Accuracy of a GPS

Bob Mehew

Well-known member
Following up some field work using another person's 10 figure OS grid ref on my GPS, I was slightly surprised to find I was over 30m out horizontally from the target.  (The target was a fossil some 20cm long by 5cm wide but fortunately someone else turned up who knew its location.)  It occurred to me that one GPS may well have different capabilities from another (even if they are the same model) such that the notional accuracy is affected. 

So does any one have any data on GPS readings from different devices for the same location?  Or indeed reading some old threads, data from the same device for the same location but at different times?
 

graham

New member
Bob

There are a number of good articles on GPS in wikipedia which discuss these parameters.

A modern hand-held unit, GPS only (not GLONASS) but with EGNOS enabled will expect to get a reading to an accuracy of around 7 m. It may be better than that (down to 2m) but don't rely on it. So, if the person who originally took an NGR for your fossil was on an older hand-held unit without EGNOS, then 30 m would be about par, especially if there weren't many satellites around. My first Garmin only actually read to 6 figures (100 m).

We took a whizzy dGPS for a walk around Cheddar the other day. On a good day, this thing will be accurate to centimetres, but that was not a good day & we couldn't get enough satellites for much in the way of a decent fix.

Currently I have three different fixes for the entrance of Great Oone's Hole. One from a surface traverse down by Willie Stanton but only to an 8 figure NGR in the 'fifties & two using a modern hand held unit. Interestingly the eastings are spot on the same but the Northings vary by quite a lot. We couldn't get a fix at all with the dGPS, no satellites that day.
 

Ship-badger

Member
Hi Bob. I don't have recorded data; but I use a GPS frequently for work, and I can tell you that using 4 different GPS, but all the same model, to record the same point will give 4 different readings. Most modern GPS are more accurate than earlier models, so using a modern model to locate a waypoint recorded on an earlier model might give rise to an error of that magnitude. When looking for Geocaches with my modern top spec GPS I often encounter errors of 20m, which I put down to the location having been originally recorded on an early model machine.
 

jarvist

New member
This sounds like there might be a 'Datum' / coordinate system error here - perhaps one of the GPS devices is still using it's internal WGS84 datum, and one is using OSGB36? That can easily give a ~100m error depending on location.

Internally, all GPS fixes are lat/long on the WGS84 datum. I think that's the most transferable number between GPS systems. Also, there's not a single mapping between coordinate systems, different devices may make different approximate conversions.

The OS publish a great + understandable guide to the coordinate systems in the UK:
http://www.ordnancesurvey.co.uk/oswebsite/gps/information/coordinatesystemsinfo/guidecontents/index.html

Also, the 'reported accuracy' on Garmin GPS devices is rather optimistic!
 

Bob Mehew

Well-known member
As I don't have access to the originator of my data, I can't say whether any of the suggestions may have some truth.  But what I fear is that different modern models will produce slightly different data such that the oft cited 3m accuracy value is mythical as on board features create an even larger error when using different devices.  My concern is that even though someone may have provided what they think is an accurate location, using a different GPS could well induce an "apparent" error.

I had hoped that some expeditions might have stumbled across this problem.  Perhaps if no one can provide such data it might be worth organising a GPS "fixation" at Hidden Earth with people bringing their model along and providing a 10 figure reference for a fixed location.  Then we can work out what happens when you take a given reference and go looking for that entrance.
 

RobinGriffiths

Well-known member
Geocaching may be a case in point here. Generally speaking, using the gps in a HTC Desire (old now),  you can expect to be within 4m maybe of the location posted by the cache's owner. With 4m errors attributable to that accumulated via both owner and seeker, then actual precision may be around the 2m mark.
 

jarvist

New member
Bob Mehew said:
I had hoped that some expeditions might have stumbled across this problem. 
My perspective is one of never trusting the coordinate converters in GPS units; we log all cave entrances in the device-native WGS84 Lat/Long, which is also what you get when you directly download from the device. Conversion to the local grid format (UTM) is done offline, with a desktop computer program.
Of course, this means maintaining two datasets - one of the WGS84 lat / long & one of survex/grid X,Y UTM offsets. When I go to a new caving region, I usually try and convert the locations into WGS Lat/Long ahead of time, for use on a GPS. You can also sanity check these data by plotting on Google Earth / maps (satellite view) and see if the location looks sensible.

In the UK you can use the canonical transforms offered by OS themselves:
http://www.ordnancesurvey.co.uk/oswebsite/support/os-net/coordinate-batch-converter-file-formats.html

With the various representations out there for local coordinates, and the datum bogey man, I don't really see what else one can do.
 

Bob Mehew

Well-known member
Thanks for the information and advice Jarvist and Robin; it is as I expected, there is some inconsistency due to variation within the instrument.  And geocatching is a clear indicator of the size of the error which I thought could be much larger. 
 

graham

New member
I have been reading up on this a bit, recently. EGNOS enabled receivers are now (since 2011) used in life-critical applications, such as landing aeroplanes this means that their receivers need to be accurate to within 1.5 m, horizontally and vertically.

I'm still not convinced by the altitude thing, as it depends very much on the geodetic model used - and the aircraft and the airfield will be using closely matched models for sure - but it does mean that, in the field, an EGNOS enabled handheld unit should give x & y coordinates to within about 2m consistently, providing it is picking up a decent number of satellites. This ought to be good enough for our purposes as cavers and cave surveyors.

I will agree with the comments above, though, that data should be collected in Lat-Long and converted using something better than the programs generally present in the units.

If you have a unit that picks up both GPS and GLONASS (as well as Galileo as satellites become available) and is WAAS and EGNOS enabled then you should be reasonably sure of a good number of satellites in view and good accuracy on the ground, more or less wherever you are.
 

hamsterlord

New member
jarvist said:
This sounds like there might be a 'Datum' / coordinate system error here - perhaps one of the GPS devices is still using it's internal WGS84 datum, and one is using OSGB36? That can easily give a ~100m error depending on location.

Been caught out by this a few times, whenever the battery is changed my old garmin defaults to WGS84 and has to be manually setup each time..
Need something more up to date,  probably going to get an eTrex30 soon.
 

McMole

New member
Experience shows that with a good view of the sky, and therefore a good spread of satellites, reproducibility can be better than 5 metres and with EGNOS perhaps half that. EGNOS corrections are broadcast from an equatorial geostationary satellite which isn't visible if you've got much of a hill to your south.  Like Graham I'm not convinced by altitude figures. I have read that the vertical inaccuracy is about 50% more than the horizontal, but actual results are worse than that ie about double.

One issue when creating databases is what to store and publish for coordinates. WGS84 is a given (or ETRS89 if you want to future-proof against continental drift!), but we should supply NGR as well. There are two possible ways to convert WGS84 to NGR. To match the actual grid used by the OS which has built-in distortions because it was surveyed by hand over 60 years ago, you need to use their on-line converter. The alternative is to use the mathematically precise and much simpler converter built into your handheld GPS receiver. The first will match published OS maps, the latter will lead another GPS user more closely to the right spot on the ground. In Assynt the difference is about ten metres. I think the latter is the better choice.

I'd expect all users to have their GPS receivers set to NGR in order to use OS maps, so we do need to supply NRG as well as WGS84 and it should be a NGR that leads the user to the right spot and not one some distance away. Not everyone will do as Jarvist recommends and use WGS84 all the time.

Some numerical results:-
66 measurements taken over several years on top of a pillar in front of our club hut gave a range of 10m and standard deviation of 2.1m for E/W and a range of 20m and standard deviation of 3.3m for N/S. The range for altitude was 39m with a standard deviation of 6.0m. All readings were taken after selective availability was turned off. I used a Garmin GPS12 at first and later a GPSmap 60CSx. EGNOS was not available.

At the nearby OS passive station the ranges and standard deviations for 13 measurements again over several years were 8, 8, 17m and 1.9, 2.8, 5.8m for E/W, N/S and altitude.
 

maxf

New member
Correct about the big vertical vs horizontal accuracy discrepancy

Work involves me processing huge amounts of gps data and this is always the trend




Sent from my BlackBerry 9900 using Tapatalk
 

kdxn

New member
Try to find out the vintage of GPS fix before you use it.
Really old GPS fixes were subject to Selective Availability - a deliberate degradation of the GPS system - could be as bad as 100m.
 

wookey

Active member
We (CUCC) have loads of info on this issue, with many fixes of the same points (in Austria) over many years using different devices and datums, since 1996, so the first 5 years of data are subject to SA, and obviously chipsets have improved over that time, but users haven't much. We also have quite a lot of points surveyed with very accurate GPS/theodolite, and surface surveys to other GPSed locations. It would be possible to generate a lot of useful info on real-world accuracies and how they have changed over 18 years from this dataset.

To summarise the results would involve a lot of peering at runes and writing a proper article. But the dataset and original notes and writeups is available to you too: http://expo.survex.com/ with info on checking out the dataset under 'expo handbook'. http://expo.survex.com/handbook/update.htm#quickstart

Cliffs and steep hilsides matter a lot in GPS accuracy. From memory: old pre-SA data can be up to 1km off when you first get a fix on a hilside. If logged for an hour or two it should be inside 100m. We generally find post SA data from two GPS fixes to be within about 30m of each other.
We also teach everybody to use WGS84 datum and post-convert, mostly because few GPS units have the Old austrian BMN grid and entering the conversions is different (and error-prone) on every device. Even proper conversions which give good X/Y locations give altitudes approx 30m off (consistent offset, not error range). If you care there is long explanation here: http://expo.survex.com/handbook/survey/coord.htm

I haven't done a more recent analysis of how good the repeatability of more recent fixes is. It would be an interesting thing to do.
 

Alkapton

Member
I use very cheap phone (huawei g300) if I look at google earth then I am typically within 2m of what the phone says I am.  My question is will the reported ngr be as accurate as my position shown on map. 

I don't see why it should not be but this thread has me wondering.
 

Bob Mehew

Well-known member
Alkapton said:
I don't see why it should not be but this thread has me wondering.

That was partially the reason for asking the question.  I have no clear answer as it is clear that sometimes it can be significantly wrong unless you take extreme precautions not viable with a hand held device.  But often it will not be that far out.  Perhaps I was just unlucky with the example which triggered the question. 

One thing which has come out clearly is that along with taking a Lat Lon reading (not Grid Ref or what ever the country you are in may use) it is also very useful to take a photo of the GPS and the target (thanks Wookey for that insight). 
 
Top