Ontology Concept-Based Management and Semantic Retrieval of Satellite Data

1 Introduction

Satellite data centers store, classify, and retrieve satellite data based on satellite, sensor, date, time, latitude, longitude, etc. In addition to this, application domain-specific concept-based data retrieval facilitates more domain-specific data. Domain-specific concept-based retrieval requires a semantic concept-based satellite data storage process. The concept-based retrieval system broadens the horizon of scientific application research and enhance interdisciplinary research. An effective concept-based satellite data retrieval can be achieved with the support of semantic technology.

Semantic technology is a powerful technology to represent domain knowledge in machine-understandable format [5]. For knowledge representation, semantic technology uses a powerful technique called ontology. Ontology is an explicit formal specification of a shared conceptualization [7]. Ontology provides concept definitions, hierarchies, and relationship between concepts of a domain. Ontologies enhance the performance of current information retrieval systems. Ontology provides solutions for the issues of modern information systems [15, 18, 38, 40].

Ontology-based semantic representation of satellite data enables semantic concept-based satellite data processing and retrieval. Remote sensing technology provides data pertaining to several fields such as Earth observation (EO), environment, marine, meteorology, etc. Semantic representation of satellite data requires multiple ontologies of various application domains. The effectiveness of the semantic retrieval system depends on ontology-based knowledge management structure. There is a need for an effective multiple ontology-based satellite data management methodology that facilitates improved semantic retrieval of satellite data.

The current research work describes an approach that supports ontology concept-based satellite data retrieval and three different strategies for ontology-based satellite data management. The three data management strategies are evaluated using popular real and benchmark metrics for SPARQL queries. The current work also provides a SPARQL query interface to execute ontology concept-based semantic queries on the satellite data and presents results on geographic maps.

The rest of the paper is organized as follows. Section 2 gives a discussion on background work; Section 3 illustrates the methodology; Section 4 describes performance evaluation and semantic concept-based query results; and Section 5 concludes.

2 Related Research Work

Satellite data play a vital role in many applications such as weather and forecasting, environmental monitoring, urban planning, etc. There are several satellite data receiving and distributing centers across the world. The following paragraphs describe a few popular satellite data centers and their retrieval system.

The Indian National Centre for Ocean Information Services (INCOIS) is the central repository for marine data in India. It receives voluminous oceanographic data from a variety of in situ and remote sensing observing systems. INCOIS is one of the major marine data service providers. The INCOIS search interface [9] provides data based on product, sensor, satellite, and date. INCOIS also provides data through a Live Data Access (LDA) service [8]. INCOIS LDA service provides data based on dataset, latitude, longitude, and time.

The Meteorological and Oceanographic Satellite Data Archival Center (MOSDAC) is an efficient data portal to satisfy the vast meteorological and oceanographic data needs of researchers. MOSDAC is a major meteorological and oceanographic data provider. MOSDAC provides two choices for satellite data access: metadata search [21] and advanced meta search [20]. The former searches satellite data based on the satellite, sensor, and parameter. Advanced search provides a keyword-based search for the satellite data. It performs search based on the satellite, sensor, parameter, processing level, frequency, resolution, temporal values, and keywords.

The National Remote Sensing Centre (NRSC) is one of the centers of the Indian Space Research Organization (ISRO). The NRSC/ISRO Open data and product archive [4] facilitates users to select, browse, and download data from this portal. This portal search interface retrieves data based on satellite, product, latitude, and longitude.

The National Oceanic and Atmospheric Administration [24] manages a constellation of geostationary and polar-orbiting meteorological spacecraft. It provides data through four data centers: (i) National Centers for Environmental Information (NCEI), (ii) National Oceanographic Data Center (NODC), (iii) National Climate Data Centre (NCDC), and (iv) National Geographical Data Centre. The NCEI provides quick links [23] to access data. The NODC search interface [25] provides data based on keyword, time, and content type. The NCDC retrieval system [22] provides data based on parameter, time scale, month, state, and city.

The National Snow and Ice Data Center archives and distributes digital and analog snow and ice data. The data center provides data based on data item, time, latitude, and longitude [26]. The above satellite data center retrieval systems provide data based on satellite, sensor, parameter latitude, longitude, and time.

Several research organizations and communities introduced ontologies in satellite remote sensing technology to improve the processing and utility of satellite data. Due to vast applications in diverse fields of remote sensing technology, several ontologies of remote sensing domain are developed by many space research organizations.

The Semantic Web for Earth and Environmental Terminology (SWEET) [36] is a collection of ontologies developed by the National Aeronautics and Space Administration Jet Propulsion Laboratory. SWEET developed 200 ontologies with 6000 concepts. SWEET covers a wide range of concepts and relations among the concepts in the domain of Earth and the environment [32]. Marine Metadata Interoperability [19] is a project funded by National Science Foundation. It provides several ontologies, vocabularies, and semantic services to achieve interoperability in marine data.

Many researchers are using ontologies for semantic representation of satellite images and retrieval [2, 6, 14, 37]. Little work has been done toward ontology-based semantic processing of satellite gridded data, stored in scientific file formats.

Jiapeng et al. [11] proposed an ontology-based production model of parameter products of satellite remote sensing data. The model describes the design of satellite ontology, remote sensing data ontology, and parameter products ontology. In this method, a production system architecture of land surface parameter products is designed based on the ontology model.

Jingzun et al. [12] used ontologies for remote sensing quantitative retrieval. To achieve this, the approach extends the RS-SECI (Remote Sensing – Socialization, Externalization, Combination, and Internalization) model. The method uses four ontologies: general ontology, domain-specific ontology, geospatial data ontology, and geospatial process ontology. The ontologies are used for geospatial processing to represent knowledge of the remote sensing domain. The remote sensing quantitative retrieval model defines process flow and relation between processes in the retrieval model. The retrieval model focuses on noise reduction and the production of application-specific products.

Sunitha et al. [35] presented an ontology-based semantic approach to address heterogeneity in sensor data vocabulary. The methodology reduces the heterogeneity between various sensor networks by mapping sensor vocabulary with SSN ontology.

Karpathiotaki et al. [13] presented the Prod-Trees platform and semantic-enabled search engine for EO products. The Prod-Trees platform allows the users to submit free-text queries, navigate the ontology browser, select applications terms defined in the supported ontologies, and, finally, search for EO products by specifying EO netCDF parameters and controlled (bounding box, time, range) search criteria.

The summarization of the research work in semantic-based satellite data management is as follows [11]: focuses on semantic modeling of land surface parameter products [12], uses ontologies for noise reduction and to produce application-specific data products [35], deals with heterogeneity in vocabulary of data products from various sensors and the semantic search engine [13], and provides information about an application term and its related terms. The current research describes and implements a methodology for ontology-based satellite data management and concept-based satellite data retrieval.

3 Semantic Satellite Data Management and Retrieval

The current existing systems provide effective satellite data retrieval that well executes queries on satellite data based on satellite, sensor, date, time, latitude, longitude, etc. Obtaining more application domain-specific concept-based data from massive satellite data is a challenging task. For example, wind and precipitation data can be received through the application interface of data service providers. Retrieval of more specific concept-based data like fresh gale, no wind, fresh breeze, whole gale, etc., is complex. In order to make the retrieval system more efficient and to meet scientific user needs, the retrieval system should support semantic concept-based satellite data retrieval [39].

Ontology is a good technology that supports semantic concept-based data retrieval. Ontology provides concept hierarchy and concept description. Figure 1 shows the concept hierarchy of weather ontology.

Figure 1:

Weather Ontology Hierarchy.

The Resource Description Framework (RDF) [30] is one of the techniques to represent domain knowledge and store instances of an application. The application knowledge can be represented in the form of ontology concepts by defining concepts, concept hierarchy, and relationships between the concepts. This knowledge stored in RDF is known as T-BOX. The application data/instances stored in RDF is known as A-BOX [34].

Queries on A-BOX data provide data pertaining to the low-level concepts fresh gale, no wind, fresh breeze, whole gale, etc. T-BOX can address concept, concept hierarchies, and inference-related queries on application domain knowledge. The support of T-BOX enables the execution of refined semantic queries on A-BOX.

The current research work focuses on ontology concept-based satellite data management and retrieval. Remote sensing is a multidimensional and multirealm application area. Ontology concept-based satellite data retrieval requires multiple ontologies. For example, the weather and climate application area has climate and forecast parameter, weather, precipitation, etc., ontologies. There is a need to select appropriate ontologies based on the satellite data. To achieve effective and efficient semantic concept-based satellite data retrieval, there is a need for

• RDF representation of ontologies (T-BOX);

• Transforming satellite data as instances of appropriate ontology concepts (Building A-BOX);

• Ontology-based satellite data management.

3.1 Building A-BOX

Data received from different satellites are stored in different scientific file formats such as HDF, NetCDF, GRIB, etc. To achieve semantic concept-based satellite data retrieval, there is a need to link satellite data from the scientific file formats to the ontology concepts. Linking can be done by taking the support of concept definitions in the ontology. The results are represented as instances to the ontology concepts. Figure 2 shows instance creation using an ontology concept definition.

Figure 2:

Instance Creation from Satellite Data Using Ontology Concept Definition.

As shown in the Figure 2, the ontology concept “FreshGale” is defined as a wind speed between 17.1 and 20.7 m/s; that is, an instance of the concept “FreshGale”.

3.2 Ontology-Based Satellite Data Management

The RDF representation of ontology and satellite data instances builds T-BOX and A-BOX. The semantic data retrieval efficiency depends on the strategies used to manage ontologies and instance data (T-BOX and A-BOX) in RDF triple stores. RDF triples can be stored in native triple store or in relational databases in triple format. The native RDF store uses a file system to store triples, and the database RDF stores use relational or object relational databases as the backend store [10] to store RDF triples. Database RDF stores such as MySQL, Oracle, PostgreSQL, etc., create model tables to store RDF triples. A model table deals with a set of RDF triples.

A native triple store is more straightforward than triples stored on relational databases. However, database management systems have many significant features such as performance, robustness, reliability, and availability. Thus, the relational database is a very good solution for storing and querying RDF triples [1]. Thus, database triple stores are an effective method to store RDF triples. The current research work adopts database triple stores for ontology-based satellite data management.

The ontologies and instances can be managed in three difference styles in database RDF triple stores. The current section discusses and evaluates these strategies to manage ontologies and satellite data instances. Inference is a mechanism that derives new triples from specified model tables. Inference plays an important role in semantic query performance. Entailment is an object that stores newly derived triples. The inference engine needs a rule set to derive new triples. The rule set can be either predefined or user defined. Several predefined rule sets are available in the literature to perform inference based on the properties transitive, reflexive, sameAs, subClassOf, inverseOf, etc. The three ontology-based satellite data management strategies execute inference on T-BOX and A-BOX using a predefined rule set.

3.2.1 Strategy 1

In this approach, a single model table is used to store ontologies and satellite data instances (T-BOX and A-BOX) in an RDF triple store, as shown in Figure 3. Inference engine derives new triples from the model table and stores an entailment object in an RDF triple store. The limitation with this approach is that modification to T-BOX data causes loss of A-BOX data.

Figure 3:

Strategy 1: Ontology Concept-Based Satellite Data Management.

3.2.2 Strategy 2

To overcome the limitation of strategy 1, strategy 2 adopts a dual model table that separates A-BOX and T-BOX data as shown in Figure 4. An inference engine executes on two model tables. Strategy 2 overcomes the limitations of strategy 1; however, it is difficult to maintain:

• Changes in an ontology leads to entire T-BOX processing.

• Deletion of an ontology leads to loss of entire T-BOX data.

Figure 4:

Strategy 2: Ontology Concept-Based Satellite Data Management.

3.2.3 Strategy 3

To overcome the limitations of strategy 2, strategy 3 adopts a separate model table for every ontology of a domain and a separate model table for A-BOX as shown in Figure 5. As the number of ontologies that are used in any application is not predefined, a dynamic T-BOX data management technique is required to achieve flexibility and enhanceability. The current research work creates model tables at runtime and maintains all model table details of T-BOX in a log table. An inference engine executes on A-BOX data and multiple T-boxes data with the support of a log table.

Figure 5:

Strategy 3: Ontology Concept-Based Satellite Data Management.

3.3 Semantic Retrieval and Visualization

Querying A-BOX data enables concept-based satellite data retrieval. T-BOX improves the effectiveness of the retrieval system. Using inference, the retrieval system can execute concept hierarchy-based queries.

SPARQL is a formal RDF query language. The current research work uses SPARQL query language to execute semantic queries on the knowledge base (T-BOX, A-BOX, and inferred data). The current research work facilitates the execution of semantic queries on the satellite data knowledge base using SPARQL query language.

The SPARQL query engine return results either in xml, text, csv, JSON, or tsv formats [33]. These formats do not enforce readability and understandability. SPARQL results need to be presented in a user-understandable format that supports effective analysis. The current research work plots SPARQL query results on geographic maps. SPARQL query visualization using geographic maps is done in two stages:

• SPARQL query result preprocess;

• Plot data on geographic maps.

Figure 6 shows the semantic query execution and visualization process.

Figure 6:

Semantic Query Execution and Visualization.

3.3.1 SPARQL Query Result Preprocess

To plot the results of semantic query on geographic maps, the latitude and longitude information is required. The SPARQL queries should contain retrieval needs along with latitude and longitude data. The SPARQL query results in conceptual data along with latitude and longitude data in string format associated with the respective Uniform Resource Identifiers (URIs), as shown in Table 1. The preprocess operation removes the URIs and converts the string representation of latitude and longitude into numeric. This numeric format is used further to plot the data on geographic maps.

Table 1

Sample Results of SPARQL Query Represents Latitude and Longitude.

Lat	Lon	– – – –
“1.232E1” <http://www.w3.org/2001/XMLSchema#double>	“7.3900000000000006E1” <http://www.w3.org/2001/XMLSchema#double>	– – – –
“-2.075E1” <http://www.w3.org/2001/XMLSchema#double>	“7.3900000000000006E1” <http://www.w3.org/2001/XMLSchema#double>	– – – –

3.3.2 Plotting Data on Geographic Maps

Plotting data on a map is done in two levels: display maps and geographic maps. Appropriate tools or libraries are used to display maps and plot concept-based data on these maps. The current work on the visualization process uses the R tool [28].

4 Performance Evaluation

The current research work adopts oracle semantic store [16], Apache Jena API [3], and OWLPrime [27] technologies for implementation. The three ontology-based satellite data management strategies are evaluated using real and benchmark metrics through SPARQL queries.

There are several real and benchmark datasets and respective SPARQL queries available in the literature to evaluate RDF storage and retrieval performances. Various real and benchmark data sets along with the benchmark generator algorithm are described in Ref. [31]. DBpedia [29] is a real data set, and BSBM and SP2 [17] are popular and widely accepted as benchmark data sets.

Various performance metrics benchmarks are defined in the state of art. The BSBM performance metrics are Query Mixes per Hour (QMpH), Queries per Second (QpS), and Load Time (LT). The SP2 performance metrics are Arithmetic Mean (AM) and Geometric Mean (GM) of the elapsed time of SP2 benchmark queries. DBpedia measures Query Processing Time (QPT) of individual queries.

The three ontology-based satellite data management strategies are evaluated to measure the performance of the semantic retrieval system. The evaluation process takes three ontologies and respective satellite data of different sizes. The ontologies considered for performance evaluation are weather, climate, forecast, and environment. The three different sizes of data, 1L, 5L, and 20L, are taken as test bed, where L=lakhs triples.

To evaluate the performance of data management strategies with respect to concept-based satellite data retrieval, a query set of 25 concept-based semantic queries has been chosen. This addresses various user requirements and involves various SPARQL query constructs. Table 2 gives the semantic queries and their SPARQL representation. Table 3 shows the mapping of various SPARQL query constructs to semantic queries.

The test is done in a Windows 8 desktop system of memory 12 GB and processor 3.4 GHz. The evaluation report is presented below.

4.1 Performance Evaluation Results

4.1.1 BSBM Metrics

QMpH and QpS metrics have been used to measure performance. A BSBM test drive is executed with 20 warm-ups and 50 runs on the three strategies. It is observed that strategy 3 improves the overall QMpH, as shown in Figure 7.

Figure 7:

QMpH Analysis of Three Strategies.

QpS analysis results indicate that strategy three results in better performance for large-size A-BOX satellite data. The results are shown in Figures 8–10 for various sizes of triple stores.

Figure 8:

QpS Analysis of Three Strategies on Three Ontologies and A-BOX of Size 1L.

Figure 9:

QpS Analysis of Three Strategies on Three Ontologies and A-BOX of Size 5L.

Figure 10:

QpS Analysis of Three Strategies on Three Ontologies and A-BOX of Size 20L.

4.1.2 SP2 Metrics

The SP2 performance metrics are AM and GM of QPT. In the current evaluation process, the AM and GM are measured in milliseconds. The test results in Figures 11 and 12 indicate that strategy 3 improves query performance.

Figure 11:

AM Analysis of Three Strategies.

Figure 12:

GM Analysis of Three Strategies.

4.1.3 DBpedia

The DBpedia performance metric is QPT in milliseconds of individual queries. It is observed that strategy 3 shows improved results. Figures 13–15 show the evaluation results.

Figure 13:

QPT Analysis of Three Strategies on Three Ontologies and A-BOX of Size 1L.

Figure 14:

QPT Analysis of Three Strategies on Three Ontologies and A-BOX of Size 5L.

Figure 15:

QPT Analysis of Three Strategies on Three Ontologies and A-BOX of Size 20L.

4.2 Semantic Query Execution and Visualization

The current work developed a SPARQL query interface using Java. Figure 16 shows a SPARQL query interface. SPARQL query execution on oracle semantic store is implemented using Apache Jena. The SPARQL query interface enables the user to write a SPARQL query. The interface also gives the option of choosing the shape files on which the results have to be plotted.

Figure 16:

SPARQL Query Interface.

An ontology concept-based satellite data retrieval and visualization process and sample query results are shown below. The following are the sample semantic concept-based queries and their SPARQL representation. Figures 17–20 show the results of Q1, Q2, Q3, and Q4.

Figure 17:

Q1 Results.

Figure 18:

Q2 Results.

Figure 19:

Q3 Results.

Figure 20:

Q4 Results.

Q1. Locate fresh gale

PREFIX we: <http://www.isro.org/Weather#>

select ?lat ?log where {?s rdf:type we:FreshGale; we:hasLatitude ?lat; we:hasLongitude ?log}

Q2. Locate strong breeze

PREFIX we: <http://www.isro.org/Weather#>

select ?lat ?log where {?s rdf:type we:StrongBreeze; we:hasLatitude ?lat; we:hasLongitude ?log}

Q3. Locate storm

PREFIX we: <http://www.isro.org/Weather#>

select ?lat ?log where {?s rdf:type we:Storm; we:hasLatitude ?lat; we:hasLongitude ?log}

Q4. Locate catastrophic hurricane

PREFIX we: <http://www.isro.org/Weather#>

select ?lat ?log where {?s rdf:type we:CatastrophicHurricane; we:hasLatitude ?lat; we:hasLongitude ?log}

5 Conclusion

The applications of satellite data in various domains and their advancements in the technical world are delivering advanced solutions to the societal applications. Semantic technologies are playing a vital role in knowledge management and information retrieval performance, and delivering refined results to the various communities. Bringing semantic technology into satellite data processing achieves an effective concept-based retrieval system that delivers concept-based results to the users. An ontology concept-based satellite data retrieval system provides more enhanced results to the scientific community that enables application-oriented and multidisciplinary research. The current research work integrates ontology-based semantic technology with satellite data processing techniques. The research work discussed ontology concept-based satellite data management strategies. The performance of the semantic query retrieval system is evaluated by taking real-time data from MOSDAC and semantic queries related to weather, climate, forecast, and environment ontologies. Further research could be conducted on optimization of semantic query processing on satellite data.