[OSM-talk] quality of drawn way

[Mike Collinson]

At 09:59 AM 10/05/2007, Steve Chilton wrote:
>Does anyone have any hard facts about relative accuracies of NPE and
>Landsat, or is it variable and unquantifiable?
>I traced quite a few small areas of water in Lake District from NPE the
>other day and was wondering whether to use the shape as per NPE but
>shift each one to fit the Landsat for position. Windermere and Coniston
>Water were already there and seem to fit pretty much to Landsat (may of
>course have been traced from Landsat).
>Any thoughts?

They are complementary tools.

I find Landsat images to be very accurate quantitatively but it is easy to make qualitative mistakes, such as continuing a road down the path of a stream or losing features when the contrast changes such as entering a wood or town.  When aligned, I can generally work in a 50 x 50 km area and surprisingly get features down to almost the Landsat pixel resolution of 14 metres when I do a quality control check by GPS traversing lake and river banks.

NPE, (digitised out of copyright maps available to UK mappers), is the reverse. I estimate that anything outside a roughly 3km x 3km alignment box can be up to 50m out. But for relative positions and identifying features as well as using feature names, it is just great.

To amplify, I've mapped a large number of different areas and usually run a reference line, driving or cycling some long feature visible on Landsat.  I them use that to align images within JOSM.  To date, I've always found that the Landsat misalignment, if any, is systematic and good for the whole area of interest (say at least 25 km in any direction).  I'd hesitate to say that is a hard fact as I've yet to determine whether the change between area A with a 30m misalignment and area B with perfect alignment is gradual or abrupt.

Arriving recently in the UK and applying the same principle to NPE, I find it is only possible to align very, very locally. I'm not familiar with the stretch properties of  map paper over time but the size and local randomness of the misalignment makes me suspect that it is due to inaccuracy in the original survey and not paper or scanning artifacts.  I'll be paying more attention to the apparent location of the grid lines to confirm that.

If anyone is interested in looking at this further, go to Wharfedale, Yorkshire, UK http://www.openstreetmap.org/index.html?lat=54.12364148710909&lon=-2.048039022841875&zoom=11 in JOSM.  The roads are GPS surveyed and marked source=survey. The Landsat data needs shifting a few pixels due south. The NPE match is good overall but wanders at the local scale.

Mike