Sensor observations are usually o ered in relation to a speci c purpose, e.g., for reporting ne dust emissions, following strict procedures, and spatio-temporal scales. Consequently, the huge amount of data gathered by today's public and private sensor networks is most often not reused outside of its initial creation context. Fostering the reusability of observations and derived applications calls for (i) spatial, temporal, and thematic aggregation of measured values, and (ii) easy integration mechanisms with external data sources. In this paper, we investigate how work on sensor observation aggregation can be incorporated into a Linked Data framework focusing on external linkage as well as provenance information. We show that Linked Data adds new aspects to the aggregation problem, e.g., whether external links from one of the original observations can be preserved for the aggregate. The Stimulus-SensorObservation (SSO) ontology design pattern is extended by classes and relations necessary to model the aggregation of sensor observations.
Sensor observations are collected with a speci c purpose in mind and, therefore,
measuring follows strict procedures and spatio-temporal scales [
Building up on our previous work on exposing standardized observation data
as Linked Data [
The main contributions of this paper are threefold. We (i) present a Linked Data model for aggregated sensor data, (ii) discuss the e ects of aggregation on links from and to observations, and (iii) outline the role of provenance in this setting. The implementation of the extensions discussed in this paper are ongoing and the 52 North semantics community7 plans to release an updated prototype in fall 2011.
The remainder of this paper is structured as follows. In section 2.1, we introduce the concept of aggregated observations and provide background information about Linked Sensor Data and provenance information in observations. Section 3 discusses the implication of aggregation on Linked Sensor Data. Here, we present the required extensions to our Linked Data model for observations. Additional investigations address the e ects of aggregation on external linking, and issues on data provenance. In section 4, we set our work in relation to current e orts to provide observations as Linked Data and to provide provenance information in observation data. The paper concludes with a summary and an outline of the remaining steps for implementing aggregated observations as Linked Data; see section 5. 2
In this section, we provide a brief overview on related work. At rst, we introduce de nitions and related work about aggregation of observations. Second, we introduce the concepts of Linked Data. Finally, we describe related work about provenance of sensor data. 2.1
Aggregation of observations in space and time is essential to derive information
that is useful for a certain application purpose and to integrate observation data
with di ering spatio-temporal resolutions. Yet, spatio-temporal aggregation of
7 Implementations and documentations can be found at http://52north.org.
observations in the Linked Data context has not yet been addressed. However, in
other communities, e.g., the database community or in environmental sciences,
spatio-temporal aggregation has been a research topic for years and is sometimes
also referred to as scaling of observations and environmental models. Vega and
Lopez [
In this paper, we largely follow the de nition of aggregation8 by Jeong et. al.
[
Besides spatio-temporal aggregation as introduced above, extracting high
level events from observations is also done by aggregating observations. Treating
the high level events as observations again enables an easy integration into
existing infrastructures and tools. Considering the blizzard example as described
in [
Several approaches for Linked Sensor Data in the Web are already available
[
There are several approaches available for providing provenance information in
the Web. The W3C's Provenance Incubator Group9, predecessor of the new
Provenance Working Group10, compiled a list of requirements to support
provenance in RDF, which includes for example that every observation should have an
URI identi er [
Vocabulary 12 http://openprovenance.org/
Besides general approaches for provenance information in the Web,
providing provenance information in Linked Sensor Data has recently gained attention.
Provenance of sensor data can be de ned as information about the source of the
sensor data as well as information about transformations applied to the original
data [
In this section we introduce an approach to enable the aggregation of
observations in the Linked Data cloud. First, we present an extension of the
StimulusSensor-Observation (SSO) ontology design pattern [
Following our previous work [
Figure 1 shows the classes and relations from the pattern extended by the Linked Data model for sensor data, and the new elements that have been added in order to account for aggregation. In a nutshell, we reuse the following de nitions: { FeatureOfInterest : entity that comprises observable properties. { ObservedProperty : property that inheres in a feature of interest. { ObservationCollection: set of observations, grouped by a distinct criteria. { Observation: (social) construct that connects observed properties with sensors, sensing results, and sampling times. { SamplingTime: time instant or interval at which an observation was made. { Result : symbol representing an observed value. { Sensor : entity that performs observations.
{ Procedure: description that speci es how observations have to be carried out
In order to account for aggregated observations as Linked Data, we extend the SSO pattern with the following elements: { isAggregateOf : a relation that allows one observation to be aggregated out of others; e.g., an observation of daily P M 10 concentration being an aggregate over hourly measures, or an observation of P M 10 in Munster, Germany being an aggregate over various Point of Interest (POI) measures. { SensingDevice13: a sensor, which is a physical measuring device; e.g., a particular air sampler including a special lter P M 10. { AggregationProcess : a sensor, which implements a concrete aggregation procedure (see below), for example the process that calculates regional P M 10 concentrations based on several P M 10 concentration observations and additional calibration parameters. 13 The concept of a SensingDevice is also captured as part of the W3C SSN-XG ontology. However, it is not part of the SSO pattern, which is applied in our work. We decided to introduce the SensingDevice in particular opposed to the notion of the AggregationProcess in order to stress the di erence between a single physical measurement instrument and the aggregation process that combines multiple sensory inputs to a new observation. { AggregationProcedure: the speci c procedure used for aggregating several observations into one; e.g., calculating the MEAN of 24 hourly observations of P M 10 concentration, or a Kriging interpolation method
The relations between the classes presented in Figure 1 act as links in our
model and de ne the multiple navigation paths and external references; see also
[
Aggregating linked observations a ects the links from and to the observations. Questions like 'Are the links to a feature of interest still valid, if observations taken at speci c points are spatially aggregated to an area?' or 'Which new links can be established after aggregation of an observation?' need to be answered. First of all, the links which are de ned in our observation ontology need to be checked for consistency and changed, if necessary14. Table 1 shows examples of objects (i.e., link targets) of the links from observations before and after aggregation of point observations to an area in space and a period in time.
Independent of the concrete example, for each aggregation the target of the ldm:hasSamplingTime link changes from an instant in time (original observations) to a period in time (aggregated observations), if the observations are aggregated temporally. Also, the DUL:includesObject link will always point from the aggregated observation to an instance of an AggregationProcess instead of pointing to a speci c SensingDevice from the original observations. In environmental applications, the ObservedProperty is usually a continuous phenomenon, which is sampled at certain locations in space or time, e.g., P M 10 concentration. If only spatial and/or temporal aggregations are applied, the ldm:aboutProperty remains the same. In case of a thematic aggregation (see Section 2.1), the ObservedProperty changes. An example is the blizzard as a combination of high wind-speed, heavy snowfall, and low temperatures: the ObservedProperty of the blizzard observation is the phenomenon of the blizzard, whereas the original observations point to the phenomena wind-speed, snowfall and surface temperature. Similar examples could be constructed for landslides or forest res.
Though the ObservedProperty might be unchanged during an aggregation process, the sso:isPropertyOf link changes, if the observations are aggregated in 14 Here, changing links means that triples of the original observations might be removed and replaced by other triples in the aggregated observations for the same relationship. For example, the hasSamplingTime relationship usually links to a point in time in the original observations, but to a time period in the aggregated observations, if the observations are aggregated temporally.
Link in Ontology ldm:hasSamplingTime DUL:includesObject ldm:aboutProperty sso:isPropertyOf
Object Before Aggregation Object After Aggregation TimeInstant (08/05/2011; 23:15 CEST)
TimePeriod (one day) SensingDevice (air sampler) ObservedProperty (PM10) SamplingPoint (N 51 57.466 E 007.37.433)
AggregationProcess (block kriging of PM10 measures) ObservedProperty (PM10) GeospatialRegion (area of Munster) space. For example, aggregating the point measurements to an area causes the FeatureOfInterest to change from a sampling point to an upper level feature like the area of the city of Munster. Finally, the sso:involves link points to an aggregate computed during the aggregation process. Originally, the sso:involves link has pointed to the measurement values from the source observations. Besides changing the original links, additional links might be added pointing to the aggregated observation. As introduced in our model, the isAggregateOf link points from an aggregated observation to the original observations. Furthermore, other observation collections might contain the aggregated observation resulting in new ldm:hasObservation links to the aggregated observation. Also, other higher level features like cities, administrative areas, etc. might be linked to the aggregated observations.
Formalizing the changes of links during aggregation is challenging and often domain speci c. However, we consider the identi cation and formalization of such changes as crucial to provide a (semi-)automated aggregation of observations in the Linked Data cloud in future and are currently working on such a formalization. 3.3
In a Linked Data context where di erent communities might have interest in interlinking their datasets, it is important to publish trust-able datasets. Provenance information favors trustworthiness of data because users are able to analyze the historic changes and reproduce them [24]. Especially when aggregating observations in Linked Data, it is important to be able to retrieve information about the original observations as well as the aggregation process that has been applied. Figure 2 shows a provenance graph that illustrates how provenance information about the aggregation process and original observations is provided in our Linked Data model and how this can be mapped to the concepts of the OPM and the Provenance Vocabulary. The reason to extend our model instead of re-using an existing solution lies in the fact that most of the provenance information needed for sensors and observations is already available, thus we avoid redundancy.
First of all, the isAggregateOf allows to trace the aggregated observations back to the original observations. Hence it can be mapped to the opmv:wasDerivedFrom relationship in the OPM. Though the isAggregateOf relationship cannot be directly mapped to a relationship in the Provenance Vocabulary, it provides basically the information that is provided by the prv:usedData link from a prv:DataCreator to a prv:DataItem. Information about the aggregation process that has created an aggregated observation is provided by the DUL:includesObject link from the aggregated observation to the AggregationProcess. This link can be mapped to the opmv:wasGeneratedBy relationship in the OPM and to the prv:createdBy relationship of the Provenance Vocabulary. The ldm:hasSamplingTime attribute provides a link to the time at which the value represents a physical phenomenon in the world. In case of observations taken by a physical sensor this corresponds to the time when the observation has been taken. However, if the observations gathered by physical sensors are aggregated by an AggregationProcess, the SamplingTime is a time period representing the value for which the aggregate is valid. This is no longer the time when the observation has been produced (time of aggregation). Thus, for aggregated observations, an additional time link might be added providing this information. Similarly, additional links might be provided for the opmv:wasControlledBy and the opmv:used relationships of the OPM, which we did not yet include, as we focus on retracing the observations and not on the users which are aggregating or using the observations.
The presented research is in line with the theoretical challenges in Sensor Web
research, which have been identi ed during an expert meeting in 2010 [25],
addressing the challenges of interoperability and integration of sensor based system
and model based systems. Our extension of the SSO design pattern as described
in Section 3.1 allows to expose aggregated observations as Linked Sensor Data.
This goes beyond the approaches available for providing Linked Sensor Data
[
Our model also allows to retrace the aggregated observations back to the
original observations and to retrieve information about the aggregation process
applied, thus providing provenance information about the aggregated
observations (see Section 3.3). Instead of providing additional metadata as in other
approaches described in Section 2.3, we show how the provenance information
can be directly retrieved by using the links established in our Linked Data model
for (aggregated) observations. For example, Patni et. al. [
Both, sensor observations and aggregates provide estimations for physical phenomena occurring in the world. As it is not possible to observe all relevant aspects in reality, observations can only represent reality to a certain degree and thus are uncertain about reality. In studies dealing and using observations, it is crucial to account for the uncertainty. This is usually referred to as uncertainty propagation [26]. Aggregation is one mean to adjust the uncertainty in estimations. The more the data is aggregated, the less uncertainty is in the data. At the moment, we do not yet explicitly account for uncertainty in the presented work. Investigations how uncertainty can be propagated in observation processing work ows in the Web are currently ongoing within the European research project UncertWeb15 [27]. We plan to adopt their approaches and add the uncertainty to our Linked Data model.
While we are providing the model for exposing aggregated observations as Linked Data and we discuss the e ects on links from and to observations during aggregation (Section 3.2), we have not yet addressed the technological aspect of executing aggregation processes on Linked Sensor Data. However, we are currently working on extending our Spatio-Temporal Aggregation Service to also deal with Linked Data serialized as RDF. This also leads to the question to what degree observations should be aggregated before exposing them as Linked Data in order to reduce the amount of triples or whether observations can/should be provided at di erent aggregation levels as Linked Sensor Data. For example, providing high resolution sensor data as Linked Data might lead to a huge amount of triples which might cause performance problems. Thus, it might be better to aggregate the observations before and then expose them as Linked Data. 5
In this paper, we identify the need for spatial, temporal, and thematic aggregations of sensor observations and their propagation as Linked Data for an easy integration with other data sources. Aggregates of sensor observations (e.g., the monthly average ne dust concentration in a city) can be much easier utilized in applications. Facilitating the integration of such aggregated observations by providing them as Linked Data enables their utilization among di erent applications. We achieve this by: (1) extending the SSO ontology design pattern to accommodate aggregation information and including concepts such as AggregationProcedure or AggregationSensor ; (2) describing how links from point observations change after aggregation (e.g., feature of interest may change from a sampling point to a city area); (3) supporting the provenance information in the model through enabling retraceability to original observations and introduce relations such as isAggregationOf.
Our future work will follow these lines. Aside from our ongoing implementation work, we plan to exploit the combination of the proposed approach with event detection mechanisms and stream processing. Therefore, we are planning to combine the extension of the SSO ontology pattern presented in this paper with our previous work on sensor plug & play [28]. In that work, we designed a framework that enables the on-the- y integration of sensors and Sensor Web services by determining the semantic matching between sensor characteristics and service requirements. This framework can also be put to use in on-stream processing for the dynamic fusion of incoming data streams of multiple sensors to produce aggregated observations. This is similar to approaches such as [29], 15 http://www.uncertweb.org but also allows the creation of new, combined phenomena. A basic example is the combination of temperature and conductivity data streams measured by underwater sensors to derive a stream of salinity observations.
Furthermore, we are planning to extend our approach developed with representations for uncertainty as described in the Uncertainty Markup Language [30]. Our provenance information currently provides information about the aggregation procedure applied, its implementation, and about the original observations that have been used to derive the aggregated observation. In future, it has to be explored how to add additional provenance information about providers and users of the (aggregated) observations.
Our approach of aggregation in Linked Data also allows to utilize the semantics of the links and the objects. First of all, the changes to links as described in Section 3.2 can be translated into rules to check whether adding or removing links is allowed or not. In a next step, the process of adding and removing links during aggregation of observations might be automatized. Furthermore, the semantic reasoning can be used to decide, whether a certain aggregation procedure can be applied to a certain set of observations. Considering, e.g., a set of water level measurements along rivers in Germany, these should not be interpolated to Germany and the semantics can be used to recommend appropriate or disallow inappropriate aggregation processes. However, in order to realize such a system, an ontology of aggregation processes is needed which we consider to be work done in a longer time frame. We hope that our approach as presented in this paper will contribute towards such a semantically-enabled aggregation system.
The presented work is developed within the 52 North semantics community (http://52north.org/semantics), and partly funded by the European projects UncertWeb (FP7-248488, http://www.uncertweb.org/), ENVIROFI (FP7-284898, http://www.envirofi.eu/), ENVISION (FP7-249170, http:// www.envision-project.eu/), and through the International Research Training Group on Semantic Integration of Geospatial Information funded by the DFG (German Research Foundation), GRK 1498. We are thankful for discussions with members of the Munster Semantic Interoperability Lab (MUSIL), the Spatial Data Infrastructures Unit of the Joint Research Centre (JRC) of the European Commission, and colleagues from the W3C Semantic Sensor Network Incubator Group. 22. Liu, Y., Futrelle, J., Myers, J., Rodriguez, A., Kooper, R.: A provenance-aware virtual sensor system using the open provenance model. In: 2010 International Symposium on Collaborative Technologies and Systems (CTS). (2010) 330 {339 23. Lefort, L., Henson, C., Taylor, K., Barnaghi, P., Compton, M., Corcho, O., Castro, R., Graybeal, J., Herzog, A., Janowicz, K., Neuhaus, H., Nikolov, A., Page, K.: Semantic Sensor Network XG Final Report. W3C Incubator Group Report 28 June 2011. Available at: http://www.w3.org/2005/Incubator/ssn/XGR-ssn/.
Technical report (2011) 24. Boose, E.R., Ellison, A.M., Osterweil, L.J., Clarke, L.A., Podorozhny, R., Hadley, J.L., Wise, A., Foster, D.R.: Ensuring reliable datasets for environmental models and forecasts. Ecological Informatics 2(3) (2007) 237 { 247 Meta-information systems and ontologies. A Special Feature from the 5th International Conference on Ecological Informatics ISEI5, Santa Barbara, CA, Dec. 4-7, 2006 - Novel Concepts of Ecological Data Management S.I. 25. Schade, S., Craglia, M.: A future sensor web for the environment in europe. In: Proceedings of the 24th International Conference on Informatics for Environmental Protection - Enviroinfo2010. (2010) 26. Heuvelink, G.: Error Propagation in Environmental Modelling with GIS. Taylor & Francis (1998) 27. Pebesma, E., Cornford, D., Nativi, S., Stasch, C.: The uncertainty enabled model web (uncertweb). In: Environmental Information Systems and Services - Infrastructures and Platforms. (2010) 28. Broring, A., Maue, P., Janowicz, K., Nust, D., Malewski, C.: Semantically-enabled sensor plug & play for the sensor web. Sensors 11(8) (2011) 7568{7605 29. Madden, S., Franklin, M.J., Hellerstein, J.M., Hong, W.: TAG: a Tiny AGgregation service for ad-hoc sensor networks. SIGOPS Operating Systems Review 36 (2001) 131{146 30. Williams, M. and Conford, D. and Bastin, L. and Pebesma, E.: OGC 08-122r2: Uncertainty Markup Language (UncertML) (2009)