Got some good comments from several of the recent posts on GPS accuracy (one, two), comparing GPS receivers from Garmin and Holux, and thought I’d feature them here with additional comments from me.
Peter Guth (author of the freeware GIS MicroDEM) notes that measurements by his students of a NOAA tidal marker showed that its measured position was off from the specified position by a distance comparable to what I saw with the Holux GPS receiver, suggesting that perhaps the NGS benchmark position I had was off by a comparable amount. His marker has been removed during pier reconstruction and then replaced, so it wasn’t clear whether the position he had was the current position, or possibly an older one before it had been moved and replaced. My NGS benchmark showed no signs of being moved, and sits on a hill in a National Forest where it’s unlikely there was any reason to ever move it; that doesn’t mean the position I had for it wasn’t wrong.
Peter also asked about quantization of positions, i.e. what was the minimum difference in latitude or longitude for the two GPS receivers I looked at. His older Garmin units only give positions to the nearest tenth of a minute, roughly a meter or so. For both the Garmin and Holux receivers, the value was the same: 2E-06 degrees in latitude and longitude was the minimum difference, corresponding to roughly 0.2 meters N/S, 0.16 meters E/W. So the difference in measured positions is unlikely to be due to differences in positional resolution between the two units. With display set to decimal degrees, my Garmin only shows five decimal places, corresponding to an accuracy of about 2 meters or so, but downloaded data is available with more significant figures. Didn’t look at the NMEA numbers for the Garmin, but for download waypoints, accuracy is given to 8 decimal places, or 2mm accuracy; this is not a degree of precision that the Garmin, or most GPS units, can determine in real-time.
He also noted he had also seen a skewed distribution of positions in some Garmin units, similar to the diagram in this post.
PMarc suggests that I measure the benchmark position with a Differential GPS, to see whether the measured and given positions match. Don’t have access to that, but might be able to borrow a GPS unit in the future with 0.3 cm accuracy after post-processing, and I’ll update these posts if I get more data. He also asks about the stability of the benchmark, and when it was last surveyed. It’s very stable, sitting on top of a hill with a thin layer of soil on top of solid basalt and limestone. There is a fault nearby, but it hasn’t been active during recorded history, and certainly not since the benchmark was installed (1959). The last surveyed position was done in 1992 using “standard geodetic survey methods”; this is a heavily-wooded area, so establishing clear lines of sight for such a survey might have been difficult. Both receivers were pumping out raw NMEA data streams, so there was no screening of positions by either unit based on DOP.
Terry notes that in the position plots shown in the first post, the Garmin values appear to be even distributed around the center point, while the Holux points might have a directional skew. Tough to say from those posts, since the Holux points are very tightly plotted, and you’d need to expand the scale to know for sure (which VisualGPS doesn’t let you do). Follow up plots for both Garmin and Holux units using SA Watch, which Terry hadn’t seen at the time of his comment, seem to show a pretty strong NE to SW axis for Garmin points, with a less-pronounced directionality for Holux points.
Nick Hopton notes that DNRGarmin uses the Proj.4 library for re-projecting coordinates, and that might be introducing errors. Good point, and I’ll try to find a more accurate coordinate conversion program; finding a free one (or even a paid one) that doesn’t use Proj.4 isn’t easy.
Not sure what the moral of all this is; perhaps that if you’re going to use a GPS receiver to measure field positions, using tools like VisualGPS, DNRGarmin and SA Watch to evalute its accuracy, repeatability, and variability might be a good idea.
I’m fairly new to the world of GPS and GIS. I’ve had a hand-held unit for about a year, and have done a small amount of GIS work here and there stretching several years aback. I find “this stuff” useful and fascinating, but – caveat – I’m far from an expert, and some of the info here goes over my head.
That said, I find the “Waypoint Averaging” application in the Garmin Oregon series to be a good ( easy ) way of gauging the repeatable accuracy of the unit. I can first create a waypoint, then take a second sample against the same point, with the unit in the same place. After the two readings are blended together, the unit will tell me how far the waypoint has been moved. If I do this almost immediately, I haven’t seen a point move by more than a foot or two; Garmin suggests waiting an hour and a half, so that the constellation is in entirely different positions, but this seems like a less meaningful test.