Follow us on Twitter

GPS Errors


We have been seeing a number of queries regarding the accuracy of GPS coordinates plotted in MapPoint. Usually these questions come up when the maps are zoomed to maximum level, and the GPS is correctly set to use the WGS84 geoid. There is a tendency to over-estimate the accuracy of consumer GPS systems, and so the accuracy of MapPoint’s maps are implicitly blamed. But is this really the case? This page describes some simple tests with a consumer GPS system.

The test was very simple and involved walking two different tracks whilst carrying the GPS unit. The GPS unit was a Garmin eTrex Legend, set to record coordinates every 5 seconds. Data was acquired on the evening of Monday 10th January 2005. During data acquisition, the GPS usually reported an error of between 14ft (4.2m) and 24ft (7.3m) although it occasionally peaked at over 30ft (9.1m). Data was downloaded from the GPS using GPSBabel which produced a CSV (comma separated value) file that could be easily imported into MapPoint. The results are shown below:

Errors recorded by a GPS device

The first track (in black) was acquired by walking on the ‘inside’ sidewalk of a block. Any deviations (e.g. to avoid a poorly parked vehicle) were towards the center of the road. The second track (in red) was acquired by walking up and down one road, on the opposite side to the matching segment of the black track.

As you can see, there are quite a few deviations between the GPS track and MapPoint’s roads. Well, how accurate is MapPoint’s representation of the roads? It is actually pretty good. In reality, the roads have smooth curves rather than angular segments, but we are at the limits of the MapPoint’s spatial sampling. The roads that should be straight north-south or east-west are drawn correctly. Belew St is shown as a one-way street, when in fact it is not. There might be a lateral shift error or stretching error, but if this is significant we would expect to see this in the plot. For example, the black track could be made to fit the roads better if the roads were shifted a few metres to the west or north. Note that this would not be perfect. Edith St might fit better, but Belew St would not.

So the significant deviations from the road are real. Four are particularly noticeable:

  1. NW corner has a large deviation peaking at about 20m. I cut across this corner, but the result is grossly exaggerated: I never walked on the road let alone across it!
  2. West side of the black track has two major deviations, first to the west, then to the east. The road is actually a smooth curve. Even MapPoint’s angular representation is better than that produced by the GPS.
  3. The red track overshoots Belew St (NE corner). In reality I stood on the corner for about 30 seconds and did not cross Belew St.
  4. The red track has a deviation just above halfway up. This was the location of a poorly parked pickup truck that caused a deviation in my route of about 2-3 metres to the west over a length of about 10 metres. Interestingly one track exaggerates my deviation, and the other completely ignores it! The exaggerated deviation is about 20m in amplitude and over a length of 60-70m.

These are some pretty big errors! What happened? There are two main problems causing these errors. The first is the fact that the GPS receiver will occasionally lose contact with enough satellites. This might be due to obstructions such as buildings or trees. Most consumer units compensate for this by extrapolating the location according to the last recorded speed and bearing. This has the effect of exaggerating small changes in track, e.g. my deviation around the pickup truck, or the NW corner.

The second problem is the error inherent in the system as a whole. Although the GPS device never reported an error over 10m, errors of 20m were recorded. How is this possible? Well, the reported error is not a maximum as is often assumed, but a standard deviation. In other words, there is approximately a 68% chance of the actual error being less than the reported error; and a 95% chance of the actual error being less than double the reported error. I.e. When 10m was being reported, there was a 95% chance that the reported location was within 20m of the true location. In fact, consumer GPS systems are only guaranteed to be accurate to about 100m! (see references). The ‘SPS’ (Standard Positioning Service) used by consumer units is only quoted to be accurate to 100m horizontally and 156m vertically. The ‘PPS’ (Precise Positioning Service) performs better (22m horizontal, 27.7m vertical) but is restricted to military users and certain other US Government Departments. These errors are probably worst-case scenarios – e.g. at high latitudes.

Surveyors improve the accuracy of the GPS system by using differential GPS. This system uses a second GPS located at a known base station. By recording the two signals, it is possible to measure your position relative to the base station with an accuracy of a few millimeters. Integrating over time can also improve the accuracy of a GPS system, but this is impossible for a moving vehicle.

Further information on GPS Errors

Leave a Reply




You can use these HTML tags

<a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code class="" title="" data-url=""> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong> <pre class="" title="" data-url=""> <span class="" title="" data-url="">