A plethora of Earth Observation data that is becoming available at no charge in Europe and the US recently re ects the strong push for more open Earth Observation data. Linked data is a paradigm which studies how one can make data available on the Web, and interconnect it with other data with the aim of making the value of the resulting \Web of data" greater than the sum of its parts. Open Earth Observation data that are currently made available by space agencies such as ESA and NASA are not following the linked data paradigm. Therefore, Earth Observation data and other kinds of geospatial data that are necessary for a user to satisfy her information needs can only be found in di erent data silos, where each silo may contain only part of the needed data. Publishing the content of these silos as RDF graphs, enables the development of data analytics applications with great environmental and nancial value. In this paper we present the tool GeoTriples that allows for the transformation of Earth Observation data and geospatial data into RDF graphs. GeoTriples goes beyond the state of the art by extending the R2RML mapping language to be able to deal with the speci cities of geospatial data. GeoTriples is a semi-automated tool that allows the publication of geospatial information into an RDF graph using the state of the art vocabularies like GeoSPARQL and stSPARQL, but at the same time it is not tightly coupled to a speci c vocabulary.
Lots of Earth Observation (EO) data has become available at no charge in Europe and the US recently and there is a strong push for more open EO data. Linked data is a paradigm which studies how one can make data available on the Web, and interconnect it with other data with the aim of making the value of the resulting \Web of data" greater than the sum of its parts. In the last few years,
This work has been funded in part by the FP7 project LEO (611141). linked geospatial data has received attention as researchers and practitioners have started tapping the wealth of geospatial information available on the Web. As a result, the linked open data (LOD) cloud has been rapidly populated with geospatial data. For example, Great Britain's national mapping agency, Ordnance Survey, has been the rst national mapping agency that has made various kinds of geospatial data from Great Britain available as open linked data.
Recently, the geospatial semantic web has also started to be populated with EO products (e.g., CORINE Land Cover and Urban Atlas published by project TELEIOS3). In general, open EO data that are currently made available by space agencies such as ESA and NASA are not following the linked data paradigm. Therefore, EO data and other kinds of geospatial data that are necessary for a user to satisfy her information needs can only be found in di erent data silos, where each silo may contain only part of the needed data. Publishing the content of these silos as RDF graphs, enables the development of data analytics applications with great environmental and nancial value. With the recent emphasis on open government data in many countries, the development of useful Web applications utilizing EO data and geospatial data in general is just a few SPARQL queries away.
Geospatial data in general and EO data in particular, can come in vector or raster form and are usually accompanied by metadata. Vector data, available in formats such as ESRI shape les, KML, and GeoJSON documents, can be accessed either directly or via Web Services such as the OGC Web Feature Service or the query language of a geospatial DBMS. Raster data, available in formats such as GeoTIFF, Network Common Data Form (netCDF), Hierarchical Data Format (HDF), can be accessed either directly or via Web Services such as the OGC Web Coverage Processing Service (WCS) or the query language of an array DBMS, e.g., the array-query language SciQL4. Metadata about EO data are encoded in various formats ranging from custom XML schemas to domain speci c standards like the OGC GML Application schema for EO products and the OGC Metadata Pro le of Observations and Measurements.
Automating the process of publishing linked geospatial data has not been
addressed yet. For example, in the wild re monitoring and management application
that we developed in TELEIOS, custom Python scripts were used for publishing
all necessary data as linked data. For this reason, we designed and implemented
the tool GeoTriples in the context of the EU FP7 project LEO5. In this paper
we present the tool GeoTriples that allows for the transformation of EO data
and geospatial data in various formats, such as data stored in spatially-enabled
relational databases and raw les, into RDF graphs. GeoTriples goes beyond
the state of the art by extending the R2RML mapping language to be able to
deal with the speci cities of geospatial data. GeoTriples is a semi-automated
tool that allows for the publication of geospatial information into an RDF graph
using the state of the art vocabularies like GeoSPARQL [
The organization of the paper is as follows. Section 2 discusses related work. In Section 3 we present the architecture of GeoTriples and our extensions to the R2RML language for the geospatial domain. We discuss how GeoTriples generates automatically R2RML mappings and how they are subsequently proccessed for transforming a geospatial data source into an RDF graph. In Section 4 we present an example where we demonstrate how GeoTriples is currently being used for realizing the precision farming application that is being developed in LEO. Finally, in Section 5 we concludes our work. 2
In this section we will present related work on methodologies and tools targeting
the transformation of existing data sources into RDF graphs. Much work has
been done recently on mapping relational data into RDF graphs. An extensive
discussion on mapping relational data into RDF can be found in [
Direct Mapping of Relational Data to RDF. A straightforward
mechanism for mapping relational data to the RDF data model is the Direct Mapping
of Relational Data to RDF 6 approach that became a W3C recommendation in
2012. According to this approach, relational tables are mapped to classes
dened by an RDF vocabulary, while the attributes of each table are mapped to
RDF properties that represent the relation between subject and object resources.
Identi ers, class names, properties, and instances are generated automatically
following the respective labels of the input data. For example, given the table
Address, the class <Address> will be generated, and all tuples will be instances
of this class. According to this approach, the generation of RDF data is dictated
by the schema of the relational database. This mechanism was initially de ned
in [
The Mapping Language R2RML. A language for expressing customized mappings from relational databases to RDF graphs is the R2RML mapping language7 that became W3C recommendation in 2012. R2RML mappings provide the user with the ability to transform existing relational data into the RDF data model, following a structure and a target vocabulary that is chosen by the user. 6http://www.w3.org/TR/2012/REC-rdb-direct-mapping-20120927/ 7http://www.w3.org/TR/r2rml/ R2RML mappings refer to logical tables to retrieve data from an input database. A logical table can be a relational table that is explicitly stored in the database, an SQL view or a valid SQL select query. A triples map is de ned for each logical table that will be exported into RDF. A triples map is a rule that de nes how each tuple of the logical table will be mapped to a set of RDF triples. A triples map consists of a subject map and one or more predicate-object maps. A subject map is a rule that de nes how to generate the URI that will be the subject of each generated RDF triple. Usually, the primary key of the relation is used for this purpose. A predicate-object map consists of predicate maps and object maps. A predicate map de nes the RDF property to be used to relate the subject and the object of the generated triple. An object map de nes how to generate the object of the triple, the value of which originates from the value of the attribute of the speci ed logical table.
R2RML processors are applications that take as input a relational database and R2RML mappings and produce RDF graphs. Some R2RML processors provide a virtual SPARQL endpoint that transparently translate SPARQL queries to SQL queries by taking into account the given R2RML mappings. Alternatively, an R2RML processor can choose to export all relational data as an RDF dump according to the given mappings and then o er SPARQL or a linked data interface over the produced data.
Let us now present some tools that are able to translate raw data sources
into RDF following the direct mapping approach, R2RML or a custom mapping
language. OpenLink Virtuoso [
However, little attention has been paid to the problem of publishing geospatial information in the Semantic Web. Most datasets that contain such information are either generated manually or by semi-automated processes.
LinkedGeoData10 project focuses on publishing OpenStreetMap11 data as linked data. Sparqlify12, developed in the context of this project, is employed for this task, using a proprietary mapping language. It creates mappings of spatial datatypes into RDF but until now, it does not discuss on how to deal with non-relational data.
8https://github.com/jpcik/morph 9http://capsenta.com/ultrawrap/ 10http://linkedgeodata.org/ 11http://www.openstreetmap.org/ 12http://sparqlify.org/
The tool Geometry2RDF13 was the rst tool that allows for the conversion of geospatial information that resides in a spatially-enabled relational database into an RDF graph. It takes as input data stored in a spatially enabled DBMS like Oracle Spatial or MySQL, and utilizes the libraries Jena and GeoTools to produce an RDF graph. Geometry2RDF follows the direct mapping approach, and allows the user to con gure the properties that connect a URI to the serialization of a geometry and allows for the conversion of the coordinates to the desired coordinate reference system. Geometry2RDF follows the direct mapping approach that is not expressive enough to deal with the speci cities of the geospatial domain. For example, since this work has preceded the proposal of GeoSPARQL and stSPARQL, it does not produce RDF graphs according to these vocabularies. For this purpose, the tool TripleGeo14 was recently developed. TripleGeo is based on Geometry2RDF and allows the generation of RDF graphs that follow the GeoSPARQL vocabulary. However, this tool has the same shortcomings with Geometry2RDF since the mapping process of the input data into an RDF graph that follows the GeoSPARQL vocabulary is hard-coded in the implementation. Thus, signi cant e ort is required for modifying such tools in order to support other vocabularies.
A di erent approach was followed by [
In this section we will present in detail the tool GeoTriples that we developed for transforming geospatial data sources into RDF. GeoTriples15 is an open-source tool that is distributed freely according to the Mozilla Public License v2.0. In this section we will present the architecture of GeoTriples and we will discuss our implementation choices. We will describe in detail how GeoTriples generates automatically R2RML mappings for publishing data that reside in spatiallyenabled databases and raw les (e.g., ESRI shape les), and how it can process such mappings for producing an RDF graph that follows the GeoSPARQL or any other vocabulary.
13http://mayor2.dia.fi.upm.es/oeg-upm/index.php/en/technologies/ 151-geometry2rdf 14https://github.com/GeoKnow/TripleGeo 15http://sourceforge.net/projects/geotriples/
Integrated
Access and Processing SQL
SciQL stSPARQL GeoSPARQL
Linked Open Data
Cloud Raw Data
Web Services Shape Files KML The abstract architecture of GeoTriples, from a data perspective is shown in Figure 1. GeoTriples takes as input data that are stored in a spatially-enabled database, data that reside in raw les (e.g., ESRI shape les), or the results that derive from processing the aforementioned data (e.g., the result of a SciQL query over raster or array data). Regarding the system architecture at a lower level, GeoTriples uses a connector for each type of input data in order to transparently access and process the input data. As Figure 2 depicts, GeoTriples comprises two main components: the mapping generator and the R2RML processor. The mapping generator takes as input a data source and creates automatically an R2RML mapping that transforms it into an RDF graph. The generated mapping is enriched with subject and predicate-object maps, in order to take into account the speci ties of geospatial data and cater for all transformations that are needed to produce an RDF graph that is compliant with the GeoSPARQL vocabulary. To accomplish this task, we extend R2RML mappings to allow the representation of a transformation function over input data. Afterwards, the user may edit the generated R2RML mapping document to comply with her requirements (e.g., use a di erent vocabulary).
Implementation Choices. One of the main choices we had to make during the
design of GeoTriples was to choose which RDB2RDF framework to extend. We
chose to extend the D2RQ platform which seemed to be the most mature system
for publishing relational data into RDF. It provides a mechanism for generating
and processing R2RML mappings for a variety of relational databases that are
accessible via JDBC, like MonetDB, PostgreSQL and Oracle. D2RQ provides
preliminary support for translating SPARQL queries into SQL queries given
some R2RML mappings. However, given the recent results of the system Morph
[
KML tr o c e n n o C
Mapping Generator D2RQ Engine
R2RML Processor D2RQ Engine
R2RML Mapping Document
Transforming Geospatial Data into RDF graphs using
GeoTriples In this section we will present the main functionality of GeoTriples. First we will show the extended R2RML mappings that are produced automatically by GeoTriples that take into account the spatial dimension of the input data, and then we will show how these mappings are processed subsequently by GeoTriples in order to generate an RDF graph.
Automatic Generation of R2RML Mappings. Much work has been done
recently on extending RDF to represent and query geospatial information. The
most mature results of this work are the data model stRDF and the query
language stSPARQL [
Given a spatially-enabled database or a raw le that contains geometric information, GeoTriples generates an R2RML mapping document. Let us take for example the Natura 2000 dataset of Germany that contains information about protected areas in Germany and is distributed in the form of an ESRI shape le. More details on this dataset can be found in Section 4. Conceptually, each geometric object stored in the ESRI shape le should be an instance of the class geo:Geometry, and all non-geometric attributes that characterize each geometry are thematic attributes of the corresponding feature. Following this modeling approach, we generate an instance of the class geo:Feature and an instance of the class geo:Geometry for each geometric object stored in the ESRI shape le. Geometric and non-geometric attributes that appear at the ESRI shape le are assigned accordingly to the appropriate instances. Part of the R2RML mapping that is generated automatically by GeoTriples to represent the features stored in the Natura 2000 ESRI shape le is the following: _:natura rr:logicalTable [ rr:tableName "`natura`"; ]; rr:subjectMap [ rr:class geo:Feature; rr:template "http://data.example.com/natura/Feature/id/{`gid`}"; ]; rr:predicateObjectMap [ rr:predicate nato:has_SITECODE; rr:objectMap [ rr:datatype xsd:string;
rr:column "`SITECODE`"; ]; ]; rr:predicateObjectMap [ rr:predicate geo:hasGeometry ; rr:objectMap [ rr:parentTriplesMap _:natura_geometry; rr:joinCondition [ rr:child "gid"; rr:parent "gid"; ]; ]; ].
This mapping document contains a triples map that describes how instances of the class geo:Feature are constructed. All thematic information that is stored in the ESRI shape le is assigned to the generated feature and a link between the feature and its geometry is also generated. However, the notion of a primary key is not de ned for ESRI shape les. Each ESRI shape le though is accompanied by a relational table in dBASE format that stores infromation about the geometries, so we de ne as a unique identi er for each geometric object the respective row identi er in the dBASE table. In the example above we represent this information as an extra attribute with name 'gid'.
Part of the R2RML mapping that is generated automatically by GeoTriples to represent the geometries stored in the Natura 2000 ESRI shape le is the following: _:naturaGeometry rr:logicalTable [ rr:tableName "`natura`"; ]; rr:subjectMap [ rr:class geo:Geometry; rr:template "http://data.example.com/natura/Geometry/id/{`gid`}"; ]; rr:predicateObjectMap [ rr:predicate geo:dimension; rr:objectMap [ rrx:transformation [ rrx:function geof:dimension; rrx:argumentMap (
[rr:column "`Geom`"] ); ] ]; ].
This mapping document contains a triples map that describes how instances of the class geo:Geometry are constructed. All geometric information that is stored in the ESRI shape le is assigned to the generated geometry. In addition, object maps de ne that geospatial functions like geof:dimension are applied to the serialization of each geometric object in order to produce the values that will appear at the object part of the corresponding triple.
Notice that in the above example we extended the de nition of an object map by allowing it to be the RDF term obtained by applying a transformation on the source data. Each transformation de nes the SPARQL built-in function or the SPARQL extension function to be invoked, using as an argument the sequence of RDF terms that are produced by the respective term maps.
Processing of R2RML mappings for producing RDF graphs. R2RML mappings usually consist of two triples maps; one for handling thematic information and one for geospatial information. The triples map that handles thematic information de nes a logical table that contains the thematic attributes and a unique identi er for the generated instances. The latter could be either the primary key of the table in case the input data is a relational database or a row number in the dBASE table of an ESRI shape le. Combined with a URI template, the unique identi er is used to produce the URI that serve as subjects of the produced triples. The predicate object maps are also processed accordingly in order to de ne RDF properties the value of which originate from the value of the column of the thematic logical table.
The triples map that handles the geospatial information of the input data source, de nes a logical table with unique identi er similar to the thematic one. However, according to the type of the data source, the de nition of this logical table may vary. For instance, in the case of a relational database, it is de ned by providing an appropriate SQL query that uses spatial functions provided by spatially enabled relational backend (e.g. utilize the function ST_Dimension). If the input source is an ESRI shape le, then GeoTriples will perform such transformations on the y by evaluating the SPARQL extension function using the JTS Topology Suite. 4
Linked geospatial data in the precision farming application of LEO Let us now give an example that demonstrates how GeoTriples is being used in LEO for the development of a precision farming application. The aim is to develop a precision farming application that combines traditional geospatial data with linked geospatial data for enhancing the quality of precision farming activities. Precision farming is a concept based on observing the heterogeneity within an agricultural eld and providing information about cultivating each part of the eld according to the characteristics of the soil. Precision farming aims to solve numerous problems for farmers such as the minimization of the environmental pollution by fertilizers. For dealing with this issue, the farmers have to comply with many legal and technical guidelines that require the combination of information that resides in diverse information sources. In this section we present how linked geospatial data can form the knowledge base for providing solutions for this problem. We published the following datasets as RDF graphs using GeoTriples in order to use them in the precision farming application.
Talking Fields16 is a precision farming project aiming to increase the e ciency of agricultural production via precision farming by means of geo-information services integrating space and ground-based assets. Currently, TalkingFields produces products for improved soil probing using satellite-based zone maps, and provide services for monitoring crop development through provision of biomass maps and yield estimates.
Natura 200017 is an ecological network designated under the Birds Directive and the Habitats Directive which form the cornerstone of Europe's nature conservation policy. National authorities submit a standard data form that describes each site and its ecology in order to be characterized as a Natura site.
OpenStreetMaps (OSM)18 is a collaborative project for publishing free maps of the world. OSM maintains a community-driven global editable map that gathers map data in a crowdsourcing fashion. LinkedGeoData (LGD)19, is a project focused on publishing OSM data as linked data.
LGD data is the only dataset that is already published as linked geospatial
data. For the rest datasets, we design an appropriate ontology that follows the
GeoSPARQL vocabulary. Afterwards, we use GeoTriples in order to produce
the R2RML mappings that dictate the process of generating the desired RDF
output. Finally, we use the R2RML processor of GeoTriples for translating the
input data into RDF graphs. The produced data are then stored into an
appropriate geospatial RDF store. We chose to utilize our own geospatial RDF store
Strabon20 [
Let us now see an example query that can provide a precision farming application with information about the parts of agricultural elds that are either close to a river or are located within a protected area. This information allows the precision farming application to take into account legal restrictions regarding distance requirements when preparing the prescription maps that the farmers will utilize afterwards.
16http://www.talkingfields.de/ 17http://ec.europa.eu/environment/nature/natura2000/access_data/index_ en.htm 18http://www.openstreetmap.org 19http://linkedgeodata.org 20http://www.strabon.di.uoa.gr/
Example. Select all parts of agricultural elds that are close to a river or are located inside a protected area.
SELECT distinct ?field_name ?river_name ?cell_wkt WHERE {{?river rdf:type osmo:River ; osmo:hasName ?river_name ; geo:hasGeometry ?river_geo . ?river_geo geo:asWKT ?river_wkt . ?field rdf:type tf:Field ; tfo:hasFieldName ?field_name ; tf:hasRasterCell ?cell . ?cell geo:hasGeometry ?cell_geo ; ?cell_geo geo:asWKT ?cell_wkt .
FILTER(geof:distance(?river_wkt ,
?cell_wkt, uom:meter) < 100) . } UNION { ?field rdf:type tf:Field ; tfo:hasFieldName ?field_name ; tf:hasRasterCell ?cell . ?cell geo:hasGeometry ?cell_geo ; ?cell_geo geo:asWKT ?cell_wkt . ?nat rdf:type nat:NaturaArea ;
geo:hasGeometry ?nat_geo . ?nat_geo geo:asWKT ?nat_wkt .
FILTER(geof:contains(?nat_wkt, ?cell_wkt)}}
This above query uses the geo:distance function whose arguments are the serializations of two geometries and a URI that identi es a unit of measurement. The result of this metric function is the minimum distance between these two geometric objects. This lter identi es the parts of a eld that are close to a river, so the precision farming application should adjust the fertilizer/pesticide usage accordingly. The second part of the query identi es the parts of a eld that resides inside a protected, thus the precision farming application should adjust the fertilizer/pesticide usage accordingly. Figure 3 depicts protected areas, rivers, agricultural elds and the parts of the agricultural elds that are close to a river. 5
In this paper we presented the tool GeoTriples that transforms geospatial data that reside in a spatially-enabled DBMS or raw les into RDF graphs, by extending the R2RML mapping language. This allows GeoTriples to transform geospatial data into RDF graphs without being tightly coupled to a speci c vocabulary. We plan to extend GeoTriples to support numerous types of vector data formats such as KML, as well as raster formats like GeoTIFF and netCDF.