[OSM-talk] Administrative boundaries export

Frederik Ramm frederik at remote.org
Wed Oct 2 18:58:21 UTC 2013


On 02.10.2013 18:23, César Martínez Izquierdo wrote:
> I plan to create and make easily available a world-wide administrative
> layer based on OSM data, ideally including existing administrative codes
> (ISO, NUTS in Europe, etc) for each level and producing regular updates
> (for instance once a year).

This is something I have been thinking about for a long time (but never
written any usable code).

Nominatim is probably a good starting point - a better one than MapIt I
should think.

If you're only after extracting certain relation polygons then you could
as well use osmjs (part of Osmium) and have it generate shape files for
you, or adapt the shapefile/ogr export samples in Osmium; this will not
yet give you a hierarchy, only individual boundaries, and you have to
find out the hierarchy yourself.

Finding out the hierarchy is going to be tricky. Nominatim does go to
some lengths to do that already. It sounds easy ("find all polygons with
an admin level smaller than X where this polygon I'm looking at lies
in"). But in reality you will encounter at least:

* missing polygons on all levels - sometimes simply not mapped,
sometimes missing by design, e.g. Germany has some areas where admin
levels 8, 6, and 4 coincide, these are mapped as admin_level 4, so draw
a map of all admin_level 8 areas in Germany and you have lots of holes
in them

* broken polygons on all levels; brokenness changes by the day, i.e.
what is working today may be broken tomorrow and vice versa

* occasionally (e.g. Japan) linear regional boundaries that simply go
from coast to coast without including the coastline

* occasionally (e.g. Chile) a regional boundary that is not a
multipolygon relation but instead a grouping of smaller regional entities

* sometimes small geometric inaccuracies mean that e.g. a state boundary
fails the "is-in" test for the country boundary because there's just a
little square metre somewhere that is mapped as belonging to the state
but not the country

* overlapping admin polygons of the same admin level

I think that ate the very least you need to run the evaluation regularly
and compare: Do I have new polygons this week - have others vanished,
and if so, is that because they were explicitly deleted/replaced, or
were they just accidentally broken and I should continue to use last week's?

What we would really need though, is something much bigger: A separate
database of admin hierarchies, where people could - in a crowdsourced
manner - record things like:

"There is an adminlevel 2 entities called Germany"
"It is divided into 16 exclusive adminlevel 4 entities with the
following names: ..."
"These 16 entities cover the area of Germany completely (no holes or sea
areas that would be outside of one of the entities)"
"The adminlevel 4 entity named 'Brandenburg' is divided in X adminlevel
6 entities..."

and so on. A tree of arbitrary size where people can add and edit at will.

Now you will say "but this tree could be generated from OpenStreetMap",
and I grant that one could attempt to build such a tree but it will
always be faulty and reflect the current brokenness of geometries in
OSM. One could *start* with an OSM-generated tree, but after that, the
tree must be kept separate. People should be able to add stuff to the
tree even when it is not in OpenStreetMap - "there should be an
adminlevel 8 boundary called so-and-so". A regularly-running process
would then compare the tree to OpenStreetMap, and generate error reports
that can be presented visually:

"The tree says that there should be a region called X in Germany but OSM
doesn't have one."

"There is an area here that is not covered by any adminlevel 4 area but
the tree says that taken together the adminlevel 4 areas must cover all
of the country."

"The tree claims there should be a region called X but in OSM there's
only a region with the similar name Y, which one is correct?"

and so on. - I would expect the tree to be much more stable than the
data in OSM. Most of all, the tree could be worked on independently,
even by people unfamiliar with OSM. Of course the tree could link to OSM
objects but these links would regularly be checked and perhaps even
changed by the automated comparison system.

As I said, a simple export is easy but it will have too many weaknesses
- you couldn't even say to what level a country is "complete". Some
people have started to link regional boundaries in OSM together with
concepts like "subarea" but I don't think this can replace an external
country structure tree because the tree could describe what is expected
to be there whereas in OSM you never know if some thing is missing
because it doesn't exist (cf. my example about the "missing" admin6/8
boundaries in Germany) or if it just hasn't been mapped yet, or is broken.

It would be great to see someone drive this important topic but it
certainly isn't something you can set up in a week or two ;)


Frederik Ramm  ##  eMail frederik at remote.org  ##  N49°00'09" E008°23'33"

More information about the talk mailing list