[Talk-ee] Administrative division

['Andre Grueneberg']

Jaak Laineste wrote:
> I have the scripts what I got from Emilie, not the ones in wiki. These
> worked quite fine for my test database upload.

So maybe you just send me a copy of it, so I can be sure to work on the
right code base. :) I admit, I could just start working and merge it
later -- I'll have to learn a bit (more) of Python first though. ;)
 
> But: I discovered now that my merging of corine with better coastline has a
> lot of tiny "topological dust": very small overlappings and
> self-intersecting areas. MapInfo found 2400 of them, and each can be
> manually fixed, but it took about 15 minutes to fix 30 of them.

At least in that case, I'd just import them to OSM and start fixing it
later. This can then be done in a joint effort ... all you need to do is
to provide an OSM file (or WMS service) so people can spot the issues.

> www.maakaart.ee/corine_ee_uncleaned.zip has the files in Shape format, so
> you can see them in QGIS e.g. - one file for corine polygons, and as 2 for
> error checking results. Maybe some errors can be automatically fixed, but
> not all of them. For automatic fixing some other workflow/tool is needed to
> update Corine with proper coastline. Or just forget about full merging: it
> would be easy to have have first step, "masking" for proper sea and this is
> it. In my process I tried to have full merging: I extended Corine areas to
> coastline, where there would be small holes; and I already did a lot of
> manual work to overlook strange bigger areas. Now small technical (possibly
> mostly non-visible for renderer) topological errors are left.

There're many things you can easily correct later on. It's a matter of
the effect vs. costs...

Anyway, one of the issues (even looking at them takes a while) in the
area errors file seems to stem from incorrect node selection for merging
-- I've got no clue how this can happen. Seems like a spike between two
nodes close to each other. This might even be detectable in a script.
For each way check the sequence of nodes. If node n is much closer to
node n+2 than to node n+1, then delete node n+1 from the sequence.

Okay ... as usual I gave it a try and hacked some proof of concept ...
http://andre.grueneberg.de/osm/remove_spikes.pl ... at least it's
removing 3543 "suspicious nodes" ...
But maybe I'm totally off the track ...

BTW: I'm usually not using a fully fledged GIS ... I once tries QGIS and
didn't feel to comfortable with it ... I'm just wondering when we're
getting ESRI shape support in JOSM. :)

Andre
-- 
Bear Whiz Beer : It's in the water thats why is's yellow!
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: Digital signature
URL: <http://lists.openstreetmap.org/pipermail/talk-ee/attachments/20090908/5a5443ca/attachment.pgp>