[OSM-dev] OSM based replacement for builtup_area.shp in standard style

[Martin Koppenhoefer]

2014-06-03 18:46 GMT+02:00 Christoph Hormann <chris_hormann at gmx.de>:

> This is of course a fairly ressource intensive process with all building
> and road data in the database being used as basis.  But in principle it
> would be possible to run this with periodic updates in a similar way as
> currently done for the coastline.
>
> Everyone feel free to try out these polygons in your maps.  I have not
> tried this myself in the standard style yet, it is likely that
> adaptations in either the generalization process or the style (or both)
> are necessary for best results.  Comments and examples on that are
> welcome.
>


First of all I really appreciate the work and dedication you have invested
into this topic. Using external datasets from natural earth data has helped
us in the beginning of the project to produce better maps, but it is IMHO
clearly desirable to move these few uses in the future to osm-only derived
datasets. Your blog post and process is for sure an important contribution
to this discussion.

As a sidenote I wanted to point out that there are currently already 281547
ways (I suspect most of them to represent areas) tagged with place=* (not
much compared to a total of 3.1M places in OSM) which could serve as an
alternative to your (preprocessing intensive) process if more mappers could
be convinced for this concept of mapping settlement extensions.

The reasons why settlements mostly aren't mapped as areas lie mainly in the
current main map style sheet (read: "mapnik"), which ignores places on
areas due to technical limitations and because the place-nodes are needed
for routing and label placement, something that might change with the
progress of the project and new ways of structuring the data (e.g.
place-nodes and place-areas could co-exist, with appropriate rules how to
avoid redundancy, e.g. a place-relation combining the two and getting the
tags, the place-node could get the role "centre" or "central_point" or
"label"...). Currently doing both, place on an area and on a node, is
perceived as "duplication" and violation of the "one feature, one
object"-rule, but this is really just a question of definitions (e.g. the
node could be called "place centre" and the area "place extension"). I
think it is obvious that a node will never be the best representation for
something as large as a settlement (go and ask the Nominatim coders ;-) ).

The main problems of your approach --- it is still only "guessing" and the
derived dataset might have a high probability to identify "built-up" areas
but it will not in all cases be able to tell which settlement an area
belongs to and it is a ressource intensive process that reasonably cannot
be done on the fly --- could be overcome with explicit place polygons. On
the downside it will of course take us quite some time to manually map all
those settlements, and it is not even clear if people are interested in
general to map this kind of feature (in addition to landuse etc.). If the
main style picked this up and rendered areas with
place=(city,town,village,hamlet,isolated_dwelling) as "built-up-area" this
would surely help promoting the cause ;-)

cheers,
Martin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstreetmap.org/pipermail/dev/attachments/20140603/9a7994d9/attachment.html>