Databases Reference
In-Depth Information
without making their intended meaning explicit. Suppose further that you would
like to query for
towns near forests
such as in the SPARQL query shown below,
and you plan to use the retrieved towns to conduct your analysis.
2
[...]
SELECT distinct ?town ?forest
WHERE {
?town
geo-pos:lat ?lat ;
geo-pos:long ?long ;
a dbp-ont:Town .
?forest
omgeo:nearby(?lat ?long "25mi") ;
a dbp-ont:Forest .
}[...]
No matter what the query will return and how you will process and analyze
the data from those thousands of towns, your results will be misleading at best.
Most likely you will have overlooked that among all those small populated places,
your dataset will also contain Los Angeles, Stuttgart,
3
and other metropolises.
The reason for this apparently odd result is that the class
city
and
town
are
defined to be equivalent by Californian law. In fact, most of US states have their
own legal definition of these terms. While some rely on maximum population
as a criterion, others do not [47]. The situation for forests is even more com-
plicated. Lund [66], for instance, lists over 900 different definitions for
forest
,
afforestation
, and related terms. These definitions are not without consequences
but often legally binding. In the past, loop holes in these definitions have been
used for massive deforestation.
4
Finally, the alert reader may be wondering why
a radius of 25 miles is used in the example above to define
nearby
. First, as with
many other terms, the semantics of nearby is context-dependent. Second, unfor-
tunately, most of today's Linked Data represents geographic features by their
centroids (geometric center points) instead of polygons. Thus, even if a GIS
would represent a particular town and forest by two adjacent polygons, their
centroids may still be miles apart; see [8] for more details on spatial queries over
Linked Data.
As these examples show, understanding what the authors of a scientific study
or data providers in general mean by apparently obvious terms is a di
cult task.
Without better geo-ontologies, semantically annotated (meta) data, and more
complex ontology alignment services that can map between local ontologies,
2
This query will fail as the class
Forest
(or similar classes) are not defined in DBpedia.
However, querying for Mountains, for instance, would work.
3
Stuttgart is described as the 'sixth-largest city in Germany' in DBpedia but
classifieds as a town via
dbpedia:Stuttgart rdf:type dbpedia-owl:Town
;see
4
Readers interested in deforestation and in combining SPARQL with spatial statistics
in R may want to check the new Linked Brazilian Amazon Rainforest Dataset[52].
