Monday, 31 May 2010

OS Locator validation mapping for UK OpenStreetMap data

ITO have produced a mapping layer which can be used within OpenStreetMap editors to check for differences between OpenStreetMap data and Ordnance Survey Locator (which has been released as Open Data).

Within the OpenStreetMap editor the new layer appears as a series of boxes with associated names for all streets that are in OS Locator but which are not in OSM within the same bounding box with a small buffer area added.

Normally a box in this layer indicates an error or omission in OSM, however there are also some errors in the OS data. The not:name can be added to the OSM data in these situations and the box will not appear in subsequent analyses and there is less risk that someone transcribing incorrect OS data into OSM at a later date.

An example of an error in OS Locator is the mispelling of Arthur's Terrace in Ipswich has been entered as 'Athurs Terrace'. I have added a tag 'not:name=Athurs Terrace' to the street and it now on longer shows on the analysis.

How many errors are there in OS Locator? I have been through data for Ipswich and have found 7 errors where OS Locator data does not match the relevant street sign. I have added not:name tags to all of these, follow the links to see how they work. It is also useful to add a 'note' field with a description of why the not:name tag has been added.
  1. Berry Road (should be Belfry Road)
  2. Gwendoline Road (should be Gwendoline Close)
  3. Lagonda Drlve (the 'i' in Drive has been replaced with a '1')
  4. The Garrods (should be Garrards)
  5. Inglelow Gardens (should be Ingelow Gardens)
  6. Athurs Terrace (should be Arthur's Terrance)
  7. Fen Bright Circle (it is actually Fen Bight Circles)
In addition to the above I am investigating a number of other potential errors at present which will probably add another 5 or so errors to the list creating a total of 12 out of some 3000 named road names in the town meaning that OS Locator is 99.7% accurate which isn't bad. This test does not of course show the streets which are in OSM but missing from OS but that is a separate test for another day and OSM should perform pretty well on that one.

To access the above mapping from Potlatch add the following string of text into the 'custom' layer box.!/!/!.png

As a technique I would recommend doing a first pass getting rid of the bigger and the less complicated boxes and then do a second pass on the remaining smaller ones as it can be difficult to make sense of multiple overlapping boxes in one pass.

The analysis is current run each night based on the Geofabrik GB data published earlier the same day so changes to OSM will show up in the map rendering not the next day but the day after that. You may therefor choose to do a pass across an area one day and then do a second pass a couple of days later.

Also, do note that the OS bounding boxes seem to have a small rounding error resulting in edges of the bounding box that do not always exactly match the end of the relevant street.


Anonymous said...

Looks good. Any chance we can get this in slippy format somewhere?

Anonymous said...

Currently looks like it's checking against the "name" tag. It would probably be a good idea to also have it check against other tags too. At the very least, "name:en" and "name:cy" - not sure if there are also names in the Locator set in Scots and Gaelic, but if so you could add "name:sco" and "name:gd" to the list. Not sure about "name:ga", since the data probably doesn't extend to Northern Ireland.

Harry Wood said...

Neat. For people wanting to take a quick look: e.g. Edit Darlington in Potlatch with OS Locator Validation Layer

(You can launch Potlatch directly passing tileurl as a URL parameter)

Peter Miller said...

Slippery map format coming soon - just doing some final tweeks to it and then it will be up.

~socks said...

This is great, but I'm doing a small village in Wales, where the street signs are bilingual. OS Locator doesn't seem to be consistent about using Welsh or English. Is there any chance you could look in the "name:cy" and "name:en" keys for what's in OS Locator as well the simple "name" key?



Andrew Chadwick said...

Just tried the layer in JOSM, and it works beautifully in conjunction with the StreetView one. Only minor annoyance under JOSM is having to visit preferences to switch between custom slippymap layers.

Would it be possible to elide apostrophes and other punctuation before comparison? Possessives can occasionally break the match.

I'll second Socks and Anon on internationalized names. Also it'd be great if the tool could match on some of the standard key variations (e.g. alt_name, int_name...) from ?

Another possibility could be to abbreviate before comparison, which might catch variations on the (normally abbreviated, round here) "Saint". Amongst others.

Anonymous said...

Andrew, as I understand it, the OS Locator dataset is unabbreviated, so this shouldn't be a problem. Any discrepancies that arise would be because of people abbreviating names in OSM when they shouldn't be.

Ed Avis said...

Slippy map here:

Harry Wood said...

This evening I've tidied up the wiki a bit and spun out separate pages for OSM and OSL differences tileset, and OSM and OSL differences tileset, linked off OS Locator. Hopefully clearer. There's some details not covered on the wiki at the moment still though. major/minor terminology and the not:name tag.