[Talk-us] Nominatim in CDP

[Minh Nguyen]

On 2014-06-24 05:52, Brian Quinion wrote:
> Hi,
>
> So from my point of view as a nominatim developer the issue is that
> there are (at least?) 2 types of city in the USA.
>
> There are administrative cities and there are postal cities.  There
> are also census areas, which may or may not be their own category of
> thing.
>
> At the moment all of these are tagged in exactly the same way:
>
> boundary=administrative
> type=boundary
> plus an admin_level
>
> There is nothing that lets us tell them apart so we are left to pick
> the 'best' city.

"Postal cities" aren't currently tagged as areas, only tag values on 
individual POIs and buildings. Clifford's first example is a building 
that lies outside the Redmond city limits but still carries a Redmond 
address, which is noted in addr:city.

Would it be possible for Nominatim itself to synthesize a "postal city" 
boundary for its own use? I'm unfamiliar with Nominatim's inner 
workings, but it appears to models streets as polygons. How about 
something similar at the city level, taking into account the city limits 
and any matching addr:city values within a certain radius?

There's at least some support on this list for making all CDPs 
boundary=census or somesuch instead of boundary=administrative. It's 
been used extensively in some areas. [1] The Standard stylesheet no 
longer renders automatic "catch-all" labels on large areas, but perhaps 
the openstreetmap-carto developers would entertain a style rule for 
boundary=census and/or place=locality areas.

> Adding individual 'city' tags to houses doesn't help much.  Do you
> mean admin city or postal city? Same issue. In most cases we actually
> want to find and link to both.

addr:city tags always indicate the postal city. boundary=administrative 
areas always indicate the city's administrative limits. Anything else 
should really use is_in:city.

> So - my suggestion would be to introduce tagging that makes the
> distinction.  Ideally to create boundary=postal or something like that
> rather than tagging every road or house with duplicate information.

A "postal city" in the U.S. is just one of possibly several names 
associated with a ZIP code, which is an area by first approximation but 
technically a route. [2] boundary=postal sounds a lot like the USPS's 
ZCTAs; otherwise, who knows the exact boundaries of their own ZIP code? 
I mean, are there "Welcome to Boston Mass 02134" signs anywhere? 
Administrative boundaries are contentious enough on this list, but 
realistically, ZCTAs could only ever be added and edited by bots. Sounds 
like the perfect candidate for an external database. ;-)

In adding a tag like addr:city, I'm making a statement about an 
individual POI's address carrying a certain city. That's often more 
verifiable (via signage, business cards, or takeout menus) than a 
statement that every address in a large polygon carries that address. 
How precise is that polygon?

The standard way to avoid maintaining redundant data is through a 
relation, like associatedStreet relations, but I'm not keen on adding 
every building or street in town to a relation.

[1] http://taginfo.openstreetmap.org/tags/boundary=census#map
[2] 
https://lists.openstreetmap.org/pipermail/talk-us/2009-December/002567.html

-- 
minh at nguyen.cincinnati.oh.us