-
Expected precision using RTK
I’m using a Leica System 500 to do some RTK work, with the base station no more than three miles away. I’m finding that my shots can vary wildly from day to day. Okay, wildly in this case means 0.20′ vertically. For this particular point, the shots were taken 2.5 months apart, at the same time of day. Approximately 60 observations each. I know that GPS isn’t always great with elevations, but what is throwing me off is that each point lists a Vertical Accuracy of 0.04′ yet these two particular shots vary by 0.20′. Is this expected behavior?
-
20 Replies
-
EdWidy, post: 352420, member: 10963 wrote: Is this expected behavior?
Yep.
-
Thanks. I’ve always taken GPS data with a grain of salt – and I shall continue doing so.
-
If you have to do RTK at that distance then 3minutes obs (180 records) would be better and return to the spot some time later – ie. if you have a couple of dozen points to check do them all once and then do them again (in the same order – if you reverse order then the furthest one gets two readings within a few minutes.
If there are a lot of points it would be better to establish a local base (say 12 minutes obs on static) from your main base, move the base station up and then do local RTK in that area. If you then want a further check do a 180 obs. RTK on your original base point. There is always a lot of uncertainty once you start taking the RTK beyond a mile or so – 5 times it’ll be fine, the sixth time will be garbage. The trouble is you never know when the sixth time is coming up GPS doesn’t count the same way we do!
Never believe the manufacturers statistics – the vertical accuracy is probably actually the spread of the values it read at that point. If it knew what the REAL error was then it could correct your reading to be true (just like Networked RTK doesn’t). -
I usually try and avoid discussions concerning technical matters such as GPS. It is all truly witchcraft to me, but we use it. But as a techno-peasant I will relate experiences I have consistently found when using RTK for determining vertical values as in a topo.
First off, what I believe we see when two points, observed at different times over a period time, is not only positional differences in the elevation determined by the “witchcraft”; we also see ALL of the combined errors (differences) in the setups when we key-in the antenna heights for both the base and the rover. While a few hundredths here and there shouldn’t add up to all that slop, it does as times.
And then there is the elevation value determined by your software as it juggles through thousands of vectors as we walk along with this stuff trying to keep up with our velocity. I personally don’t usually see values like 0.2′ very often, most are in the neighborhood of half of that (0.1′). But when you think about it, determining precise locations from SVs whirling around thousands of miles away, that ain’t too bad.
In consolation, what I look at is the actual elevation differences between two points observed and compared over a period of time. While some elevations seem to be “off”, the differences in the determined elevations don’t vary that much; usually just a few hundredths. Like say two curb shots 200′ were shot two months ago showing elevations as 1241.15′ and 1242.15′. Returning a few months later and trotting about with the RTK may very well show there elevation as 1241.25′ and 1242.28′. While the RTK elevation indicates 0.10′ and 0.13′ differences with the data collected a few months ago, the two points reflect differential elevations that are within 0.03′ of the previously recorded data.
Point is, elevations with RTK can be elusive. Throwing a base up, pushing go and expecting vertical values to be anything other than “close” might be a little disappointing. We always use tight vertical control and still utilized spirit levels and TS when the vertical elevation is an important factor in the survey.
-
I used to hire a 500 and what you describe would be typical of such.
But take Paden’s notes on board as it’s the relativity, that is from experience with 500, ‘up to it’.The 500 is long in the tooth, but I did get excellent results.
Wasn’t best around trees, but with care one could still grab shots that at first were not fixing.Those units, I found were excellent, and I’ve stretched them to 12 miles but then not interested in 3D only 2D.
As Chris says, set it to take 2 or 3 minutes of observations and repeat.
That’s how I collect data now. Pays off.
( I don’t have a Javad!) -
Are you using your own base station?
-
Is your base and rover both 500s. I had problems when I tried to use the older model as a base. When I tried to stake out a location between two points is when I found out you can’t mix the 500 with the older unit.
The second problem I had, and corrected, was the pole was out of adjustment.
A third thing I would try in your case is, mask out any low satellites that your base or rover is seeing. Masking out in my language is to cut out any low satellites. The rover will see them, but won’t try to use them.
I did not like the Leica systems, and don’t have any Leica equipment now and never will again. -
The points about using your own base station and masking are good ones. It’s always better and more consistent to use compatible equipment. If you use somebody elses base station then, if you can, take a 10 second observation on a tripod or bipod a few feet away and measure the distance between the two with tape and level (3ft. tradesman level would do). The you have some idea if there is a systematic shift between the two. The antennae constants/centres are often the cause of shifts between different receivers.
On masking, I normally mask off at 15 degrees, dropped to 13 only if there isn’t any choice or the work is rough only. If you are in a wooded area you might want to set the mask even greater, to cut out “one-signal” readings which have squeezed through the foliage.
-
Some good thoughts so far. A few more tidbits to consider…
Placing the base and rover less than 10 or 20 feet apart can cause very real problems for some units. If you are haunted by Bluetooth or radio problems try maintaining separation and see if it stops.
Mixing and matching antenna models is nothing new. Some sellers advocate it as a way to save money. While some combinations work, it can come with a cost. RTK is based on the age old principal that screwing things up the same way every time gives consistent results. Multipath and other forms of interference impact every antenna different. Results can be difficult to quantify and predict.
Matched units are getting very inexpensive. If you can afford to get the same color equipment I would go that route. Same model even better… -
The “accuracy” your getting is typical of RTK. As several have already said, average several positions with differing satellite geometries to get a more “accurate” position. The “precision” reflected by the 0.04 ft. doesn’t really have anything to do with how close the positional value is to the “known” positional value of the point. It it typically the VDOP of the pseudorange position. In other words it is a component of the aposteriori covariance matrix which may or may not be scaled by anything depending on your software. It is a value that is computed completely independently of the “known” positional value of the point. The diagonal of the aposteriori pseudorange covariance matrix has four diagonal elements: X,Y,Height and T for time. These elements are squared and then summed. The square root of that sum is the GDOP of the solution. The square root of the sum of X and Y is the HDOP. The square root of the V element is the vertical precision, the 0.04 ft in your case. At least this is how it is usually done. A similar process can be performed with the covariance of the phase solution. The results are usually pretty much the same. So the vertical precision is a value that is completely a function of the satellite geometry. I always take these values only as a rough guideline because, as you noticed, they quite often don’t really give you much of an idea of how close the computed position is to anything else. In fact it is not unusual to have a fairly low vertical precision and still not be very close to the known value of the point. The whole subject of weighing of satellite measurements and setting up their covariance matrix as part of a stochastic model is very poorly understood and produces only a very rough stochastic model. About all I use it for is to tell if I have a fixed solution or not. The whole stochastic modeling method works much better with conventional angle and distance measurements.
-
EdWidy, post: 352420, member: 10963 wrote: I’m using a Leica System 500 to do some RTK work, with the base station no more than three miles away. I’m finding that my shots can vary wildly from day to day. Okay, wildly in this case means 0.20′ vertically. For this particular point, the shots were taken 2.5 months apart, at the same time of day. Approximately 60 observations each. I know that GPS isn’t always great with elevations, but what is throwing me off is that each point lists a Vertical Accuracy of 0.04′ yet these two particular shots vary by 0.20′. Is this expected behavior?
That’s too much considering your own base is being used, you are taking long timed locations, RTK will do better than that, but if you are doing continuous topo, single epoch shots, then yes .2′ would not be surprising.
Although I can’t speak to the Leica 500
-
My experience with Leica System 500 was that it was way better than that. I’m wondering if maybe at that distance you’re dropping some corrections – if you lose the radio for a couple seconds it won’t necessarily notify you but the position will drift.
-
My experience with Topcon RTK is it is almost always better than 0.20 vertical. I would say 0.10 isn’t uncommon but it usually does better than that. Are you using a geoid model in your data collector?
-
My experience with RTK is that the vertical accuracy is usually better than 0.2 ft., but 0.2 ft is not unusual and I have seen positions as bad as 0.3 ft. vertically. Several years ago I ran some 24 hour tests of RTK positions between two known network base stations both positioned to very high accuracy by months of static data. For about a minute the vertical position drifted to as bad as 0.3 ft vertically then reverted to better accuracy probably when a low satellite dropped out of the solution. One minute in 24 hours is a very small percentage of the time span but if you happen to be taking an RTK position during that interval, that is what you would get. They sometimes get that bad although they usually are better. There is no mistake involved or software glitch. RTK positions can get that bad although they usually are better. So for vertical positioning certainty with RTK it seems necessary to take several positions with different satellite geometries.
-
The “accuracy” your getting is typical of RTK. As several have already said, average several positions with differing satellite geometries to get a more “accurate” position.
I wish to have an input on this, the reason why we use RTK is to avoid the longer time duration using static method. If you are going to re-observe the point again at another date or later in the day then it kind of defeat the purpose of using RTK.
Might as well set up a static session and get a stable reading. That’s the dilemma of GPS – RTK vs Static :stakeout: -
FrancisH, post: 352743, member: 10211 wrote: I wish to have an input on this, the reason why we use RTK is to avoid the longer time duration using static method. If you are going to re-observe the point again at another date or later in the day then it kind of defeat the purpose of using RTK.
Might as well set up a static session and get a stable reading. That’s the dilemma of GPS – RTK vs Static :stakeout:If you want a valid point using static you need to observe it more than one time also.
-
To me it seems important for me to keep in mind that RTK is certainly not the answer to all my positioning needs. It reminds me of microwave cooking. It works when you want something in a hurry and the food is OK, but if you really want to do some good cooking you go to the stove and take a longer time of it. So if I have to have strong vertical accuracy, I take repeat measurements separated by time and I use the time between measurements to position something else. As has been discussed elsewhere in the board repeat RTK measurements separated by time usually give pretty much the same results as static sessions over the same time interval. The changing geometry is what does the averaging rather than the one after another one second RTK measurements with virtually no change in geometry.
Incidentally, in the test I mentioned above where I got 0.3 ft. vertical error in one minute of a 24 hr session of RTK measurements, I like to point out that one minute in 24 hrs is 1/60 x 1/24 = 0.000694 or 0.07% of a 24 hour session. So getting that much vertical error happens very infrequently; however, if it happens during the time I take the RTK measurement I still have 0.3 ft vertical error.
I’ve done a lot of airborne lidar processing and noticed a similar phenomena. Sometimes a set of GNSS positioned lidar points from a specific flight line on a flat area such as a parking lot will suddenly jump vertically a few tenths then gradually smooth out. This is usually caused by a satellite approaching the horizon then suddenly dropping out of the solution when it drops below the mask. You can smooth these things out by figuring out which specific satellite is causing the problem and eliminate it for that interval, but this can be a very tedious and time consuming process and usually not worthwhile.
-
Smoothing things out and eliminating the satellite(s) close to the horizon does work BUT it requires the operator to remember to log all the data and not just record RTK coordinates. Good practice says that you should always log everything; you can always throw it away afterwards but you can’t get back what you never had.
Most of the time you can mess on with the data for hours and only tweak the Static results by a few mm., but once in a while it proves really useful. My experience is that chopping short sections out of the graphs is a waste of time. Once the satellite record starts to get “bitty” then chop the whole section and see if it improves matters. If there isn’t any change then you might as well put it back into the solution, otherwise you end up with cutting out everything!
-
A lot of good info posted so far as would be expected here.
One thing not mentioned (I think) is that there are times during some days that the DOPs can be quite high for various reasons. I was recently involved with a static control survey. Both days, according to the mission planning software, there was significant spike in the DOPs that lasted about 45 minutes. We didn’t collect data during those times. While not likely, your questionable vertical could have been taken during a period of bad DOPs. Mission planning can still be our friend.
-
Sometimes when I am able to compare an RTK shot with a Static session of same point and their vertical values are off by several centimeters I just throwaway the RTK shot. I don’t analyze the data because it would be a waste of time. The math behind it are too complex for me. I would rather re-observe the point or compare it with my TS results.
Log in to reply.