[postgis-users] How to design a database for continents, countries, regions, cities and POIs?

[pcreso at pcreso.com]

Hi Michal, 

One suggestion...

There are two ways (at least :-) to do this in a RDBMS. You can have the spatial relationship implicit in the feature geometries, so a spatial query is used, for example, to determine the cities within a country:

select * from polygons a, polygons b
where a.type = 'city' 
and b.type='country'
and b.name='Italy'; 

While flexible & effective, relying on spatial queries for quick searches with polygons with many thousands, or even millions of records may not be ideal.

The other approach is to explicitly predefine these relationships, so a column for each polygon feature stores the parent id. Simplistically assuming the "parent" of a city is the country containing it, rather than navigating the hierarchy, the above query becomes:

select * from polygons
where type=city
and parent_id = (select id from polygons
                               where type = 'country'
                               and name = 'Italy');

Even with both structures optimised & indexed, the latter is likely to be much faster. No join is required. Given the country containing a city is a pretty static relationship, I suggest predefining to optimise query performance makes sense. 

If you store the heirarchies as predefined levels then a heirarchical search using the recursive "with" capability- see:
http://www.postgresql.org/docs/8.4/static/queries-with.html
is perhaps possible, to invoke searches up & down the tree. 

So use Postgis to determine the parent id using a spatial function, then store this as an indexed id.

HTH,

  Brent Wood


I'd say there are several approaches you could take to build a viable database, the optimal one is defined by your use case: the sorts of queries you want to apply.


--- On Tue, 4/10/12, Michal Kubenka <mkubenka at gmail.com> wrote:

From: Michal Kubenka <mkubenka at gmail.com>
Subject: Re: [postgis-users] How to design a database for continents, countries, regions, cities and POIs?
To: pcreso at pcreso.com
Cc: "PostGIS Users Discussion" <postgis-users at postgis.refractions.net>
Date: Tuesday, April 10, 2012, 9:59 AM

Actually what we need is some hierarchical base for relationship between countries, cities, regions, etc. Main goal of the application will be collecting data from many sources about specific cities, regions, countries and so on, and store it in database. Let's say we will have city Rome, we collect some info about this city into database from couple sources. And we need to know that Rome is in province Rome, sub-region Lazio in region Lazio, country Italy. So system should be flexible to allow create such relation from real world.

That's why I would choose two tables:
1) `polygons` - which can store countries, regions, sub-regions, provinces etc.2) `points` - which can store cities and POIs

Thanks.
Michal K. 
On Mon, Apr 9, 2012 at 8:11 PM,  <pcreso at pcreso.com> wrote:

Are you planning to store multiple versions of these polygons, for zoom layers?


Generally you need a high res version (eg: coastline) when zoomed in (large scale) and a lower resolution version when zoomed out (you can't see & don't need the detail.

This may or may not have an impact on your eventual data model, but it is worth ensuring you take this into account during the data modeling process. You can have a model where each feature has multiple geometry columns associated with it in the one table, or an approach which has the geometries in separate tables, using ID's to link to the aspatial attributes. The former is a simpler, monolithic solution, the latter is more complex but allows more use of tablespaces & underlying Postgres optimisation. 


You may also find you need to carry out joins (identify relationships between types
 of polygon, eg: cities within counties within states within countries, and this may perform better with a denormalised structure with separate tables for different categories of polygon. 

One example you might look at is the OSM data model. Not quite what you are describing, but a robust & well tested model for global roads & related spatial data, which does not use Postgis at all. 


http://booki.flossmanuals.net/openstreetmap/_draft/_v/1.0/the-osm-data-model/

--- On Mon, 4/9/12, mkubenka <mkubenka at gmail.com> wrote:


From: mkubenka <mkubenka at gmail.com>
Subject: [postgis-users] How to design a database for continents, countries, regions, cities and POIs?

To: postgis-users at postgis.refractions.net
Date: Monday, April 9, 2012, 11:31 PM

I'm brand new to GIS
 programming and I am designing a GIS application. Target
is to create system with continents, countries, regions (including states,
sub-regions, provinces), cities and places in cities. Each of this elements
will contain some text information and related stuff. As database we are

going to use PostgreSQL with PostGIS.

My question is how to design database for this system? I was thinking of 2
tables polygons and points, but I'm not sure if it's good way of thinking.

--
View this message in context: http://postgis.17.n6.nabble.com/How-to-design-a-database-for-continents-countries-regions-cities-and-POIs-tp4715669p4715669.html

Sent from the PostGIS - User mailing list archive at Nabble.com.
_______________________________________________
postgis-users mailing
 list
postgis-users at postgis.refractions.net
http://postgis.refractions.net/mailman/listinfo/postgis-users



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20120409/16e23f2c/attachment.html>