The informative value of analytic processes by geographical information systems depends on the accuracy, consistency and completeness of the gathered data fed into the system. By feeding Big Data into it, such requirements are hard to maintain, as the provenance, veracity, velocity, structural and semantic heterogeneities of the gathered spatiotemporal data have to be addressed. Exploitation and integration of Big Data in such ways is an ongoing challenge. We present fundamentals of a well-de ned and collaborative information integration approach based on semantic web technology, established ontologies and linked APIs that speci cally emphasizes a spatio-temporal relation and enable a new generation of geographical information systems. We employ the concept of smart web services for dynamically composed work ows in order to cope with the characteristics of Big Data value streams and generate more elaborated data.
Geographical information systems (GISs) are important tools for decision support based on spatio-temporal data. These tools are used in various elds like civil planning, emergency management, agriculture or environment and nature protection. In this paper, we introduce a novel approach of how GISs can exploit and integrate Big Data based on semantic web technology, established ontologies and linked application programming interfaces (APIs). Due to improved and pervasive sensor technology and data created by mobile devices and users of social Web applications, the amount of spatio-temporal data is increasing. At the same time, the reliability of these data may be uncertain and need to be taken into consideration when used in GIS. In addition, spatio-temporal data from di erent sources may use di erent schemas to describe locations, like addresses, relative spatial relationships or di erent coordinates reference systems. The quantities measured and units used for data values may also vary across heterogeneous and uncontrolled data sources. Due to these developments, GIS are facing challenges in all four dimensions of Big Data: { Volume: The prevalence and omnipresence of sensor technology and ubiquitous data sources inposes challenges regarding data volumes to be integrated. { Variety: Unstructured data are new kinds of data for GIS, which require innovative methods of data interpretation for analyzing, interpolating, predicting and visualizing. { Velocity: In order to permanently integrate acquired sensor data in GIS, the common batch processing of these systems have to be technically and conceptually reorganized in order to enable real-time analysis and activity recommendations. { Veracity: The integration of volunteered geographic information (VGI) and other user-created content as well as integration of remote sensing analyzed image processing data, which may be incomplete prevent the assumption that collected data are complete and correct at any given point on time.
By feeding Big Data into GISs, we have to take these characteristics into consideration with a special focus on the requirements imposed by GISs, including the provenance information of data. In our approach, we use semantic Web technology to i) describe data sources and data transformation services for GIS in a machine interpretable way. This enables smart web services to ii) compose a work ow that generates the result set of more elaborated data with respect to accuracy, consistency and completeness. The informative value of analytic processes by GISs depends on the data generated by the composed work ow used as input. When integrating heterogeneous sources of spatio-temporal data with di erent units of measurement, property de nitions or coordinates reference systems, the data have to be transformed into a uni ed result set across all sources. The schema of this result set has to be exible and de ned on demand in order to match the requirements of di erent use cases. This requires a structure for meta data that represents the relations of the data that should be integrated. By describing heterogeneous data sources for GIS together with input and output parameters of available data transformation services semantically, work ows for processing these data can be composed dynamically in order to ful ll use case speci c requirements. The approach presented in this paper enables domain experts to select a combination of data sources and de ne the data structure needed for a speci c use case with respect to the quantities, units, granularity, precision of measurements, period and area under investigation. On the other hand, all provenance information has to be retained in order to make the values of di erent sources comparable. Using semantic Web technology for managing spatio-temporal data also enables semantic analysis for unstructured and remotely sensed data. We investigate a semantic work ow composition approach for integrating Big Data in GISs and hypothesise that the increasing amount of geographic data will signi cantly improve the scienti c ndings of GISs closer to reality. However, the level of improvement strongly depends on a common understanding of concepts across heterogeneous data sources. This leads to the following research questions: RQ1 Which type of provenance information is relevant in order to make the processed values of heterogeneous data sources comparable within a uni ed result set? RQ2 How does a collaborative approach of describing sources and services for spatio-temporal data scale for Big Data processing work ows in GIS? Our approach to answer these research questions is based on the related work presented in Section 2. We present fundamentals of a well-de ned and collaborative information integration approach in Section 3, show a concrete use case in Section 4 and discuss the preliminary results in Section 5. 2
In addition to the related work discussed in [
Transforming data from heterogeneous data sources into a uni ed schema and
the interoperability of distributed systems is still an ongoing research topic where
web services are commonly used for converting data. As an example, Stolz and
Hepp [
2 http://purl.oclc.org/NET/ssnx/ssn 3 http://www.opengeospatial.org/ 4 http://www.opengeospatial.org/ogc/markets-technologies/swe 5 http://www.opengeospatial.org/standards/sos 6 http://www.opengeospatial.org/standards/waterml Quantities, Units, Dimensions and Data Types Ontologies (QUDT)7 can be used as a common standard for describing units and their conversion. 2.2
Vettor et al. [
The work presented in this section expresses that exploitation and integration of Big Data in a way that addresses provenance, veracity, velocity, structural and semantic heterogeneities of spatio-temporal data, especially for GIS, is an ongoing challenge. Based on the related work introduced in this section, we present our collaborative information integration approach for spatio-temporal data in GIS in Section 3. 3
In our approach, we present fundamentals of a well-de ned and collaborative information integration of spatio-temporal data used in GIS. We address the di erent requirements on more elaborated data for data analysis by describing sources of spatio-temporal data and APIs using semantic web technology and established ontologies. We employ the concept of smart web services for dynamically composed work ows in order to cope with the characteristics of Big Data
value streams. For contributing to the exibility and usability of GIS, we pose the following requirements: R1 Collaborative: Users should be able to add sources of spatio-temporal data and relevant APIs to the GIS.
R2 Semantical : All sources of spatio-temporal data and relevant APIs have to be described in a machine interpretable way.
R3 E cient : Value streams of spatio-temporal observations have to be transmitted and processed with the least possible amount of overhead in order to cope with high volume data.
The rst step is to build a collaborative system based on SMW for managing meta data. This system does import and reuse commonly used vocabulary in the domain of GISs like SSN, QUDT and GeoVocab10, which does also cover the Basic Geo Vocabulary11, to semantically describe data sources and the Hydra vocabulary12 for APIs of transformation services. This information is used to dynamically build work ows consisting of the data sources and (smart) transformation services needed to ful ll the data conditions requested by the data consumer. For an e cient processing of the spatio-temporal data, only the meta data of these value streams are modeled with the exibility of semantic web data formats like RDF, while the observed values are transmitted and transformed with the least possible amount of overhead. The scope of our work within the general architecture of a GIS is shown in Figure 1.
Users create description pages of available data sources or APIs in SMW and the system provides a representation of this information in RDF using our speci ed ontology annotated with common vocabularies. Smart web services can therefore use this information instantly. We provide a RESTful API that enable domain experts to query the quantities, units, granularity, precision of measurements, period and area under investigation for a speci c use case. The response of this API call can be the plain observations within the result set (e.g. JSON, CSV, XML), the observations with semantically described meta data (e.g. JSON-LD, RDF/XML) or a hyperlink to a relational data base that holds the values of the result set, depending on the needs of the data consumer. As the RESTful API used for the uni ed data access is described semantically itself, all parameters and allowed values can be queried by a consuming application which allows for continuous integration of more functionality. A concrete use case of our approach is described in Section 4. 4
Analyzing the characteristics of suburban heat islands (SUHIs), that means heat island that occur within cites on hot days, requires a set of records for thermodynamic temperature values from heterogeneous sources for the area and period of 10 http://geovocab.org/ 11 https://www.w3.org/2003/01/geo/ 12 http://www.hydra-cg.com/spec/latest/core/
SpatioTemporal Big
Data
Volume Veracity Velocity Variety
Geographical Information System
Linked APIs and
Cognitive Apps Ontologies
Linkage Semantic MediaWiki - Reuse Ontologies - Model Interrelation - Define Sources, APIs, Workflows
SSN Scope
QB QUDT etc.
Consistent Complete Accurate
Decision Support System investigation. As a rst demonstration of our approach, we gather data from the regional environment authorities of Baden-Wurttemberg13, the German weather service14, mobile measurements on an urban railway15 and remote sensing data from satellites operated by European Space Agency and National Aeronautics and Space Administration. With the gathered data as our training set, we plan to perform predictions of SUHIs within the city of Karlsruhe, Germany, and evaluate the prediction with our test data set which is classi ed as measurements from SUHIs in the same city. By varying the sources used as input for the predictions, we are going to evaluate the impact of these sources on the decision support in GISs.
At this stage of our work we have developed the ontology that describes the sources of spatio-temporal data in SMW reusing the vocabularies of SSN and RDF data cube to model the meta data of observations, QUDT for units and quantities and the Basic Geo Vocabulary for the coordinates reference system. Using this collaborative information integration approach, we are able to include the heterogeneous sources of thermodynamic temperature values for our use case and provide meaningful, machine interpretable meta data. We have implemented a RESTful-API that can provide uni ed result sets for downstream decision support systems by exploiting the semantic descriptions of data sources and data transformation services in SMW.
Our current work focuses on employing the concept of linked APIs and smart web services that use our semantic meta data in order to dynamically compose 13 https://www.lubw.baden-wuerttemberg.de/lubw 14 http://www.dwd.de 15 http://www.aero-tram.kit.edu/ the work ows that cope with the characteristics of Big Data value streams and generate more elaborated data. As an example, the unit conversion of thermodynamic temperatures in our use case may be invoked automatically based on the meta data provided by our SMW, the unit conversion rules de ned in QUDT and a rule engine like SPARQL Protocol and RDF Query Language (SPARQL) Inferencing Notation16 that executes these rules. For the evaluation of the automatically processed data, like the transformation of temperature data of weather stations, we have also used open re ne17 with the RDF plugin18 to manually create test data. 5
Based on semantic web technology, established ontologies and linked APIs, we have presented fundamentals of a well-de ned information integration approach. Our work supplements the work introduced in Section 2 with respect to a collaborative and continuous integration of data sources and APIs that speci cally emphasis a spatio-temporal relation. Exploitation and integration of big spatiotemporal data in a new generation of GIS strongly depend on a common understanding of concepts across heterogeneous data sources. We have shown how to address this issue by combining the dynamics of an collaborative approach with the expressive power of established ontologies. For the evaluation of our approach we are going to observe experimental results in the SUHI use case that show how collaborative created descriptions of spatio-temporal data sources and data transformation services can be used to generate a machine interpretable ontology and employ the concept of smart web services for dynamically composed work ows. We believe that these preliminary results will indicate that our approach enables users even without a web engineering background to easily add sources and services for an existing GIS. For a meaningful evaluation of the research questions de ned in Section 1, we have to create more examples of dynamically composed work ows from big spatio-temporal data. The values transformed from di erent sources to a uni ed result set have to be investigated together with domain experts with respect to their comparability among each other depending on the provenance, accuracy, consistency and completeness. With an increasing number and amount of data sources and spatio-temporal values that has to be processed, we have to prove that our approach does also scale for Big Data in GIS work ows.
Acknowledgements. This work was supported by the German Ministry of Education and Research (BMBF) within the BigGIS project (Ref. 01IS14012A). 16 http://spinrdf.org/ 17 http://openrefine.org/ 18 http://refine.deri.ie/rdfExport