[Proj] Google Earth Accuracy

[Irwin Scollar]

Jan Hartmann  writes:

>I can confirm Irwin Scollar's conclusions for the Netherlands: we have
>here a list of about 5000 "official"  points, both in the national grid
>(RD, Rijksdriehoeksnet) and in WGS84 as converted by the official
>transformation program. Quite a few (church towers and the like) are
>easily localisable in Google. From what I have seen, there are
>discrepancies with Google, but they are in an order of magnitude of
>meters.

Are the points in that list available on the Web 
without charge?  If so, please provide a link.

>On anything with a scale smaller than 1:10000 there is no real
>problem, but on cadastral levels (I have been working with scales 1:2500
>and larger), you need to be careful when using Google, e.g. in
>archaeological settings. As in the UK, GE data have been derived from
>aerial photographs, manufactured a few years ago on behalf of Google
>itself.

A simple linear correction for an image over a 
small area given the known coordinates of a 
visible point in GE, seems sufficient to bring 
accuracy within near-cadastral levels if the 
source images are from good aerial orthophotos 
like those in many parts of England and 
Germany.  The satellite images from Digital Globe 
used in many more places by Google Earth are not 
quite as useful in this case, but linear 
correction can bring them into fair agreement 
with 1:5000 maps.  Their Spot imagery is rarely 
useful for better than +/- 10 meters, and the 
Landsat imagery in many other places is fairly useless.

Ethan Alpert writes:

>How come no ones questioning the accuracy of the 
>sensors that collected the imagery in Google 
>Maps? I happen to know that satellite imagery is 
>not very accurate unless you have decent ground 
>control and a DEM. Even then it's not that 
>great. It seems pointless to try to determine
>accuracy in Google Maps unless you understand 
>where the imagery came from and how it was produced.

Even without this knowledge, a bi-variate 
polynomial stretch can compensate to some extent 
for sensor error without requiring too many 
control points.  If still better correction is 
required, then a local affine stretch scheme like 
that used for matching old maps with new ones 
having different datums can be used at the 
expense of considerable hand-labour in identifying corresponding points.

>I would expect the situation for Germany, where 
>they produced the aerial photgraphs the same 
>way, to be the same, although I don't have any precise data for that country.

There are 16 sources of trig point data, one from 
each of the mapping agencies of the German 
"Länder". None of these were free of charge the 
last time I looked, so I haven't tested them.

Very accurate results of near-cadastral quality 
were obtained in Kanton Zürich in Switzerland 
where the very high price of land dictates 
digital maps at sub-decimeter levels even though 
the GE imagery is not as sharp as that from 
southern England or the Netherlands. At 10 cm per 
pixel, the match was surprisingly good.

GE uses Plate Carée as is documented on their KML 
developer's web site. I suspect that this is done 
for speed of calculation in an interpreted Web 
environment as well as for relaxed requirements 
for making mosaics from images of different 
qualities.  For small areas, it seems good enough.

Irwin Scollar