Article Preview
TopIntroduction
Built upon the huge quantities of live and archived sensor data available in the Web, the so-called Sensor Web (Botts, Percivall, Reed, & Davidson, 2006), the Semantic Sensor Web is a framework for providing enhanced meaning for sensor observations so as to increase interoperability (Sheth, Henson, & Sahoo, 2008). Such meaning is represented through ontologies (i.e., machine-processable formal models) that support data description, discovery and integration.
The inclusion of ontologies into Sensor Web infrastructures has led these infrastructures to evolve into Semantic Sensor Web ones, i.e., software infrastructures capable of processing sensor data that are modeled with ontologies and distributed on the Web.
The UML models designed by the Sensor Web Enablement Working Group (OGC, n.d.) in the Open Geospatial Consortium (OGC) (e.g., the Observations and Measurements (Cox et al., 2010) and the Sensor Model Language (Botts & Robin, 2007)) are frequently used in sensing infrastructures and have sometimes been transformed into ontological models. Furthermore, the recent development of the W3C Semantic Sensor Network (SSN) ontology (Lefort et al., 2011) will push forward the use of common ontological models for representing sensor networks.
However, current Semantic Sensor Web infrastructures have further representational requirements than just modelling sensor networks. This is because sensor data are distributed, which requires being aware of the different sensor datasets available; are heterogeneous, which requires knowing the schemas used in these datasets; and are web-accessible, which requires having descriptions of the web services that provide dataset access.
Nevertheless, none of the data models proposed fully covers these requirements; they are partially covered in some models but some requirements have not been dealt with yet (i.e., dataset schema descriptions).
Furthermore, in addition to the types of data just mentioned, Semantic Sensor Web infrastructures need to manage further data (e.g., geographical, organizational, or domain-specific data); thus, one of the dilemmas that developers face is to know which is the most appropriate way to model domain-specific features of interest and their properties, so that they can link domain data with sensor data.
The main contribution of this paper is the definition of a core ontological model for Semantic Sensor Web infrastructures that allows describing sensor networks (by extending the SSN ontology), sensor data sources with their underlying schemas, and the web services that these sources expose.
This core ontological model provides a shared vocabulary to discover, access, and integrate information within the infrastructure and to interoperate both across the internal infrastructure services and between an infrastructure and the external sources that adopt alternative approaches (e.g., the OGC Sensor Web Enablement ones) (Botts et al., 2006).
A second contribution is a set of guidelines to model domain-specific features of interest and properties; these guidelines are based on frequent requirements and, in contrast to the current practice, they encourage to take advantage of the expressiveness of the underlying logical formalism and to add logical constraints to the model.
As an example of application, the core ontological model is currently being used in a Semantic Sensor Web infrastructure that is being applied to a use case in the domain of coastal flood emergency planning (Gray et al., 2011). This development required us to define a set of ontologies to represent those data that are specific to that domain, and to this end the mentioned guidelines have been applied.
As any other engineered product, ontologies may have defects; some of these defects may appear during their development, but others only emerge once they have been deployed. The ontologies described in this paper have been evaluated through different aspects (i.e., vocabulary, syntax, structure, semantics, representation, and context) to ensure their suitability for their intended use scenario; in this paper we present the details of such an evaluation to guide future evaluation efforts.