[Geocoding] OSM Nominatim: additional data source for addresses in the US (and elsewhere?)

[Sarah Hoffmann]

Hi Tobias,

On Thu, Mar 22, 2018 at 04:20:54PM +0100, Tobias Wrede wrote:
> today I tried to look up an address on openstreetmap.org. It was "35516
> galen place, fremont". I got a result from Nominatim in the right spot on
> the map (confirmed by Google Maps). I was then very surprised to not find
> the address in the OSM data. Only the residential "Galen Place" is mapped,
> no individual houses let alone addresses though, nor any interpolation.
> 
> After some discussion on the German forum
> (https://forum.openstreetmap.org/viewtopic.php?pid=690815) it was speculated
> that Nominatim draws from TIGER database (|SELECT place_id||FROM
> location_property_tiger |in the debug output).
> 
> Could someone enlighten me how and in which locations Nominatim uses
> additional data to OSM? What is the priority?

Nominatim currently uses 3 additional data sources:

- US postcodes (from Tiger data)
- UK postcodes (OS CodePoint databse, iirc)
- US housenumbers from Tiger data

All these sources are secondary, i.e. they are only used
when data is missing in OSM. The postcodes are added to addresses
that do not have a postcode already and are also returned as a
direct result when searching for a postcode.

The house numbers are only used to pinpoint a more specific
location on the street. That means that the search only works
if the street itself is already in OSM. Pretty much all of the
result is then derived as usual from OSM data and only the location
is taken from the Tiger data.

In theory, Nominatim has the ability to add more external sources
for postcodes and house numbers. In practice, it is unlikely that
the official instance on nominatim.openstreetmap.org will ever get
more external sources becuase a) OSM coverage has vastly improved
since the original sources where added, b) licenses are always a
pain and c) nominatim.osm.org is mainly meant for mappers who might
get a wrong sense of completion when external data is used.
(Naturally, the situation is different for private installations
where external sources might be helpful but that's a different topic.)

Kind regards

Sarah