[Geowanking] RDF vs GML in Open Source vs Google

Chris Goad cg at mapbureau.com
Mon Jul 25 07:57:48 PDT 2005


I agree with Anselm  about the ways that the open source geocommunity can
continue to contribute in the new world that Google/Microsoft is bringing
us. In particular, the synthesis of diverse kinds of information tied to
geography ("Modelling" in Anselm's term) is full of possibility, involves
grassroots contribution of data, and is not subject to capture by any one
enterprise.

I do, however, want to respond to a remark that Anselm made in passing:


>  I  would argue that RDF2D has failed to do much so far.

I don't disagree, and my purpose is not to defend any particular RDF-based
approach to geo (eg RDFGeom2d) But I do believe that, regardless of the
current  level of adoption of any particular approach,  the general notion
of an RDF-based  alternative to GML has a lot to offer in the current
circumstances.

The fundamental problem with GML is its impermeability. It is not possible
to inject information about geographic features from another application
area (geology, crime statistics, desert turtle simulation ...) into the
innards of a GML file, and expect any of the tools to keep working.
Instead, one is supposed to define an "application schema"  , and this is a
daunting task for most.  The rules are intricate, and the result must be
formulated using the constructs of XML Schema - a specification widely
feared and loathed for its complexity (see
http://lists.xml.org/archives/xml-dev/200507/msg00057.html). In any case,
the barrier to entry for mixing GML with any other kind of information is
high. So, in practice, GML is mostly left as is, and a bag of nonstandard
integration tricks is used to glue it to the rest of the information flow.
(See http://www.mapbureau.com/gml/ for a longer version of this argument)

RDF is just the opposite. Mixing information from diverse vocabularies is
easy; in fact, that's what RDF is mostly for: a formalism for data
integration.

GML, impermeable and somewhat ponderous as it is,  seems a particularly bad
fit for the new geo-world that we are entering,  where mashups of diverse
applications are common, and, as Anselm says, velocity will be king.

The viability of a lighweight RDF approach is already demonstrated by the
success of the small w3c geo vocabulary for location. What is needed next is
a generally accepted extension of this idea to lines and areas.  RDFGeom2d
is my own proposal in this direction, but there are others. Anselm is right
that nothing in the way of widespread adoption or agreement has happened so
far. I wonder to myself, how can progress be made?  My own work in this area
takes place within my private Fabl universe, and is not particularly helpful
in speeding adoption. I understand that some of the work at @semantics
involves RDF for linear and area geography. Maybe if there were a few  open
source tools in widely used languages for parsing and displaying one or
another dialect of geoRDF, this would help.

Also, I'd like to mention for the benefit of those who don't spend their
days perusing OGC specs that the OGC and GML should by no means be
identified.  The most successful of the OGC specs, WMS, has nothing to do
with GML, and, although WFS specifies GML as the payload when serving
features, the WFS idea would certainly apply to other representations of
geography (although SPARQL running over a RDF geo store might prove
perfectly adequate).


-- Chris










More information about the Geowanking mailing list