[Tagging] boundary relations and the subarea property

[André Pirard]

On 2015-11-26 20:24, Colin Smale wrote :
> I use the subarea member because it makes cross-checking easy. Have
> all the lower-level boundaries in my higher-level admin area been
> added to OSM?
>
> Unfortunately the various admin levels do not always form a strict
> hierarchy. A small area at (lets say) admin_level=10 might be enclosed
> spatially by entities at level 8, 7, 6, 5 etc but it only has a direct
> administrative relationship with one of them, which might not be the
> next-highest level (next-lower number).
>
Unlike those who seem to want to fight against subareas, I don't want to
fight against admin levels.
But, thinking twice, as I once wrote, the levels are unnecessary and
even cumbersome as you say.
The boundary tree can be made only with subareas.
The only problem is pairing similar levels.
And it can be done very efficiently by using e.g. level_type=* names
like municipality and province.
And then you can pair very simply a municipality like Brussels with the
other municipalities.
Even though Brussels is not in a Province.

The only problem is that programs like Nominatim should recognize the
subareas.
And, as a bonus, they could call a village a village instead of a "this
or maybe that".
Limite de village ou arrondissement municipal La Reid, Theux, Verviers,
Liège, Wallonie, 4910, Belgique
<http://www.openstreetmap.org/relation/5687890>
That is wrong: La Reid is a "section".
Limite communale Theux, Verviers, Liège, Wallonie, 4910, Belgique
<http://www.openstreetmap.org/relation/2416084>
Limite de village ou arrondissement municipal Theux, Verviers, Liège,
Wallonie, 4910, Belgique <http://www.openstreetmap.org/relation/5687891>
The first is right, but the second is wrong, a "section" again and hence
the names should be Theux, Theux, ...
Village La Reid, Becco, Verviers, Liège, Wallonie, 4910, Belgique
<http://www.openstreetmap.org/node/766816145>
I still wonder how Nominatim can invent that. Obviously, click it, Becco
is a village inside section La Reid
<http://www.openstreetmap.org/relation/5687890> and it should be "Becco,
La Reid, Verviers, ..." and not a section inside a village

Obviously, Nominatim would have a simpler logic if it used the subareas
(and if roads weren't split).

> Finding the boundaries of all districts within a county (UK example)
> becomes trivial with the explicit parent-child link. Otherwise its
> like finding all boundaries with admin_level=8 which are at least 99%
> contained by the higher-level boundary. That sounds computationally a
> lot more complicated to me. Why not 100%? Because sometimes the
> boundaries at different levels are not imported/drawn from the same
> source, leading to the boundaries not being exactly coincident. So
> there needs to be some tolerance in there.
(1)

I have been very surprised that "Limite de département Bracknell Forest,
South East, Angleterre, Royaume-Uni
<http://www.openstreetmap.org/relation/113682>" (certainly not a
"département") is a subarea of county Berkshire.
But also that Berkshire has no admin_level and that's why Nominatim
doesn't place it in Berkshire but in South East England. They call
Berkshire a "ceremonial county". It's now 55 years I'm acquainted with
England and I've never heard anything else than "county" before.
Typically English.  And I'm sure most Englishmen would say that
Bracknell is in Berkshire without restriction.
Should we remove subarea Bracknell?
Now, let's not cheat...
Englishmen, please name the départements, well, parts of Berkshire. 
Unitary Authority OSM seem to call them, Borough as I see on one of
their Web pages.
Easy some will say: you just need this tool
<https://osm.wno-edv-service.de/boundaries/>. Problem is that it's
nowhere mentioned in OSM.org's Help, first reason being that there is no
public help. And you must subscribe to OSM!!! Well, let's get over that
and use the tool.  After trying to move the map, you'll want to search
for Berkshire, won't you? Ahem, try. Got it?
Now let's search for Berkshire in OSM.org: Ceremonial Berkshire,
Royaume-Uni <http://www.openstreetmap.org/relation/88070>  (Not in South
East England? How do they not know).
And now let's look down the left pane, and read Relation XXX as subarea.
We've got them. A bit techie admittedly, but remember it's a 0-lines
program (app for newcomers).

And now what we see is that only the subareas know not only the
Lieutenant's but also the Postman's Berkshire.
I would ask Nominatim to support subareas for those who don't like
things like is_in=Brussels, Brussels, Belgium
or is_in=Liège, Liège, Liège, Wallonia, Belgium.

Cheers

André.


(1) The fastest algorithm to see if a point is inside an (convex) area
is to turn a clock-hand-like arrow around the point with the arrow head
following the way and see if it turns 0° or 360°.  Fortunately, doing
that at 4 90° spaced points is enough for a gently shaped area. Repeat
that for all necessary "presumably inside" points.

> //colin
>
> On 2015-11-26 19:51, Martin Koppenhoefer wrote:
>
>> I just noticed that a lot of boundary relations have the lower
>> ranking parts included as members with the "subarea" role.
>> This role is documented here:
>> http://wiki.openstreetmap.org/wiki/Relation:boundary
>>
>> But I wonder how it got on this definition page. Was this discussed
>> anywhere? I don't think it's a good idea to add all those lower
>> entities in nested relations (they are already spatially structured,
>> this is redundant and makes the relations more complicated for no
>> good reason).
>>
>> I propose to remove this property from the definition page and move
>> it to the talk page.
>>
>> Comments?
>>  
>> Cheers,
>> Martin
>>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/tagging/attachments/20151127/57ab14d4/attachment.html>