Skip to content
Prev 8532 / 29559 Next

georeferencing point shape

On Tue, 15 Jun 2010, Tom Gottfried wrote:

            
There is a problem with the known points too, in that their surveyed and 
GPS interpoint distances differ somewhat:

dist(cbind(gps_x, gps_y)[17:20,])
dist(cbind(x, y)[17:20,])
dist(cbind(gps_x, gps_y)[17:20,]) - dist(cbind(x, y)[17:20,])

This means that with only 4 GPS points, it is hard to model the 
transformation with more terms:

df <- data.frame(gps_x, gps_y, x, y)
lmx <- lm(gps_x ~ x + y, data=df[17:20,])
lmy <- lm(gps_y ~ x + y, data=df[17:20,])
nx <- predict(lmx, data.frame(x, y))
ny <- predict(lmy, data.frame(x, y))
dist(cbind(nx, ny)[17:20,])
dxy <- dist(cbind(x, y))
dnxny <- dist(cbind(nx, ny))
all.equal(dxy, dnxny, check.attributes=FALSE)

More GPS measurements would have helped, but at this scale, GPS are always 
going to be approximate. You cannot measure the centres of the GPS and 
surveyed points accurately either, I'm afraid. To fix things, you would 
need GPS measurements in the precision of the data set, so in most cases 
like this DGPS rather than GPS. If one of the surveyed points is clearly 
visible on a registered high resolution image, you could use that instead 
of the GPS - something like the corner of a building (measured to about 
1cm and known in the projection of interest). However, planning this 
before doing the fieldwork was the only effective remedy.

Roger

PS. Do you know that the GPS was using the ellipse you specify? Which 
datum are you assuming (WGS84)?