that Shawn used the application he had available in the field to make the transformation. I avoid doing calcs in the field, so I probably would have rolled the total station data along with the RTK gfile into a single Star*Net adjustment, holding the base position fixed. Star*Net would handle the transformation for me.
Yes, Star*Net would be a very effective way to solve the problem if you have the RTK vectors with their covariances instead of just coordinates.
So many ways a test like this could be modified and extended. How about RTK points collected at much shorter occupations? Duplicated points and various intervals? The same collection technique at a different time? In a different place? Levelling of the points for more positive comparison of the GPS derived elevations. Remeasurement of the terrestrial observations. Etc. Etc. I keep checking my maps for this "Someday Isle" people hear me talking about. I'll get there yet.
I do think that Shawn has shown that RTK points can be very repeatable, in the horizontal at least, without using extraordinary procedures.
I didn't mean 'hide' in any nefarious sense. From a strictly math standpoint the procedure described should produce better than expected horizontal and worse than expected vertical. Hence the 'predictable' comment regarding the observation by Kent.
> I do think that Shawn has shown that RTK points can be very repeatable, in the horizontal at least, without using extraordinary procedures.
Actually, I got the idea that he just surveyed the series of RTK points in sequence, one after the other. I'm not sure how great a way to test repeatibility that is, really. I do agree that if all RTK users took at least 5-minute occupations, the world would be a better place. :>
Sorry, Not Accurate
All 4 points are in precise agreement relative to each other. But accuracy implies precise with the World,, but that cannot be proven.
Paul in PA
Sorry, Not Accurate
I would submit that, because this small experiment compared two unique measuring methodologies (RTK and terrestrial total station), this is evidence of accuracy, not precision. Perhaps some scalar error would be hidden that cannot be identified in the case of a closed system consisting of RTK only (just as an EDM with improper scalar settings can provide an excellent closure). However, because this compared two different types of measurement, the RTK proved to be quite accurate in a local context.
Based on this test, we have no way of confirming how this network relates to the base 12 miles away, so I would agree that total station measurements would only suggest there is precision in the 12 mile vectors to the site. The points were observed by RTK over a period of about 50 minutes, so they are somewhat correlated. There could have been some atmospheric condition that would cause the 12 mile vectors to be scaled by some unknown amount that affected each vector identically. So the local network could be off by some distance from the base, while all of the local points are relatively accurate with one another. You'll notice that I didn't provide any comparison of coordinates from the base to the rover points, only a comparison of local coordinates generated by RTK from a base 12 miles away with points determined by total station observation.
For a comparison of total Network accuracy, see the link I mentioned above.
Triumph-LS Precision and Localization
I titled the paper "Precision and Localization" because of the comparison of repeat observations, but I would suggest that because I was observing points from an independent network established under substantially different conditions, equipment, and processing software, that this reveals something of the accuracy as well.
Sorry, Not Accurate
Not sure if this paper has been posted here before, but some here may enjoy reading it. There is a research project going on at Oregon State University under an NGS grant that is designed to test the precision and accuracy of various methods and observation times as part of the overhaul of the NOS NGS-58 and 59 Guidelines.
Not to Hi jack the thread, but what software are you using to broadcast correction?
