GPS accuracy

I have read various threads (some from back in 2009), and have a general feel, but nothing that goes to core of my question. It also may be that the accuracy of GPS devices has changed (improved ?) over the years.

If I hold a smartphone flat (e.g. GPS capable iPhone), while standing over a specific location [1], how much can I trust the GPS coordinates that it records ?

This question is complicated by the fact that iPhone/iOS implements 3 complimentary GPS approaches. First, true satellite GPS, second cellular-tower based AGPS, and third wifi access-point based skyhook-style GPS.

Assuming that I have #2 and #3 disabled, and I’m using only the satellite receiver, is the recorded GPS considered valid enough to reposition a road intersection in OSM ?

The question also speaks, indirectly, to the quality of the original road coordinates. Were they traced from satellite photos vs were they loaded from government survey data files vs some other source.

One thought that crosses my mind is to track down the surveyor that placed the disc, and see if they could tell me what coordinates they recorded vs what I record.


[1] specifically, standing over the survey disc, which is nailed to the pavement, at the exact center-point of the intersection of 2 roads.

You must only use true GPS for OSM mapping. AGPS normally doesn’t mean the use of cell tower data, but rather distribution of ephemeris data over the internet, rather than waiting for it to arrive from the satellites, but cell tower data is inadmissible. Also beware of devices that snap the coordinates to an internal map.

For a single reading, I would say the road would have to be out by 200ft before you could justify moving it. For a half hour average with the whole sky visible, I’d still think you would want an error of more than 50ft to justify a move. In some cases it may be better to accept a large error in order to keep the local geometry correct.

Given that government survey data is public domain in the USA, I expect that to have been used for most US roads. Even then, the coordinates may have significant errors. I don’t know to what extent roads have been resurveyed with modern, radio based, devices, but the original triangulation surveys could have quite large large scale errors.

The main relevance of GPS to OSM is that it enabled amateurs to create usable maps without relying on copyright data, particularly in places, like the USA, where government mapping data was restricted. Although it is still restricted, access has been given to data that is generally better than GPS, and aerial imagery is generally better than GPS.

To add a little more food for thought on this subject … The road I live on is a rural paved road. It carries a moderate amount of traffic (DOT numbers show 500 travels/day is each direction). The pavement is 20-21 ft wide, i.e. one 10-ft lane in each direction. The full R/W width appears to be 100-ft. So 40-ft shallow culverts on both sides.

The difference between my smartphone GPS and Bing satellite appears to be 5-15 ft (once it has stabilized, but even then it floats around a bit).

Hopefully vehicle navigation systems have enough smarts to recognize that the map tracks may deviate from reality, and that reality may deviate from what the vehicle’s GPS is presenting.

In this type of situation, being off anything more than 100-ft seems unacceptable. +/- 50-ft is probably a more reasonable goal. I’m thinking that most of the intersection points I have measured are falling in the +/- 15 feet (~5 meters) difference from what OSM (and Mapnik) are telling me. Short of having a very high quality unit, that I could leave on that spot for 30 minutes, I’m not sure anything better can be obtained.

You’re in Florida, so your road probably originally came in via the TIGER import, and is probably not well aligned.

Feel free to use the GPS traces that you capture to ensure that aerial imagery is properly aligned within the OSM editor, and align the road in OSM to that. Individual GPS traces can be inaccurate, but the average of traces will be less so. Older traces are more likely to be from devices that don’t support WAAS / EGNOS, and are likely to be less accurate.

However, having lots of traces allows you to easily see outliers, and iD’s GPS trace layer shows direction too, so you can take into account the hysteresis effect that you might get with fast moving vehicles and corners too. Unless you’re on the side of a mountain you shouldn’t see regular offsets in the same direction in GPS traces, so you can get accurate data by “manually averaging” in the OSM editor. Also look at e.g. the Strava layers for point cloud data.