[Talk-us] Neighborhoods / Zillow

[stevea]

>OSM has pretty poor neighborhood coverage in the 
>US. We have around 1100 place=neighbo[u]rhood. 
>Geonames has ten times that at 11,000 (feature 
>class P.PPLX - not sure if all of those are 
>neighborhoods) and Zillow has 7,000. Both these 
>data sets are provided under (different) CC 
>licenses. Could we use either Geonames or Zillow 
>to drive improvement to neighborhood name 
>coverage in OSM? I am not proposing an import, 
>but a local MapRoulette challenge might work 
>where people with local knowledge accept / 
>reject proposed neighborhood points, or 
>something along those lines.
>
>Martijn

I don't use points (a POI with place=* or 
neighbourhood=* tag) but rather named polygons 
which surround/define a given named residential 
area.  These seem to work just as well:

What I've done in my city is to get the (public 
domain) digital city data for how parcels are 
grouped together into polygons defining 
residential neighborhoods, with names in the 
name=* tag (and even numbers for each residential 
neighborhood, which I've put into the ref=* tag). 
These get an additional landuse=residential tag, 
and voilá, OSM (the database), mapnik and 
Nominatim all capture/display/index each 
neighborhood properly (Nominatim nicely and 
correctly as "Residential area.")

The same data sets also contain outer-parcel-edge 
boundaries for commercial and industrial 
districts, which of course get landuse=commercial 
and landuse=industrial tags (respectively), as 
well as THEIR name=* (and ref=*) tags.  As a 
result, our city displays very nicely, all 
neighborhoods/districts show up in Nominatim, and 
the OSM database contains definitive, correct 
polygons, straight from a public domain source 
(the city GIS department).

There are a very small number (two, three?) of 
"additional" data points which my neighbors use 
as community names (like "East Park" or 
"Midtown") which the city doesn't actually 
define, but people who live and/or work there do. 
For these, I use place=locality, name=* tags, and 
they render with a slightly different font (and 
smaller type size) than the 
neighborhoods/districts above.  For these, I 
place the point at a significant "cultural 
centroid" for those small sub-communities 
(place=suburb is too big, though I have also 
defined four of those in my city of 60,000 -- 
suburb points also display with 
distinct/different typeface/size, and at 
"certain" zoom levels which make it clear they 
are suburbs).  From both an "in the OSM DB" and a 
"how does mapnik display this" (in addition to 
how Nominatim indexes), I believe this is 
completely correct, and they look nice, too.   I 
sincerely believe anybody who lives in these 
neighborhoods would agree.

I would guess many medium- and larger-sized 
cities have these sorts of datasets available: 
they are just big polygons that surround a 
neighborhood or commercial/industrial district: 
no "single point" required.  While these might 
take up more space in OSM's database, the extra 
points for the polygon-defining way makes them 
quite exact, and mapnik's rendering is in the 
very center of each polygon:  a nice way to do it.

I invite you to take a look (within the City limits):

http://www.openstreetmap.org/?lat=37&lon=-122&zoom=14&layers=M

I don't think Zillow or Geonames should be leaned 
on too heavily (if at all) to define these: 
where neighborhoods begin and end is very much a 
local thing, and usually the City itself (or the 
County for unincorporated areas) or people who 
live locally are best at defining these.  That's 
why I'd say MapRoulette is a poor candidate for 
doing this:  you won't get local knowledge, 
you're just crowd-sourcing what effectively 
becomes an import among many, and they don't 
really know whether the data are high quality or 
not.

SteveA
California