Random Post: Linking Placename Authorities
RSS .92| RSS 2.0| ATOM 0.3
  • Home
  • About

    OpenStreetmap and Linked Geodata

    October 14th, 2010

    I’ve been travelling overmuch for the last six weeks, but met lots of lovely people. Most recently, during a trip this week to discuss the Open Knowledge Foundation‘s part in the LOD2 consortium project, had a long chat with Jens and Claus, the developers and academics behind Linked Geo Data, the Linked Data version of the OpenStreetmap data.

    linked geodata browser

    The most interesting bit for Unlock is the RESTful interface to search the data; by point, radius, and bounding box, by feature class and by contents of labels assembled from tags. So it looks like Opensearch Geo as much as Unlock’s place search api does.

    Claus made up a mapping between tags and clusters of tags in OpenStreetmap, to a simple linkedgeodata.org ontology. Here’s the mapping file – warning, it is quite large – OSM->linkedgeodata mapping rules. Pointed him at Jochen Topf’s new work on OSM tag analysis and clustering, Taginfo.

    As well as the REST interface, there is a basic GeoSPARQL endpoint using Virtuoso as a Linked Data store – we ran containment queries for polygons returning polygons with reasonable performance. There is a fracturing in the GeoSPARQL world both in proposed standards and in actual implementation.

    So we want to be able to return links to LinkedGeodata.org URLs in the results of our search. Right now Unlock’s place search returns original source identifiers (from geonames, etc) as well as our local identifiers, for place-names and shapes. In fact Unlock could help with the mapping across of Linkedgeodata.org URLs to geonames URLs, which are quite widely used, an entry point into the bigger Linked Data web.

    Another very interesting tool for making links between things on the Linked Data web is SILK, by Chris Bizer, Anja Jentsch and their research group at the Freie Universitat Berlin. The latest (or still testing?) release of SILK has some spatial inference capacity as well as structural inference. So we could try it out on, for example, the Chalice data just to see what kind of links can be made between URLs for linkedgeodata things and URLs for historic place-names.

    We’ve been setting up an instance of OpenStreetmap for Unlock and other purposes at EDINA recently. Our plan with this is to start working from Nominatim, which has a point-based gazetteer for place-names down to street address level, and attempt to extract and/or generalise shapes as well as points corresponding to the names. We’re doing this to provide more/richer data search, rather than republishing original datasets in some more/differently interpretable form. So there’s lots of common ground and I hope to find ways to work together in future to make sure we complement and don’t duplicate.

    OpenSearch Geospatial in progress

    March 15th, 2010

    One promising presentation I saw last week at the Jornadas SIG Libre – Oscar Fonts’ work in the Geographic Information Group at the Universitat Jaume I building on OpenSearch Geospatial interfaces to different services. OpenSearch geo query of OSM

    The demonstrator showed during the talk was an OpenLayers map display hooked up to various OpenSearch Geo services.

    Some are “native” OpenSearch services, like the GeoCommons data deposit and mapmaking service, the interfaces published by Terradue as part of the European GENESI-DR earth observation distributed data repository project.

    The UJI demo also includes an API adapter for sensationally popular web services with geographic contents. Through the portal one can search for tweets, geotagged Flickr photos, or individual shapes from OpenStreetmap.

    Oscar’s talk highlighted the problem of seeming incompatibility between the original draft of the OpenSearch Geospatial extensions, and the version making its way through the Open Geospatial Consortium’s Catalog working group as a “part document” included in the next Catalog Services for the Web specification.

    The issues currently breaking backwards-compatibility between the versions are these:

        geo:locationString became geo:name in the OGC draft version.
        geo:polygon was omitted from the OGC draft version, and replaced with geo:geometry which allows for complex geometries (including multi-polygons) to be passed through using Well Known Text.

    1) looks like syntactic sugar – geo:name is less typing, and reads better. geo:locationString can be deprecated but supported.

    2) geo:geometry was introduced into the spec as a result of work on the GENESI-DR project, which had a strong requirement to support multi-polygons (specifically, passes over the earth of a satellite, which crossed the dateline and thus were made up of two polygons meeting on either side of the dateline).

    geo:polygon has a much simpler syntax, just a list of (latitude, longitude) pairs which join up to make a shape. This also restricts queries to two dimensions.

    This seems to be the nub of the discussion – should geo:polygon be included in the updated version – risking it being seen as clashing with or superfluous to geo:geometry, leading to end user confusion?

    There is always a balance to be met between simplicity and complexity, Oscar pointed out in his talk what I have heard in OGC Catalog WG discussions too – that as soon as a use case becomes sufficiently complex, then CSW is available and likely fitter for the job. geo:geometry is already at the top end of acceptable complexity.

    It’s about a year since I helped turn Andrew Turner’s original draft into an OGC consumable form. Anecdotally it seems like a lot more people are interested in seeing what can be done with OpenSearch Geo now.

    The OGC version is not a fork. The wiki draft was turned into a draft OGC spec after talking with Andrew and Raj Singh about the proposed changes, partly on the OpenSearch Google Group. The geo:relation parameter was added on the basis of feedback from the GeoNetwork and GeoTools communities. There’s been a Draft 2 page, as yet unmodified, on the OpenSearch wiki since that time.

    In order to build the confidence of potential adopters, these backwards-incompatibilities do need to be addressed. Personal point of view would be to update the wiki draft, deprecating locationString and including both polygon and geometry parameters.

    I was impressed by the work of Oscar and collaborators, though wondering if they are going to move in to aggregation and indexing, search-engine-style, of the results, or just use the OpenSearch interface to search in realtime fairly fast moving sources of data. I wish I’d asked this question in the session, now. It all offers reinforcement and inspiration for putting OpenSearch Geo interfaces on services nearby – Go-Geo!, CKAN. The NERC Data Discovery Service could benefit, as could SCRAN. We’ll get to see what happens, which I’m glad of.