The Semantic Sensor Network Ontology, Revamped

Abstract

The Semantic Sensor Network Ontology, popularly known as SSN, was developed by an Incubator Group of the World Wide Web Consortium (W3C) over 2009 to 2011. Subsequently, the W3C and the Open Geospatial Consortium (OGC) joined forces to update the SSN as informed by experience, to harmonize it with OGC's O&M, and to publish a new version to be endorsed as both a W3C Recommendation and an OGC standard in late 2017. The major contribution of the new SSN is a modular structure designed to be more convenient for ontology engineers and data custodians. It also slightly extends the coverage of the previous SSN with new terms for sampling and actuation. SSN retains the ability to comprehensively represent: sensors in terms of what they can sense, and what and how they do sense; observations in terms of what they measure and what values they nd; systems (or networks) of sensors in terms of sensor components and how they are deployed; and real-world objects (called features of interest, OGC-style) in terms of their physical properties, what can sense them, and what observations of them have been made. A few little-used SSN terms have been deprecated, and several others have been renamed. For a comprehensive description of new SSN the reader is referred to the speci cation [10]. A full description of the scope, design rationale and additions, with examples of its application are presented in [11].

Keywords

• sensor
• ontology
• OGC standard

Introduction

The Semantic Sensor Network Ontology, popularly known as SSN, was developed by an Incubator Group of the World Wide Web Consortium (W3C) over the period 2009 to 2011. It was well-timed for the rapid growth of industrial interest in the Internet of Things that ramped up around 2012, when the concept was expanded beyond RFID alone to include arbitrary sensors and actuators. The original SSN, hereafter referred to as SSNX, was largely an experimental design. It was in uenced in part by the by the Observations and Measurements (O&M) model (issued in 2007 [ 6 ] and later standardised by ISO[ 13 ] and the Open Geospatial Consortium (OGC) [ 8 ]). It was intended to serve research projects in, for example, agriculture [ 30 ], smart campus [ 26 ], environmental modelling [ 29,16 ], IoT [ 4 ], streaming [ 22 ], and climate modelling [ 18 ]. These projects developed or exploited semantic technologies, and were seen to bene t from a quality, shared ontology for describing sensors, networks of sensors, sensor capability, and the phenomena they sense. Arguably, it was designed for an ontology-expert audience and not for the emerging cohort of professional software engineers developing linked data infrastructures. Indeed, it was linked to the OGC's SensorML standards [ 5 ] through patterns showing how SSNX terms could be used to annotate entities and relations expressed SensorML's XML [ 19 ], but to the knowledge of the authors this design was seldom employed.

In 2014 the W3C and the OGC decided to work together to jointly progress the application of linked data techniques to spatial data, recognising the relatively low uptake of traditional spatial data infrastructure standards based on XML schemas and SOAP or REST interaction protocols. The Linking Geospatial Data workshop [ 1 ] was convened in 2014 to gauge the demand and scope for a joint standards activity. Later that year the Spatial Data on the Web Working Group was established, chaired by the rst author, with Ed Parsons of Google [ 31 ]. Amongst many other goals, the Working Group was to bring SSN forward as a formal recommendation of the W3C and a standard of the OGC. The key requirement for the new standard was to address the perceived complexity of the SSNX, considering modularisation as a possible approach to facilitate ease of use.

In this paper, written 18 months after the standard was published, we highlight the signi cant features of the new SSN, particularly aiming to provide insight into the rationale behind its design. We also take a look at how the new SSN is being used through a very brief survey of the literature. 2

Disentangling the upper ontology

Firstly, and most signi cantly, it was readily agreed that the reliance of SSNX on the Dolce Ultralite (DUL) upper ontology [ 24 ] through an owl:import statement added more confusion than clari cation, both through the introduction of very abstract terms and also through the dependency for representation of concrete measurement values. While it was recognised that the upper ontology gave some valuable structure and coherence to the design of SSNX, it was widely believed to get in the way of learning and using the SSNX. The rst step of redesign was to move axioms relating SSN terms to DUL terms away from the SSN ontology to a new Dolce Ultralite Alignment module, called SSN-DUL that itself imports SSN and DUL. We were able to update the DUL URL at the same time, as DUL had moved in the intervening years and this frequently annoyed SSN users. By this simple redesign SSN could be readily used either with or without DUL. 3

Re-thinking the O&M relationship

For the SSNX it was determined very early that an ontology of sensors needs to be able to talk about the observations sensors make. In O&M, arguably the central concept is an Observation, and this was conceived as an act [ 7 ]. In common language, as well as in related OCG documentation (such as https: //www.opengeospatial.org/standards/om), while we may use the verb to observe, we usually talk about making an observation, and the Observation itself is the artefact or record of the act of observing. The SSNX conceived the concept Observation as an information construct or social construct, more precisely, as a DUL situation that conveniently groups together the things you need to know about the what the Sensor did when it took a measurement. However, by the time of the SSN development in the context of a joint working group with the OGC, the conception of an observation as an event was entirely familiar. Furthermore, in the near aftermath of the long-running debate about the signi ers of URIs (http range 14 debate [ 32 ]), and declining interest in upper ontologies, there was little will for distinguishing physical world concepts from a social or digital interpretation of them. Consequently, SSN reverted to the O&M conception of an Observation as an event in explanatory annotations. On the other hand, the SSN has retained the property madeObservation and its inverse madeBySensor (previously observedBy) that connects Sensor to Observation. Unfortunately by this SSN has inherited the ontological confusion of O&M, stating that a Sensor makes an event Observation. In any case, in SSN the distinction is not signi cant as the relationship of Observation to other SSN concepts is una ected; but it does make a di erence to its integration with other ontologies including PROV [ 17 ] (where it is a Prov:Activity) and DUL (where it is now a dul:Event) [ 11 ]. 4

Modularity

The working group began with a starting point for modularity developed by Michael Compton, where parts of the ontology, with the topic-speci c focus as identi ed in the original SSN design, were separated into OWL ontologies with owl:imports relationships amongst them. However, this was found to be di cult to work with as the dependency between the terms from di erent modules typically led to an import of all modules, which increased the di culty instead of a simpli cation.

Modularity in DLs has been well-researched. The research has focused on extracting modules with a selected focus from pre-existing non-modular ontologies to satisfy various desirable coherence properties. For example, [ 15 ] extracts modules in EL and ALC, whereby a signature of terms de ne the focus of a module and the module is populated with axioms from the source ontology such that well-de ned dependencies of the signature terms are maintained. Alternatively, [ 3 ] develops package-based description logics that rede ne the semantics of interpretation of standard description logic together with a signature focus to extract terms from multiple independent ontologies, aiming to satisfy desirable coherence properties. Neither of these style of approaches was considered appropriate for SSN where we had the opportunity to designing modularity from scratch, and where a key requirement was always to make use of SSN easier by ontology engineers and data custodians.

One way to make it easier would be to reduce the signature, and another would be to reduce the expressivity of the ontology language and thereby reduce the burden of understanding the intent and consequence of complex axioms. The modular architecture of SSN does both these things. Furthermore, we wanted SSN modules to satisfy linked-data engineering practices that are not addressed by established modularity designs. These included { That each ontology module is retrievable from the module URI; { That the namespace for each module is the same as its URI; and { That each term is de ned in a unique module that is referenced using rdfs:isDefinedBy whenever it is used in an ontology Our SSN design satis es the desiderata of packaged description logics [ 3 ] by design, without requiring its non-standard inference.

As published, SSN has been designed with a central core ontology of terms considered to be most in-demand for use of the ontology, called SOSA for Sensor, Observation, Sample and Actuator which are its key concepts. The axioms of SOSA are simple in two respects. The signature is minimised by avoiding terms considered of only specialist use. The language of expression is minimised by using only the ALI(D) DL language. This language is convenient for performance of reasoning, and while OWL classes and object properties are declared, the only constructs that require OWL reasoning services (or RDF(S), for that matter) are inverse roles and concrete XML Schema datatypes. There is no subclass nor subproperty hierarchy. Intended domains and ranges of object properties are documented by way of schema.org's schema:domainIncludes and schema:rangeIncludes and these are explicitly declared to be annotation properties, not object properties in SOSA, hence cannot be reasoned with.

On the other hand, the interpretation of basic terms introduced in the SOSA core is constrained by more expressive ALRIN axioms in the SSN module which owl:imports SOSA. SSN also extends SOSA with further terms: 6 classes and 15 object properties to enrich the relationships amongst SOSA classes and to extend the modelling to aspects of Systems of Sensors and Deployment.

While SSN is consistent with SOSA, it cannot guarantee consistency with arbitrary other ontologies that also extend SOSA, under the usual semantics. Nor can it guarantee consistency with ABox assertions that are made in the SOSA context, with, for example, no formal domain, range or cardinality axioms. This is unfortunate, as it would be very convenient for data integration to have the property that any consistent SOSA ABox also forms a consistent ontology when combined with the SSN TBox. On the other hand, any consistent SSN ABox is a consistent ontology when combined with the SOSA TBox when instances of terms not present in SOSA are removed, so a data integration strategy of resorting to the lowest-common-denominator is simple to execute.

Figure 1 is adapted from the gures presented in [ 10,11 ] and shows the import relationships between the SSN, SSN-DUL and SOSA modules, together with some other modules introduced in the following. In the early days of the development of SSNX, there was an ontologically elegant e ort to structure the ontology around a skeleton fragment that explained what Sensors do. In particular, this skeleton represented Sensors as things that detect(s) a Stimulus in the physical world and that implement a Sensing process to produce (that is, inverse isProducedBy) a SensorOutput. However, ontology users have shown to be little interested in what sensors do when they capture data, so this was determined to be overly-complex. In the SSN this part of the skeleton was excised from the core SOSA where Observation has become of more central importance. 6

New terms

SSN has been extended very little from SSNX. While several new capabilities were derived from proposed use cases or directly proposed by Working Group members, these were largely eschewed in the interests of simplicity and generality.

However, partly in SOSA and partly in the greater SSN, the pattern for description of Observation has been cloned (with term changes) to describe the acts of Actuation and Sampling. An Actuation is made by an Actuator on a Feature of Interest, and a Sampling is made by a Sampler on a Feature of Interest.

In addition, in what may well be the most widely appreciated simple extension to SSNX, SOSA Observation has been extended by a simple datatype property hasSimpleResult that can be used to directly assert the literal values of an Observation, Actuation or Sampling, without further reference to external ontologies or via property chains or paths.

About a dozen SSNX terms have been renamed. 7

Other SSN modules

In addition to ssn-dul, several small alignment ontologies to popular external ontologies have also been released with SSN. One, the SSNX alignment module, is intended to ease the transition of existing SSNX applications to SSN by providing, as far as possible, subclass and subproperty relations that, interpreted by a reasoner, can infer SSN instances from SSNX instances. Similarly, PROV-O[ 17 ] and OBOE [ 23 ] alignments are provided.

A System capabilities module extends SSN by the terms and axioms from SSNX that are designed for modelling the capabilities of sensors and systems of sensors to measure particular physical properties, and identifying their measurement limitations. In this case they were separated from SSN as a separate module due to an analysis of use that showed they were used less often than the remaining SSN terms and so could be treated as specialist terms for a limited audience that are better not clogging up SSN.

Finally, the sample relations extends the basic structures for SOSA terms for the sampling process, designed for asserting relationships amongst di erent sampling individuals. Please refer to gure 1 for advice on using this module.

Presentation

The presentation of the new SSN is improved over the original SSNX by taking advantage of improved tools and practices developed by the linked data community. This new presentation should, itself, improve the usability of the new SSN. In comparison with the presentation of SSNX, the new ontologies consistently use rdfs:comment and skos:definition to annotate terms with descriptive explanations, and skos:example for an example for how it might be applied, expressed in English. The ontologies are documented with the Vocabulary for Annotations (vann) and Vocabulary of a Friend (voaf) terms for use with linked data tools.

Further, updated examples are used in the speci cation to support understanding. A section on Common Modelling Questions has been included as a response to the lessons learnt over the years of experience with SSNX. 9

SSN in the Wild

SOSA has been experimentally evaluated in the OGC [ 25 ] and some datasets and ontologies using SSNX have been updated. For example, measurements made by a meteorological station at an experimental farm in 2017 [ 27 ] are now published and queryable via SOSA [ 12 ]. The new sampling terms have been applied to describe a collection comprising millions of geological samples at Geoscience Australia [ 2 ]. Additional applications in research are published in the SSN workshop held at ISWC in 2018 [ 20 ]. 10

Conclusion

One-and-a-half years after the publication of the new SSN standard, and a halfyear since its publication in the academic literature[ 11,14 ], we nd that the 2011 SSN Incubator Group Report [ 19 ] and 2012 journal paper [ 10 ] are being cited at a greater rate than SSN. This may well be due to the time lag to create awareness of the update, but it would seem plausible that the added visibility due to its status as a standard of two standards organisations should have overcome the lag by now. It is hoped that this paper will encourage the shift to the new SSN.

While the simple core SOSA has not yet been taken up by schema.org [ 9 ], there are some clear equivalence mappings to relevant schema.org terms including sdo:variableMeasured with sosa:observedProperty; sdo:measurementTechnique with sosa:procedure; and sdo:object with sosa:featureOfInterest. Work will be undertaken to harmonise these vocabulary fragments in the near future.

Meanwhile, the complementary emerging W3C Web of Things standard [ 28 ], that is protocol and security-focussed, o ers a context extension method by which SSN can be employed for data description. The comparable OGC SensorThings standard, that addresses both an interaction protocol (like Web of Things), and a small vocabulary, has been published. It refers to part of the prior O&M which is mirrored in SOSA[ 21 ].

References

1. Archer , P. : Linking geospatial data . Tech. rep., W3C , London, UK (5-6 March 2014 ), https://www.w3.org/ 2014 /03/lgd/report

2. Australia , G.: Register, retrieved 29 May 2019 , http://pid.geoscience.gov.au/ sample/"

3. Bao J. , Voutsadakis G., S.G.H.V. : Package-Based Description Logics , LNCS , vol. 5445 , pp. 349 { 371 . Springer, Berlin, Heidelberg ( 2009 )

4. Bermudez-Edo , M. , Elsaleh , T. , Barnaghi , P. , Taylor , K.: IoTLite: A Lightweight Semantic Model for the Internet of Things . In: Proc Int'l IEEE Conferences on Ubiquitous Intelligence Computing etc , (UIC/ATC/ScalCom/CBDCom/IoP/SmartWorld). pp. 90 { 97 . IEEE, Toulouse, France ( July 2016 ), https://doi.org/10.1109/ UIC-ATC-ScalCom-CBDCom-IoP-SmartWorld . 2016 .0035

5. Botts , M. , Robin , A. : OGC SensorML: Model and XML Encoding Standard . OGC Encoding Standard, version 2 .0, Open Geospatial Consortium (Feb 04 2014 ), http: //www.opengeospatial.org/standards/sensorml

6. Cox , S.J.D. : Observations and Measurements - Part 1 - Observation schema . OpenGIS Implementation Standard OGC 07-022r1, Open Geospatial Consortium (Dec 08 2007 ), http://portal.opengeospatial.org/ les/?artifact id= 22466

7. Cox , S.J.D. : Observations and Measurements - XML Implementation . OGC Encoding Standard, Version 2.0 OGC 10-025r1, Open Geospatial Consortium (Mar 22 2011 ), http://portal.opengeospatial.org/ les/?artifact id= 41510

8. Cox , S.J.D. : Geographic information - Observations and measurements . OGC Abstract Speci cation, Version 2.0 OGC 10-004r3, Open Geospatial Consortium (Sep 17 2013 ), http://portal.opengeospatial.org/ les/?artifact id= 41579

9. Guha , R. , Brickley , D. , Macbeth , S. : Schema. org: Evolution of structured data on the web . ACM Queue 13 ( 9 ) ( 2015 ), https://doi.org/10.1145/2857274.2857276

10. Haller , A. , Janowicz , K. , Cox , S.J.D. , Le Phuoc , D. , Taylor , K., Lefrancois , M. : Semantic Sensor Network Ontology . W3C Recommendation, World Wide Web Consortium (Oct 19 2017 ), https://www.w3.org/TR/vocab-ssn/

11. Haller , A. , Janowicz , K. , Cox , S. , Lefranois , M. , Taylor , K., Le Phuoc , D. , Lieberman , J. , Garca-Castro , R. , Atkinson , R. , Stadler , C. : The modular SSN ontology: A joint W3C and OGC standard specifying the semantics of sensors, observations, sampling, and actuation . Semantic Web 10 , 9 { 32 ( 2019 ). https://doi.org/10.3233/SW-180320

12. IRSTEA , F.: Weather with SOSA/SSN ontology , retrieved 29 May 2019 , http: //ontology.irstea.fr/pmwiki.php/Site/Weather2017

13. ISO: ISO 19156: 2011 Geographic information - Observations and measurements . International Standard , ISO (Dec 2011 ), https://www.iso.org/standard/32574. html

14. Janowicz , K. , Haller , A. , Cox , S.J. , Le Phuoc , D. , Lefrancois , M. : Sosa: A lightweight ontology for sensors, observations, samples, and actuators . Journal of Web Semantics 56 , 1 { 10 ( 2019 )

15. Konev , B. , Lutz , C. , Walther , D. , Wolter , F. : Semantic modularity and module extraction in description logics . In: Proc. ECAI . pp. 55 { 59 (01 2008 ). https://doi.org/10.3233/978-1- 58603 -891-5-55

16. Le-Phuoc , D. , Dao-Tran , M. , Xavier Parreira , J. , Hauswirth , M.: A native and adaptive approach for uni ed processing of linked streams and linked data . In: The Semantic Web { ISWC 2011 . pp. 370 { 388 . Springer Berlin Heidelberg, Berlin, Heidelberg ( 2011 )

17. Lebo , T. , Sahoo , S. , McGuiness , D. : PROV-O: The PROV Ontology. W3C Recommendation , World Wide Web Consortium (Apr 30 2013 ), https://www.w3.org/ TR/prov-o/

18. Lefort , L. , Haller , A. , Taylor , K., Squire , G. , Taylor , P., Percival , D. , Woolf , A. : The ACORN-SAT linked climate dataset . Semantic Web 8 , 1 { 9 ( 09 2016 ). https://doi.org/10.3233/SW-160241

19. Lefort , L. , Henson , C. , Taylor , K.: Semantic Sensor Network XG Final Report. W3C Incubator Group Report, World Wide Web Consortium (Jun 28 2011 ), http: //www.w3.org/2005/Incubator/ssn/XGR-ssn- 20110628 /

20. Lefrancois , M. , Castro , R.G. , Gyrard , A. , Taylor , K. (eds.): Proceedings of the 9th International Semantic Sensor Networks Workshop , vol. 2213 . CEUR workshop proceedings (9 October 2018 )

21. Liang , S. , Huang , C.Y. , Khalafbeigi , T.: SensorThings API part 1: Sensing . Tech. Rep . 15 - 078r6 , OGC (07 2016 ), available at http://docs.opengeospatial.org/is/ 15-078r6/ 15 - 078r6 . html#18

22. Llaves , A. , Corcho , O. , Taylor , P., Taylor , K.: Enabling RDF stream processing for sensor data management in the environmental domain . International Journal on Semantic Web and Information Systems (IJSWIS) 12 , 1 { 21 ( 2016 ). https://doi.org/doi:10.4018/IJSWIS.2016100101

23. Madina , J. , Bowers , S. , Schildhauer , M. , Krivovc , S. , Pennington , D. , Villa , F. : An ontology for describing and synthesizing ecological observation data . Ecological Informatics 2 ( 3 ), 279 { 296 ( 2007 ), https://doi.org/10.1016/j.ecoinf. 2007 . 05 .004

24. Masolo , C. , Borgo , S. , Gangemini , A. , Guarino , N. , Oltramari , A. , Schneider , L. : The WonderWeb Library of Fundational Ontologies and the DOLCE ontology . Technical report, LOA-ISTC ( 2003 )

25. OGC environmental linked features interoperability experiment engineering report . Tech. rep. , OGC (11 February 2019 ), https://docs.opengeospatial.org/per/18- 097 . html#Standards Best Practices

26. Ploennigs , J. , Schumann , A. , Lecue , F. : Extending semantic sensor networks for automatically tackling smart building problems . In: ECAI ( 2014 )

27. Roussey , C. , Bernard , S. , Andre , G. , Corcho , O. , De Sousa , G. , Bo ety, D., Chanet , J.P. : Weather Station Data Publication at IRSTEA: an Implementation Report . In: Joint Proc 6th Foundations , Technologies and Applications of the Geospatial Web and 7th Semantic Sensor Networks . vol. 1401 . CEUR, Riva del Garda , Italy (Oct 2014 )

28. Web of things (WoT) thing description: W3C candidate recommendation 16 May 2019 . Tech. rep., W3C (May 2019 ), https://www.w3.org/TR/2019/ CR-wot - thing-description- 20190516 /

29. Stocker , M. , Shurpali , N. , Taylor , K., Burba , G. , Ronkko, M. , Kolehmainen , M. : Emrooz: A scalable database for ssn observations . In: SSN-TC-ORDRING 2015 , Joint Proc Semantic Sensor Networks, Terra Cognita, Ordering and Reasoning . vol. 1488 , pp. 1 { 12 . CEUR Workshop Proceedings (10 2015 ). https://doi.org/10.13140 /RG.2.1.4066.9921

30. Taylor , K., Gri th , C. , Lefort , L. , Gaire , R. , Compton , M. , Wark , T. , Lamb , D. , Falzon , G. , Trotter , M. : Farming the web of things . IEEE Intelligent Systems 28 ( 6 ), 12 {19 (Nov 2013 ). https://doi.org/10.1109/MIS. 2013 .102

31. Taylor , K., Parsons , E.: Where is everywhere: Bringing location to the web . IEEE Internet Computing 19 ( 2 ), 83 {87 (Mar 2015 ), https://doi.org/10.1109/MIC. 2015 . 50

32. W3C: Httprange14webography ( 2013 , retrieved 27 May 2019 ), https://www.w3. org/wiki/HttpRange14Webography