In 2008, IRD started to work on setting up a knowledge base (named Ecoscope) about Ecosystem Approach to Fisheries domain (EAF) in the context of a marine ecology laboratory studying Exploited Marine Ecosystems in different regions of the world. This application was meant to fit the needs of researchers by improving knowledge and related information resources management [14,12]. Among other goals, researchers expected an information system enabling to provide an inventory of available data sources (ecological observations, satellites images, pictures, articles, reports..) and facilitating data rescue, data access, data processes (indicators..) as well as the ability to summarize related knowledge through fact sheets about domain entities (ecosystems, species..) connected with hyperlinks (based on ecosystem relationships). Beyond metadata, data management and related interoperability issues (OGC, TDWG. . . ), this project was then a real opportunity to set up an ontology for EAF domain in order to link existing information resources with real-world entities (EAF domain concepts). To achieve these goals, semantic Web standards and reference RDF schemas have been taken into account (SKOS, Dublin Core, FOAF, OBOE, Darwin Core. . . ) and a first version of RDF schema for EAF domain has been set up. These ontologies have been instantiated to describe our information resources and some knowledge about entities that researchers are studying. A first Website has been set up on top of this knowledge Base. Related Web pages consist mainly in fact sheets about domain entities (ecosystems, top predators and related preys species, fishing vessels..persons) where users can find related information resources (spatial layers, articles, pictures, indicators..). Knowledge can as well be summarized through networks of entities like food webs with dedicated visualizations tools. This is made possible by querying the knowledge base where Linked metadata and data (in underlying databases) are tagged with related species URIs. Proof has been done that Semantic Web languages can be used to fit the needs of our colleagues. Moreover, in the context of iMarine FP7 project, we started to deal with partners having similar projects (FLOD from FAO, Worms, FORTH). We then set up a SPARQL endpoint and OpenSearch access to share the content our knowledge bases with other applications (search engines, text mining applications..). We will present our current application and related technical choices as well as futur plans to connect additional data sources to enrich this knowledge base and make it available for our partners. In particular we will describe some use cases related to biodiversity management issues.
In this paper, we present our work on knowledge management applied to the
domain of Ecosystem Approach to Fisheries (EAF). According to FAO [
We used this definition to create a conceptual model as schown in the UML diagram class of Figure 1.
This UML diagram has been used to create a first set of top-level classes
and properties for EAF domain (identifying real-world entities according to [
Similar objects have been created for hundreds of species, fishing vessels and fishing gears. . . which are all part of different marine ecosystems.
EAF requires thus the management of knowledge related to marine ecosystems and their abiotic, biotic and human components. However, this knowledge comes from different scientific studies driven by researchers which generate various information resources. These entities have to be described as well.
Real-world entities of our domain (like species, fishing vessels, habitats. . . ) are
related to different kinds of information resources (pictures, spreadsheets, databases,
satellites images, sensors data. . . )[
As many other laboratories working on ecological studies, the inventory of available data sources consists of various data types (which are subclasses of ”information resource” class in Figure 3): – ecological observations from
fieldwork (observers on-board fishing or scientific vessels collecting samples for data analysis: size, stomach content, isotopes, contaminants, fatty acids. . . ). These data are usually managed in spatial databases (Postgres / Postgis) or spreadsheets. – satellites images to characterize environmental parameters of species habitats. These data are usually managed with binary formats (e.g. netCDF files for series of images). – pictures that are collected by on-board observers or scientists, – articles, reports. . . published by researchers, – processes to run data analysis with different programming languages (R, IDL, Matlab. . . ).
All these resources are usually described and managed with specific (meta)data formats that impedes basic tasks like data discovery and retrieval. In addition to knowledge management, RDF is expected to facilitate data management (seamless access to metadata catalogues, codelist mapping. . . ) by complying with widely used schemas. 3
The description and the management of the information resources has to comply with well known standards to ensure that these resources will be made available for various communities of users. In particular, we target communities related to spatial, biodiversity and statistical information. In addition to information resources management, we selected as well existing standards to manage information about agents and domain entities (species. . . ). An effort has been required to transform these (meta)data in RDF with EAF domain URIs. 3.1
Schemas for information resources and related agents
Many standards enable the description of information resources. Most of them
consist in XML schema where keywords and mother metadata elements are
described with literals (for species, characteristics, fishing vessels, agents. . . that
are observed). This is the case with OGC metadata standards for spatial
information (19115, 19119, 19110 [
In order to describe and retrieve our information resources by using common
metadata elements and URIs we decided to comply with following RDF schemas:
– DCMI [
We aim to describe and make some of our data available as Linked Open Data
by taking into account the 5 star development scheme for Open Data [
However, being able to describe information resources, processes and related agents requires URIs for ecological entities described in Figure 1. 3.2
Standards for domain entities
Among existing RDF schema relevant for our domain, we can mention:
– Previous work on ontologies for ecoinformatics [
These ontologies have been taken into account to map our ontology classes and properties as well as for raw data triplication. 3.3
RDF generation We have two kinds of RDF triples which are generated from our information resources: – statements instantiating our ontology with real-world entities (ecosystems and related components): species, fishing vessel, ecosystems. . . Each data source provides a set of entities (species, environmental parameters. . . ) which are translated into instances of our ontology, – RDF description of information resources, including related agents. More than basic descriptions, our goal here is to tag metadata with URIs of domain entities (previous item).
For now, we have been using an ”ad hoc” approach to get RDF triples from each type of information resource as illustrated in Figure 4. Moreover in some cases, previous efforts can be reused: – OGC metadata (ISO 19139) can be transformed in GENESI-DEC RDF metadata by reusing an existing XSL file from GENESI-DEC project, – EML metadata can be transformed in OGC 19139 metadata with a GBIF
XSL file, – bibliographic references metadata can be exported in RDF (BIBO compliant) from Zotero (as well as references of pictures, videos if managed with Zotero).
In this case, the real challenge consists in adding some context to these RDF metadata by relating them to URIs of domain entities. This can be achieved by entity mining approach.
Once classes and properties for EAF entities have been created and related objects (including information resources and agents) instantiated, these objects have been loaded with JENA in a triple store (Jena with TDB, preferred to Postgres for persistent storage and access) and have been made accessible through a SPARQL endpoint (the ontology is available as well at this URL: http://www.ecoscope.org/ontologies/main).
Another access has been set up to deliver search results through OpenSearch protocol with different data formats: HTML, RSS and RDF (Semantic extension).
That was needed to enable a set of use cases and make the content our knowledge base available online. 5
In this section, we describe some use cases exploiting the content of the Ecoscope knowledge base. Our first use case was the setting up of a Web portal built on top of the ontology / knowledge base (through Jena for storage and access and Struts2 to set up Web pages on top of Jena). The goal was to satisfy basic use cases like data discovery and retrieval, knowledge summary about domain entities (species, fishing vessels. . . ). 5.1
Metadata catalogue One of the goal of using RDF metadata with URIs was to enable seamless access to different metadata catalogues without having to deal with specific standards. In particular OGC metadata (19115/39 used by INSPIRE), EML metadata (GBIF), Bibliographic references, Dublin Core metadata have been transformed to comply with a common set of metadata elements (cf section 3.1). Moreover these metadata are all anotated (and thus linked) with URIs of domain entities (cf section 3.2). This approach enables to query resources related to domain entities (e.g. yellowfin tuna) without having to restrict the search to specific standards, languages or terms. The search engine suggests all the results related to this entity and cluster the results according to their types Results can be related entities (e.g. preys of yellowfin tuna) or information resources like pictures, articles, databases, people. . . (see online application).
More than inventorying existing information resources, our prior goal was to summarize available knowledge by themes / domain entities. This has been done by setting up fact sheets. 5.2
Fact sheets The main purpose of our knowledge base was to feed the content of a Web portal by providing the knowledge about entities of interest for our laboratory (ecosystems, species, fishing vessels. . . ). The main goal was the creation of fact sheets about these entities. To achieve this goal, A SPARQL query harvest all the triples related to a given entities and Jena objects are used by Struts2 to build some HTML views. Figure 6 gives an example of Web page for yellowfin tuna.
The fact sheets cluster related entities by type of relationships (is predator of, is prey of, is exploited by) and cluster related information resources by data types (pictures, spatial data and related processes / indicators).
Other visualization interfaces are available to present RDF triples as
taxonomy or network.
This use case illustrates what can be achieved with previous ontologies when
applied to management of biodiversity data. The example of a food web is very
relevant as it shows relationships between entities (species) that are either
predators, preys or both. In Figure 7, we filtered RDF triples of the knowledge base
to represent the food web related to tropical tuna trophic data (using prefuse
API [
The Figure 6 gives an example of Web page which is available online with other visualization application (e.g. Taxonomy). Another use case consists in using the Ecoscope knowledge base for the mapping of codification systems (code lists). For example, in EAF domain, species are often identified either by FAO codes in fisheries datasets or, most of the time, by Worms codification systems in ecological observations. This is an issue when researchers need to run some processes which require cross analyses between fisheries and ecological datasets. Indeed, in this case, there is a need to enable mapping between codes to merge datasets. A first application has been developed to enable such mapping at data export. We aim to deliver similar services in a generic way (in a programmatic way or through GUIs) to enable mapping between code lists of different schemas. 6
For now, RDF triples to link our data with entities and agents of the domain have been created in various ways. To make the knowledge base sustainable in the long term, we can’t afford to feed it with ad hoc approaches. There is a need to harvest information from dedicated endpoints to facilitate updates by a workflow: – databases and netCDF files will be turn in RDF through a single data server, – pictures that are collected by on-board observers or scientists, – articles, reports. . . published by researchers will be exported in RDF from
Zotero Server, – processes to run data analysis with different programming languages (R, IDL, Matlab. . . ) will be described in RDF from OGC WPS metadata.
Another important improvement is related to the introduction of logical rules to infer some knowledge. A simple example consists in infering competition relationship between species from predation relationship. Indeed, two species are competing when they are predators of the same species (preys). This first use case is going to be implemented in the framework of iMarine FP7 program. 7
Our first application has demonstrated the interest of using ontologies and knowledge bases to satisfy different needs of researchers in our marine ecology laboratory: data discovery and retrieval, knowledge management and visualization (fact sheets, food web. . . ). Such an application is worth but our current ”ad hoc” approach still requires a lot of work to fill and update the database. To fix this issue, we aim to enable a RDF export from the Web portal which is used to access raw data (relationnal databases and netCDF files). We aim now to set up a workflow to facilitate RDF generation directly from the relational databases where ecological observations are stored and used by researchers to run scientific analysis. As a second step, we want ecological observations (raw data) to be made available as well in RDF and to be linked with existing RDF triples (summarizing underlying knowledge: for example a RDF triple stating a predation relationship between two species should be related to hundreds of observations / facts proving it). The next goal consists in enabling RDF export from our main data sources, as already done with other standards (OGC, TDWG, SDMX).