Discussion Page   Home     Help   Index     Login
Accuracy of Google Satellite View #3792
Back To Discussion List Written: 2011.12.09 by: Robin Tivy

Just how accurate are the google satellite photos? I'd certainly like to hear any experiments or observations from other people. My current belief is that they are can be accurate to within 5 meters, but I've seen at least one (Manning Park) that is out by 30 meters.

In previous studies the Bivouac scientific department has determined that GPS units can be accurate to within 5 meters. See GPS Accuracy Experiments. And our standard is to enter lat-longs with 5 decimal places, which allows precision of 1 meter.

We have the ability to zoom in on satellite photos and get readings to within a couple of meters. The only problem is, if you go and measure the same with a GPS, you can be up to 30 meters different.

Here are some test cases:

  1. 2150 Macdonald, Northmost line on crosswalk
     In this experiment, I proved that at least some of the google satellite images are accurate within 5 meters. That is, they agree with a GPS within 5 meters. The way to prove this yourself is to pick a convenient point outside your house that is visible on the satellite, and has good GPS reception. I picked the point where the northmost line on the crosswalk hits the sidewalk. I went out and got the GPS reading for that, then I brought up any GMap display, and searched for my address. Eg: Search for "2150 Macdonald Street, Vancouver". Now I can zoom the satellite view and pick out the point to within 1 meter. So here are the results (today):

      From Satellite: 49.26671,-123.16828
      From GPS measure: 49.26654,-123.16833
    Conclusion: The GPS reading for longitude only differs from the satellite by .00005, which is less than 5 meters. That is as accurate as you can get with a GPS anyway. So this is not an example where google is wrong.

  2. Case 2 - Upper Similkameen Fireroad
     I suspect that there is something wrong with the Google Maps calibration of Manning Park lodge area. One clue is that all the peak locations obtained with BC Basemap seem to be to far north from where the peak would be on the map. Eg: Norheim Peak. Because it is harder to get to the peaks than to chart a road, I charted a road, and then compared my track log with the map and the satellite. I can assume the track log is within 5 meters because the shape of the road is correct.

    So I uploaded the tracklog to Similkameen Headwaters Fireroad. If you then look at the MyTopo view, you can see the GPS log is shifted south by about 90 meters. (Notice crook in road). Now look at satellite view. Its hard to tell where the creek really is. However, I can see that the track log lies about 30m east of the obvious cut in the trees which is the road. So this is a case where the satellite view is out by 30 meters. This may be due to mis-calibration by Google, or whatever. So an actual GPS reading is more accurate than you can get using Google maps.

  3. Case 3 - Manning Park Sheet - Nordheim Peak
     If you type in "Manning Park" and then click the GMap link to bring up the map, you will notice that the Bivouac position of some of the peaks like Nordheim Peak looks wrong. Bivouac shows the peak as being part way down the northern slope. Yet if you look up the peak in BC Basemap, you notice the peak exactly corresponds with Bivouac.
    BC Basemap  49.17556, -120.90472
    1:50K       49.17559, -120.90631 
    Remember that for latitude, the 5th decimal place is roughly 1 m. So The difference between the longitudes is 631-472=159 m, say over 100m. So something is wrong with either BC Basemap or the Google 1:50K. I then checked the position in Memory map, and it agreed with google. But of course they could have both used the same wrong calibration data.

    But it's hard to bust the 1:50K maps, because I'd say they are only readable to 100m anyway, so the map isn't really wrong.


n order to discuss how much readings are varying, we need to be able to know roughly how much .00001 degrees is in meters.

 - 90 Degrees is 10000 km (By definition)
 - 1 degree is 111 km, which is 111000 meters
 - so .00001 degree latitude is 1.11 meters
 - so two latitudes like 52.88827 and 52.88831 are 4 meters apart

But the distance of .0001 degrees of longitude depends on how far north you are. Eg: At the equator (0 degrees), .00001 degrees is 1.11 meters. But at 50 degrees north, it is only 0.71 meters. Here is a table:

Latitude  Degrees      Meters
0	 0.00001	1.11
30	 0.00001	0.96
35	 0.00001	0.91
40	 0.00001	0.85
45	 0.00001	0.78
48	 0.00001	0.74
49	 0.00001	0.73
50	 0.00001	0.71
51	 0.00001	0.70
52	 0.00001	0.68
55	 0.00001	0.64
60	 0.00001	0.56
89	 0.00001	0.02


#1607 - 2012.01.12 Paul Kubik - Re: second track log
The difference is possibly due to the relative satellite positions being different on the days you took the measurements because, as Steve mentioned, the satellites are not geostationary.

#1606 - 2012.01.10 Robin Tivy - Second track log much closer to satellite view
I took a second track log on Dec 27th, and it was much closer to the satellite view. That wasn't what I expected. In light of this evidence, I now have less faith in GPS track logs and more faith in the satellite view.

#1605 - 2011.12.18 Steve Grant - Books about Digital Mapping
Along with the development of mapping technology, there are also books on this subject. An example is "Map Scripting 101". It's likely these books can answer many of the questions raised on this subject.

Table of Contents

Chapter 1: Mapping Basics

Chapter 2: Plotting Markers and Message Boxes

Chapter 3: Geocoding

Chapter 4: Layer It On

Chapter 5: Events

Chapter 6: Explore Proximity

Chapter 7: User Location

Chapter 8: Geodata Formats

Chapter 9: Go Server-side with PHP and MySQL

Chapter 10: Mashup Projects

Appendix A: Introduction to Javascript

Appendix B: Mapstraction Reference

#1604 - 2011.12.13 Robin Tivy - More Experiments out on the net
There is a discussion on a website called POI-Factory which has a couple of useful postings, mixed in with some general confusion:

Here are the postings:

CaseyGuy says: "For one latitude measurement (and this was typical), my GPSr read 38.87925 deg at a spot that Google Earth said is at 38.879218 deg lat. "

R Tivy says: Since the 5th decimal place is 1 meter, the two readings are 4 meters apart (1 versus 5 meters). That conforms with what I have observed, EXCEPT for Manning Park.

The other useful posting is from "ochsfisher" who says: "Google earth is very accurate in its coordinates. i have uploaded many boat ramps to the us boat ramp file using google earth. and some of them i have reverified with my gps driving to the locations. so i would say that it is very accurate. really though your gps is only accurate up to 10 feet anyway because of government regulations. "

#1603 - 2011.12.13 Paul Kubik - Mis-calibration?
Not sure what calibration has to do with the discussion. Distortion at the edge of sat photos doesn't really have anything to do with the georeferencing of the photo. Distortion can play a role when trying to line up a GPS track to a photo feature. If the distortion is bad near the edges it will make the track appear inaccurate when it is overlayed on the photo. Georeferencing calculates a formula or other mathematical construct to project real world coordinates onto the photo surface.

#1602 - 2011.12.12 Robin Tivy - Distortion of Satellite image at edges
Paul mentioned that satellite images could be off at the edges due to distortion. That is true of air photos, but satellite photos are from out in space, so it is doubtful there could be the same distortion. So I think its not distortion, it is mis-calibration.

#1601 - 2011.12.12 Robin Tivy - Answer to Scott Nelson - RE Satellite off by 20m
In the detailed experiments I documented in 2009 on the Burrard Bridge the maximum variation, over hundreds of measurements on 3 different GPS units was 7 meters. See GPS Accuracy Experiments. The lowest "Easting" was 890 and the highest was 897, with over 95% falling between 892 and 895 (4 meters). The variance of the northing is about the same.

Of course, my experiments only show consistency over many days between units. I guess ALL of those measurements could have been off by 20m. But I've never seen a GPS be systematically off by 20 meters, as Scott suggests, and I suspect that number is wrong. I do agree that it is necessary to go on multiple days.

I assert with 95% confidence, that GPS units are not systematically off by more than 10 meters. I would like to see proof of a situation where they can be off by 20 meters.

What I need to do is view the burrard bridge point with the Satellite view.

#1600 - 2011.12.12 Robin Tivy - Rubble Creek GPS track log is within 10 meters
In the interests of science, I suffered to ski up the icy Garibaldi Lake trail to Taylor Meadows yesterday to get some waypoints from a different location. I had my GPS in my pocket, taking a track log the whole time. Then today I checked the track log against the Rubble Creek road, which you can see clearly on the satellite. My waypoints were within 10 meters of the satellite image. Please have a look on GMap for Rubble Creek Road.

So my conclusion from this is that the Garibaldi lake satellite image is correctly calibrated. I'm also still operating from the viewpoint that the GPS tracks are normally correct to within 10 meters.

I also took a specific waypoint at the parking lot, standing right in front of the parking meter. That point was at 49.95741,-123.12004=Parking meter @562m. On the google image, unfortunately you can't see exactly the parking meter, but the waypoint looks very close. It lines up with the 4th car. In summer, someone could go up there and get an exact waypoint at a specific parking lines.

I also took a waypoint in front of the warming hut in Taylor Meadows. That point is at 49.95270,-123.07395. I can't see any hut at that point on Google, does anyone know exactly how to find the warming hut on the sat. photo?

#1599 - 2011.12.12 Scott Nelson - GPS accuracy can be off by more than 5m, even if the shape is correct
I generally find that my GPS unit is sometimes off by a constant value (say 20m) but always in the same direction. This leads to a constant positional error, but the shape of the track is correct. Often the problem is that the GPS thinks it's at the wrong elevation, which can result in a horizontal position error. It's impossible to tell the difference between such an error and a projection error by Google. The only way to tell is to do multiple GPS tracks of the same road on different days to get an averaged track.

For the same reason, averaging a GPS measurement over a couple minutes isn't really going to improve the accuracy. To do that you need to average over several hours as the satellites move past you.

I have two GPS tracks from the Manning Park Lodge area and they are both abbot 15m west of the road/trail locations in the google earth images. Both tracks are from trips to Frosty Mountain. But on the one track near the top of Frosty Mountain the GPS track is too far east instead, so it's not consistent. The other track was from spring so it's not possible to tell anything in the alpine.

#1598 - 2011.12.11 Vito Gudaitis - More on GPS Satellite Orbits
It should be noted that GPS satellite orbits are inclined 55 degrees to the equator; so they range between latitude 55 degrees north and south. This means that there is no advantage in GPS reception to having a view to the south. A view to the north is just as good. This also allows the GPS system to work at the poles. If the satellites were in geosynchronous or equatorial orbits, that wouldn't be the case.

#1597 - 2011.12.11 Steve Grant - GPS Satellite Orbit
The GPS satellites are in medium earth orbit. Not low, such as the space station, but not as far out as geosynchronous. Medium orbit is about half as far above the earth surface as geosynchronous. The gps satellites do/must move relative to the ground.

#1596 - 2011.12.11 Paul Kubik - Comments on Robin's comments
1. Addressable pixels

Actually you can't zoom in to whatever scale you want. GMap4 only provides a discrete number of scales. (ZoomMap will allow any scale but it doesn't show a background map.)

2. GPS error

Not sure you can eliminate GPS error. Aren't the GPS satellites in geostationary orbit? Meaning they don't move around the sky. Therefore, the error will be invariant in a small area all other factors being equal. The main factor here being that where Robin took his upper Similkameen track is uniformly in the lee of Mount Kelly. The point is that the "lee" here means shielded from the satellites. If you're near a large mountain or in a canyon satellite signals are blocked by those features. I also seem to recall the nodes for geostationary orbit are near the equator so under north faces there will be limited numbers of satellites in view and leading to larger errors.

3. Ground reference points

This is the most germane point, I think.

#1595 - 2011.12.10 Robin Tivy - Need some more examples and hard data
I studied Paul's notes and have a few comments.

1. Addressable Pixels:
 First, it is not an issue of addressable pixels on the computer monitor, because you can zoom in to whatever resolution you want.

2. GPS Error:
 What we see in Manning Park is that it is UNIFORMLY shifted to be out by 30m. That also rules out GPS error, because if it was a GPS reception issue, the GPS line would not be so consistent.

3. Ground Reference Points
 Of all the points Paul brings up, I think it is most likely that the satellite photo is not calibrated using reliable ground references. What would be interesting would be for people to get more reference points, to compare with the satellite versus GPS.

4. Pleasing Maps Discussion
 Paul says: "I'd say it's better to just try to get pleasing maps using the technology, i.e. so that road tracks in Bivouac overlay roads accurately on the Google satellite photos. "

That question becomes: What is the best policy, when a GPS track log says one thing and the satellite is shifted? I'd say that for trails, a reliable GPS track log trumps the satellite photo. I propose that for the Similkameen road, the official location is the GPS track log, not the satellite photo, despite the fact this is not as pleasing when you look at such track log super-imposed on the satellite view.

However, in cases where a specific road junction is being discussed, the satellite would normally be used as the authorative reference. (Because we don't normally have a GPS reading).

Apart from theories, what I'm really hoping is that someone else can actually gather some hard data. What we want are more actual "reference points", where someone has a GPS reading for a satellite visible point. Only then do we know the extent of the problem. It is possible that there is just a one time error at Manning Park, and everywhere else the GPS corresponds with the satellite. (as in Vancouver). Or it may be that the vast majority of satellite photos are out by 30 meters. Which is it?

#1594 - 2011.12.10 Paul Kubik - Google is a projection
Google uses the Mercator projection. The advantage is that the earth's surface is projected onto a square map so it works nicely with rectangular computer monitors. Mercator is horribly distorted towards the poles by greatly expanding the surface area in the longitude direction.

A satellite photo is distorted towards its four edges. So to map a sat photo to a Mercator projection, another projection is used. While it is undoubtedly possible using high precision mathematical formulas to do that, the satellite photo must be georeferenced to ground control points. That means that there must be several identifiable points on a sat photo in which the lat-lon coordinates can be accurately identified. In a grid system such as the Vancouver road layout, that is easily possible. Vancouver has many survey points embedded in crosswalks etc. throughout the city.

I don't imagine there are many such survey points on a Manning Park map. So, my guess is that the lat-lon of the photo's corners are calcualted from the satellite's own spatial location above the earth. This is yet another calculation and sure to be a level of imprecision. In a series of reprojections, error can magnify. Alternatively, a photogrammetist could estimate several lat-lon points on the photo to establish the control points for georeferencing. That could be very error prone unless there were identifiable survey points in the photo.

The other thing to consider is that the computer screen is composed of addressable pixels. Depending on the scale, one pixel can represent one metre, 30 metres, 90 metres or whatever.

Finally, there is the GPS error itself. If you are in the upper Similkameen River valley with high mountain walls to the south, the relative satellite positions may not be conducive to high geometrical precision.

It's another reason not to put too much trust in technology when navigating in the mountains. I'd say it's better to just try to get pleasing maps using the technology, i.e. so that road tracks in Bivouac overlay roads accurately on the Google satellite photos.