Observational/hydrographic data of the South Atlantic Ocean published as LOD

1.Introduction

In the ecology domain, research teams collect and store biological and environmental information over the years/decades in database systems to answer their own queries. However, this information is isolated from other datasets for interoperating with and, in addition, is not ready to be accessed by machines. Particularly, in marine science the data collection is a process of cumulative logistic complexity, which makes it important to work on the curation and sustainability of the database, in both the short and long term. It is of great benefit for scientific institutions to publish their datasets following the Linked Data principles [7] not only for interlinking and easy cross-referencing but also for other purposes that are not foreseen at the moment of publication. The state of the art in the last decade shows that together with technology to collect data, semantic interoperability has further grown in importance [20]. To meet Linked Data requirements, datasets must be described with rich metadata such as controlled vocabularies in a particular form – RDF – and published as a findable resource with a unique identifier.

This paper integrates observational and hydrographic datasets based on the SOSA/SSN ontology [6,10] and BiGe-Onto ontology11 [23]. As far as we know, this work is the first to publish linked open data occurrences of a species in two geographical environments (coastal and marine) collected over two decades. Data comes from a research program focused on Southern Elephant Seals (SES) in Patagonia Argentina “Temporal and spatial distribution of the southern elephant seal colony in Península Valdés, Argentina” [15]. The program started in 1990 to study ecology and life history strategies of SES, together with the research of foraging areas and dive behavior, and to contribute to understanding the effects on the species from changes in the ecosystem of SW Atlantic Ocean. The research site is in Península Valdés (PV), which has been a UNESCO World Natural Heritage since 1999 [21].

During the annual cycle, the SES come ashore to breed and molt. The rest of the year they are at sea, traveling long distances throughout its extensive migration (up to 8 months and 12000 km. of round trip), and diving continuously to a depth of 1500 meters or more. During their terrestrial phase they frequently revisit previous years’ sites [13,14]. The behavior during the marine phase shows that SES are ideal carriers of devices, providing physical profiles, (i.e. hydrographic of the water column). For tracking the SES at sea, researchers make use of miniaturized animal-attached tags for relaying data, known as biologging domain [19], and cover animal migration and oceanographic measurements [1]. The instruments deployed on the seal return, at a low cost, large volumes of hydrographic data in regions never studied directly by buoys or oceanographic vessels and collecting large amounts of information associated with the key habitats in the South Atlantic Ocean.

This paper is organized as follows: Section 2 describes the SES database. Section 3 briefly presents the network of ontologies. Section 4 shows examples of how the data are organized. Section 5, describes the populating processes and the links to other datasets. Section 6 shows the application developed to access and analyze the dataset. Finally, we conclude by presenting an analysis of our work and perspectives.

2.SES database

Data are recorded from measurements of physical variables and locations obtained in two different stages. First stage involves an annual census, which takes place during the breeding season of SES. The second stage starts at the end of the breeding season, when SES go back at sea for foraging purposes. Below we briefly describe how data are generated and recorded in each stage. During the breeding season, the SES haul out to the beach to breed. Annual census on foot along the coast of the colony is an arduous but indispensable work to know distribution and trend of the population. The objective is to count each of the harems scattered on the beaches of the PV to determine the number of offspring born in a season. Counts carried out during 2–3 days at peak of the breeding season (October 3–7), when most of the population is ashore. All the breeding groups were counted and located along 200 km of coastline, divided into sections and each census taker is assigned to a route. The census taker must count for the number of animals and classify them by sex and age males, females, and pups. Hereinafter, we will call the procedure of counting individuals in a certain place Occurrence. Each occurrence was georeferenced (latitude and longitude) and demographic data included date and time, group size and substrate where the SES is located. All information about these censuses is recorded in a field book and then uploaded into a MySQL database. Table 1 summarizes the most relevant fields for the conducted censuses.

At the end of breeding, SES go back at sea for foraging. The trip is monitored by small computers designed and built by Wildlife Computers Inc.22 with sensors to take measurements about their location and immediate environment. The instrument is deployed when the seal is on land before the migration into the sea begins. Time Depth Recorder (TDR) records, time, depth, and temperature every 30 minutes during round foraging trip. The position is also registered when the seal ascends to the surface. Table 2 summarizes the fields that are most relevant in diving.

The census and the deployments of the instruments are carried out by the research team belonging to Centre for the Study of Marine Systems hosted in Puerto Madryn, Patagonia Argentina (CESIMAR-CENPAT-CONICET).33 The institute is engaged in oceanographic and marine research activities, monitoring information management and data acquisition activities on different platforms.

3.Ontologies used to model observations and hydrographic profiles

In this section, we briefly summarize these ontologies used for the publication of our dataset, indicating the reuse of concepts.

The core of our ontologies network is composed by SOSA/SSN [6] and BiGe-Onto [23], which can be jointly used for both hydrographic profiles and observational data. These ontologies are linked to other ones describing different sub-domains, and thus creating such network. Therefore, the resulting network is composed mainly by the following:

A list of prefixes and their corresponding URIs are listed in Table 3.

Semantic Sensor Network (SSN) ontology The Semantic Sensor Network (SSN) is a generic ontology related to sensor observations. This ontology has been updated to become a W3C recommendation, and currently it is a lightweight one dedicated to sensor and actuator descriptions. It has been called Sensor, Observation, Sample, and Actuator (SOSA) pattern. The link between SSN and SOSA is described in [6]. The classes we have reused from SOSA/SSN ontology are:

BiGe-Onto ontology BiGe-Onto is an ontology designed for modeling biodiversity and marine biogeography data [23]. Its main concept is an occurrence. Given that each census is a set of occurrences of SES at a specific time and place, we consider BiGe-Onto fits the nature of our data. At the same time, it reuses different vocabularies such as Darwin Core (DwC) [22], which is the core one in BiGe-Onto. Main reused classes are dwc:Occurrence, dwc:Taxon, dwc:Organism, and dwc:Event. Moreover, BiGe-Onto reuses foaf:Person void:Dataset and dc:Location, among others.

Since BiGe-Onto mainly describes occurrences, which depend on other concepts to exist, we also outline below some of the most important properties defined for relating such occurrences:

The Quantity, Unit, Dimension and Type (QUDT) ontologies QUDT is a collection of OWL ontologies and vocabularies [9]. The QUDT schema defines the base classes, properties, and restrictions used for modeling physical quantities, units of measure, and their dimensions in various measurement systems. QUDT also contains a set of vocabularies to define units for different domains. We have reused the unit vocabulary that categorizes units in different classes. This vocabulary also provides individuals to identify units such as qudt:M for meter and qudt:DEG_C for Celsius degree.

GeoSPARQL ontology GeoSPARQL [18] is an Open Geospatial Consortium (OGC) standard for supporting the representation and querying of geospatial data on the Semantic Web. As such, it is based on the OGC’s Simple Features model, with some adaptations for RDF. GeoSPARQL designates a vocabulary for representing geospatial data in RDF, and it defines an extension to the SPARQL query language55 for processing them, together with both a small ontology for representing features66 and geometries,77 and a number of SPARQL query predicates and functions. All these definitions are derived from other OGC standards so that they are well grounded and documented. Using the new standard should ensure two things: (1) if a data provider uses the spatial ontology in combination with an ontology of their domain, these data can be properly indexed and queried in spatial RDF stores; and (2) RDF-compliant triple stores should be able to properly process the majority of spatial RDF data. This ontology is used to describe the location of each occurrence, and the beaches involved. We reuse the classes geo:Feature and geo:Geometry, and the associated properties, like geo:hasGeometry and geo:asWKT.

The W3C time ontology The W3C Time ontology [2] enables the description of time instants and intervals. Hence it may be useful when we need to describe the timestamp, or the time associated with the measurements made by the observers of the SES. We reuse the classes time:Interval and time:Instant, and their associated properties, such as time:hasBeginning, time:hasEnd, and time:inXSDDateTimeStamp.

FRBR-aligned Bibliographic Ontology (FaBiO) The aim of FaBiO [16] is recording and publishing descriptions of entities that are published or potentially publisable, and that contain or are referred to by bibligraphic references. Its classes are structured according to the FRBR schema of Works, Expressions, Manifestations and Items. Additional properties have been added to extends the FRBR data model by linking works and manifestations. Considering that both census observations and measurements of physical variables have been used to publish results in diverse scientific journals, we have chosen the ontology FaBIO for modeling the relationships between platforms and such publications. Our choice is based on the fact that FaBIO is one of the ontologies involved in the OpenCitations88 initiative, which promotes the use of open bibliographic data by the use of Semantic Web technologies. We reuse several classes to identify the kind of published document, for example fabio:JournalArticle, fabio:Book, fabio:BookChapter and fabio:Dataset. We also reuse some of their associated properties, like prism:doi and prism:publicationDate.

NERC and GoodRelations ontologies Additionally, we reuse the vocabulary for Oceanography known as Natural Environment Research Council (NERC) Vocabulary Server [12] developed by the British Oceanographic Data Center (BODC),99 which provides access to lists of standardized terms that cover a broad spectrum of disciplines relevant to Oceanography. In particular, we use the category P01 whose terms are used to describe individual measured phenomena and S10 category to describe the sex of biological entities. We also reuse the GoodRelations ontology [8] to model sensor brands, serial numbers and models.

It is important to highlight that the previous ontologies were chosen because many of them are W3C standards (such as SOSA/SSN and OWL Time) or standards of the Open Geospatial Consortium (OGC) such as GeoSPARQL and vocabularies that are widely used by the user community involved in the domain. (e.g. Darwin Core or NERC). In the absence of a standard ontology for a specific domain, we decided to use those that according to our knowledge, are the ones that are currently under development, provide support and the documentation is adequate, such as FaBIO and GoodRelations.

4.Data model and URIs

Based on the network of ontologies described in the previous section, we are now able to create a dataset containing all the individuals describing hydrographic profiles and occurrences taken during the censuses. Now we explain the decisions taken to create resource URIs and we provide examples of resource descriptions.

4.1.Resource URIs for hydrographic profiles

This subsection presents the main URI design decisions and conventions used. Table 4 provides a summary of the main types of URIs that we generate. The first column presents the type of resources. The second column indicates the associated class which types the resources. The last column contains the name pattern used to generate the resource URIs. The base URI for our dataset is http://linkeddata.cenpat-conicet.gob.ar/resource/. Its prefix is base. Our generic name pattern to produce URIs for each object is baseURI +“/”+ nameOfClass +“/”+ objectIdentifier.

Table 4

URI generation templates for resources. First part of the table describes the patterns of URIs related to hydrographic profiles, while the second describes those related to occurrence data. Longitud and latitude numbers are expressed in decimal

Object	Class	URI pattern
SES Platform	sosa:Platform	base:platform/{sesID}
TDR	sosa:Sensor	base:sensor/{Sensor type}/id-{serialNumber}
Feature of Interest	sosa:FeatureOfInterest	base:featureOfInterest/sampleOfWaterColum/id-{DiveID}
Observable property	sosa:ObservableProperty	base:observableProperty/id-{DiveID}/avg_temp
		| avg_depth | location
Instant	time:Instant	base:instant/{beginDive}T{beginTime}
Duration	time:Duration	base:duration/sampleID-{DiveID}
Interval	time:Interval	base:interval/sampleID-{DiveID}
Observation	sosa:Observation	base:observation/id-{DiveID}/avg_temp | avg_depth
		| location
Result value	sosa:Result	base:result//id-{DiveID}/avg_temp | avg_depth
		| location
Dive location	geo:Geometry	base:geometry/point_{logitude}_{latitude}
Census observation	dwc:Occurrence	base:occurrence/ID-{ObservationID}
Observation event	bigeonto:BioEvent	base:bioevent/ID-{ObservationID}
Location	dc:Location	base:geometry/{geometryType}_{longitude}_{latitude}
Observer	foaf:Person	base:person/{personID}
SES category	vocab:Category	base:category/{categoryCode}
Taxon	dwc:Taxon	base:taxon/{mirounga_leonina}
Related paper	fabio:Expression	base:paper/ID-{paperID}

Fig. 1.

SES platform and TDR sensor description.

4.1.1.Platform and sensor

We consider SES as an oceanographic sampling platform. The individual that represents the platform is an instance of the sosa:Platform class. Each sensor hosted by the SES is represented by an instance of the class sosa:Sensor. Figure 1 presents the description of the TDR sensor. The TDR is identified by an URI generated using the sensor type, e.g. TDR plus the manufacturing number 19793. This URI is typed by the class sosa:Sensor. The sosa:host property links the sosa:Platform instance to the TDR URI. The sosa:observes property links the TDR URI to an instance of sosa:ObservableProperty; in this case it corresponds to depth. Additionally, the gr:ProductOrServiceModel class is used to model the sensor, and the gr:hasBrand property relates the sensor to its manufacturer, which is an instance of the gr:Brand class. Other properties such as gr:serialNumber and gr:category were reused to describe the type of sensor and its serial number. The link http://linkeddata.cenpat-conicet.gob.ar/resource/sensor/TDR/id-19793 can be used to explore visually the sensor URI.

4.1.2.Observation

An observation describes the context of a measurement made by a sensor. In the case of TDRs, the measurements are location, time, depth and temperature. Properties sosa:hasFeatureOfInterest, sosa:hasResult, sosa:observedProperty and sosa:madeBySensor link our specific observation with its corresponding observed property, location, sensor, and measurement value. We create an instance of sosa:FeatureOfInterest class that represents the sample of water column during dive. GeoSPARQL is used to describe the precise location of the SES during the dive. As shown in Fig. 2, the geometry of the trip made by the SES is a set of points expressed by a WKT string. This string is linked to a geo:Geometry instance by the geo:asWKT property. The geo:hasGeometry property links the sosa:Result instance to an instance of the geo:Geometry class. Visually exploring the obervation URI can be done through http://linkeddata.cenpat-conicet.gob.ar/page/observation/id-233/location.

Fig. 2.

Example of observation made by TDR on location of the sea.

Fig. 3.

Example of an interval representing a dive of 10.94 seconds duration.

4.1.3.Phenomenon time

Figure 3 presents an observation produced by the TDR. Sometimes a measurement is related to a period of time. For example, the TDR measures the duration of a dive during a immersion. The property sosa:phenomenonTime links an instance of sosa:Observation to an instance of the class time:Interval. The properties time:hasEnd and time:hasBeginning point to an instance of the class time:Interval. To connect an instance of time:Instant with a xsd:dateTime value we use the time:inXSDDateTimeStamp property. The duration of the interval is described as instances of the class time:Duration. To explore the phenomenon time URI visually use the link http://linkeddata.cenpat-conicet.gob.ar/page/interval/sampleID-233

Fig. 4.

Representation of an observation (dwc:occurrence) made during a census. The occurrence of an SES is performed by a person at a location and time. Links to Geonames, DBPedia, Wikidata, and ORCID are shown highlighted.

4.1.4.Occurrence

It is true that we can model the SES census using SOSA/SSN because sosa:Sensor can be an observation made by a human instead of an electronic equipment. Thus, we decided not to use SOSA/SSN, and instead we used BiGe-Onto since it was created to model species occurrences by means of the DwC. We believe that if we want to share the results in an interdisciplinary way, it is necessary to respect the standard adopted by biodiversity community. Using DwC will also allow the reuse of this part of the dataset to perform more complex analyzes such as marine spatial planning. We use the class dwc:Occurrence to represent the SES observations made during a census. In Fig. 4 you can see the observation of two female pups on October 3, 2001. To represent gender we use the nerc’s URI (S10:S106). The property bigeonto:has_event connects the instance of occurrence with the instance of the event bigeonto:BioEvent. In the same way, each event instance is related to an instance of the geo:Geometry class through the bigeonto:has_location relationship. On the other hand, the dwciri:recordedBy property relates the instance of the occurrence to instances of the foaf:person class that perform the observation. Finally, the occurrence is associated with an instance of the class dwc:Organism through the bigeonto:associated property and the organism belongs to a specific taxon (dwc:Taxon) whose scientific name is Mirounga leonina. The red rectangles represent the links generated for the taxon in DBPedia and Wikidata, as well as the identifier of the person in ORCID. The link http://linkeddata.cenpat-conicet.gob.ar/page/occurrence/ID-36202 can be used to explore visually the occurence URI.

4.1.5.Publication

Each publication is represented as an instance of fabio:Expression. These expressions are also split by type of publication, thus using the respective subclasses of fabio:Expression. For instance, published books are represented with the class fabio:Book and so on. The shared properties for all the publications include date of publication (prism:publicationDate), title (dc:title), doi (prism:doi), authors (dc:creator), abstract (dc:abstract) and file format (dc:format). Finally, documents reference through dc:references to the respective platforms (sosa:platforms), which have been involved in the results reported in those documents. Figure 5 shows the modeling of a publication associated with a platform. To explore the publication URI visually use the link http://linkeddata.cenpat-conicet.gob.ar/page/paper/ID-45

Fig. 5.

Representation of a publication associated with the AAPR platform, the DOI is described using prism:doi and the external link (owl:sameAs) to OpenCitations was shortened for simplicity.

5.Data transformation process

To create Linked Open Data, a conversion needs to take place from the data contained in SES database into RDF. As explained in Section 2, measurements produced by sensors, and census data are stored in MySQL server. Fields that are no longer used or that contain confidential data are excluded, for example data that is still being processed. Transformation process is done by D2RQ Platform,1010 which consists of the D2RQ Mapping Language, used to write mappings between database tables and RDF vocabularies or OWL ontologies; the D2RQ Engine, a SPARQL-to-SQL rewriter that can evaluate SPARQL queries over your mapped database; and the D2R Server, a web application that provides access to the database via the SPARQL protocol, as Linked Data, and via a simple HTML interface. In order to see the complete mapping, one can use the link https://bitbucket.org/arglod/southern-elephant-seals/src/master/mappings/buceos.ttl to the project repository.

D2RQ runs at http://linkeddata.cenpat-conicet.gob.ar in back-end to browse structured data. It also has a SPARQL endpoint to be accessed from other applications, and a SPARQL explorer to query our own database in a friendly manner. One of the advantages that D2RQ provides is that after mapping, if the database is updated, it is not necessary to rewrite the mapping. Key statistics are presented in Table 5, and were computed in December 2020.

6.Use case: Accessing and analyzing data from dives and censuses

One crucial aspect is how to access and analyze data, and especially how to get only that part of data which is of interest for a given research question. To show the exploitation of the dataset, we developed a dashboard 1414 that allows querying the statistics of the dives and the taken routes. We use the R flexdashboard1515 package that allows generating web pages based on an R Markdown1616 document. To query our endpoint we use the SPARQL Package1717 that allows to import directly results of SPARQL SELECT queries into the statistical environment of R as a data frame. The following describes each module of the dashboard.

7.Discussion

This paper presents the publication as LOD of a biological and physical dataset, collected for more than 20 years, and stored with early objective of studying the environment influence on the foraging, reproductive performance, and population trend of the SES. This dataset was initially available for a small research group, and the aim is to make it available to a global community. Our development improves the discoverability of the content of the database and could be applied at new knowledge-building and cross-disciplinary research. For example, we expect the hydrographic profiles become a useful tool together with physical samples resulting from other science programs, to assess ocean changes associated with the climate change.

The dataset comes from PV a geographic region under conservation regulations by UNESCO and there is a continue demand of the governmental authorities to develop spatial planning. This requirement helps the sustainability of the database, because it needs a high level of accuracy for SES data, and access at other databases in a user-friendly manner. Coastal management Planning and Marine Spatial Planning (MSP) are concerned with the management of the distribution of human activities in space and time in and around seas and oceans to achieve ecological, economic and societal objectives and outcomes [3]. The next steps will be to promote the use vocabulary terms for discovery databases purposes of the institute CESIMAR, to allow the availability and suitability of data, to be used at regular review cycles of the MSP process. In addition, it would be desirable to access to the physical dataset collected by tourist and commercial vessels that overlap the same range in the southwest Atlantic Ocean. These hydrographic profiles could cover changes of the environment in all influence area of the SES distribution.

Particularly, this work shows the feasibility of using the SSN/SOSA ontology for modeling hydrographic measurements of instrumented animals and observation data collected during censuses. SSN/SOSA ontology is already a W3C standard, then we find this analysis as a valuable step towards the definition of the precise semantics of the ocean biodiversity systems, which requires of a collaborative effort.

The research also provides us useful insights into the process of developing and publishing data as LOD. First of all, valuable raw data can be highly heterogeneous and, as it was in this case, stored in relational databases developed by third parties, and even no longer maintained. Therefore, understanding how these data have been structured is one of the first barriers. Automated processes for bootstrapping SPARQL-to-SQL mappings, as those provided by the very same D2RQ or related technologies, fail in their attempts to automatically generate such mappings based on the content of data sources. Thus, much of this work have to done manually.

Ontology reusing is another challenging task, particularly in the context where standard ontologies have not been developed yet. Thus, our process required of a trade-off between the needed conceptualization (concerning the domain) and the availability of reused ontologies. As defined in [4], the strategy to reuse ontologies is another important obstacle, reusing concepts or properties from external ontologies (soft reuse) seems to be a good strategy to bootstrap LOD datasets from scratch, but it is limited for reasoning purposes. As a consequence and looking toward the future works, other strategies must be considered to increase the expressiveness of model without introducing unnecessary complexity.