Geospatial information is indispensable for various spatially-informed analysis and decision-making, e.g. traffic analysis and built environment processes. Geospatial data often must be integrated for meaningful analysis, whereas such integration is challenging due to siloed data organization, semantic heterogeneity and multiple representation of geospatial data. Moreover, the visualization of geospatial data is one of the most prominent ways of utilizing geospatial data, however how to properly visualize the data is sometime difficult, as it pertains to a wide range of visualization (cartographic) knowledge. Semantic Web technologies unveil a promising way to mitigate these issues, as they provide means of data integration on the Web, and knowledge representation capacity to formally represent the visualization knowledge. In this PhD project, we investigate the potential values of Semantic Web technologies for geospatial data integration (particularly for geospatial data with multiple representation) and visualization in several cases, where the integration and visualization knowledge is formalized using Semantic Web technologies. All the case studies embody realworld meaning and entail data integration and visualization challenge, which have been addressed by state-of-the-art solutions inadequately. Preliminary results demonstrate great yet not fully unlocked potential of Semantic Web technologies for geospatial data, and also disclose challenges that need to be addressed.
geographically. Currently, geospatial data is maintained and delivered mainly by spatial
data infrastructures (SDIs). However, the data in SDIs is inadequately integrated and
harmonized, particularly the integration of geospatial data and other types of data is
rare. Geospatial data integration is complex, and a prominent intricacy is, among others,
the multiple (geometric) representations of geospatial data, which is a specific data
integration problem in the geospatial domain [
Moreover, the knowledge concerning how to appropriately use geospatial data is important. There have been many endeavors for geospatial knowledge formalization, whereas today experts from other domains still often have to look into the literature, or cooperate with geospatial experts to accomplish meaningful use of geospatial data. Visualization, as one of the most predominating ways of utilizing geospatial data, also entails much semantic intricacies, as visualizing geospatial data in a sensemaking and cartographically satisfactory way pertains to a wide range of cartographic knowledge, which is hard to transfer, interpret, and reuse, especially by non-geospatial experts.
The increasing appreciation of Semantic Web technologies in the geospatial domain
[
Therefore, this PhD project investigates the potential of Semantic Web technologies for geospatial data integration (particularly the handling of multiple representations), and formalizing geospatial (visualization) knowledge for knowledge outreach on the Web. 2
Geospatial information plays an indispensable role in a vast number of spatial analysis and spatially-informed decision making, thus sharing, integrating geospatial data on the Web are important, so is the outreach of geospatial knowledge.
From another perspective, Semantic Web technologies (especially the parts
concerning linked data and ontologies) are increasingly adopted and applied in the geospatial
domain. A recent survey conducted in 2018 by EuroSDR (European Spatial Data
Research) demonstrated that linked data has been seen as one of the most important
research issues and major movers toward future SDI [
The application of Semantic Web technologies has developed considerably in
geospatial domain in the last decade, as they address several long-standing challenges of e.g.
data integration, semantic interoperability and knowledge formalization, and provide a
promising way to connect SDIs with the mainstream IT to augment the application of
geospatial data [
Increasing geospatial data has been published or planned to be delivered as linked
data; this trend is particularly prominent for authoritative geospatial data. For instance,
Ordnance Survey, the national mapping agency (NMA) in the UK, has released several
geospatial datasets maintained by them as linked data [
Semantic Web and linked data are used for geospatial data integration to (partially)
resolve semantic heterogeneity of multi-source data and consolidating distributed
information. Such work has been accomplished mostly in the environment of SDIs. For
1 https://kcopendata.eu/sdi2030/
2 https://www.geonames.org/
3 https://www.stardog.com/
4 https://virtuoso.openlinksw.com/
5 https://inspire.ec.europa.eu/news/linking-inspire-data-draft-guidelines-and-pilots
instance, Janowicz et al. [
The capacity of knowledge representation of Semantic Web leveraging ontologies
and rules has been recognized in the geospatial domain for many years and used in a
number of studies. These studies span several research subjects of e.g. visualization,
geo-processing and information retrieval. For instance, Hofer et al. [
With regard to the visualization of geospatial data, Scheider and Huisjes [
The overall research question is what are the benefits of Semantic Web technologies for geospatial data integration and visualization?
Under this hood, we formulate several specific research questions focusing on realworld problems that can potentially better addressed by Semantic Web technologies: 1) Geospatial data is often repetitively generated despite relations between the objects. Is it possible to link geospatial objects to existing objects in the Semantic Web to diminish data repetition and inconsistency? 2) The knowledge concerning how to visualize geospatial data is important. Is it possible to use Semantic Web technologies to formalize such knowledge, and thus share it on the Web? 3) Multiple representation of geospatial data sometimes renders data integration complex and problematic. Is it possible to leverage Semantic Web technologies to formalize the knowledge of multiple representation and assist cross-detailed-level data integration? 4) Geospatial data interlinking is imperative to further unlock the potential of the Semantic Web. Therefore, it is relevant and important to investigate how to advance geospatial data interlinking. 5
In order to answer the research questions, we formulated a set of hypotheses that are used to operationalize the work, including: 1) Using linked data and ontologies can facilitate the multi-source geospatial data integration on the Web, especially instead of repetitively generating multi-source geospatial data, one could link the data to reference data (e.g. authoritative geospatial data from NMAs) to obtain more accurate location information for better visualization and analysis. 2) Coupling ontologies and semantic rules can (partially) formalize the geospatial data visualization knowledge into knowledge bases, thereby enable semantic reasoning to derive proper visualization means for the data, which can make the visualizations appropriate tools for decision making. 3) Combining ontologies and semantic constraints (SHACL) can represent complex and subtle semantic relations raised by multiple representation of geospatial data, and thus facilitate the use of geospatial data in other domains that perceive the geographic space differently. 4) For automating geospatial data integration at instance level, the knowledge graph embedding technique is useful, but geometric (location) information is also important. Thus, combining geometric information with knowledge graph embedding technique can help geospatial data integration and interlinking. 6
As geospatial information can be used in various real-world applications, the value of the data integration and visualization can be revealed in solving real-world or even long-standing problems that have not been solved before or have been inadequately solved. Therefore, the approach for addressing the research questions and testing the hypotheses is mainly case studies, i.e. spatially-informed analysis or decision making. The evaluations are/will be mainly performed by comparing Semantic Web-based solutions to traditional solutions.
Specially, to test hypothesis 1, we use the case of web maps for natural reserved areas, as this type of geospatial objects often have intrinsic connections with other geospatial objects, whereas state-of-the-art approach of data modelling neglects such relations. Linked data can be used to relatively position the natural reserved areas to reference objects (e.g. roads, rivers, cadastres).
To test hypothesis 2, we use the case of heritage building protection mapping, where we use ontologies and SPIN (SPARQL) rules to formalize the visualization knowledge and distributed linked data retrieval. The rationale of using SPIN rules rather than e.g. SWRL rules is that it is often that the visualization rules include non-monotonic semantics, e.g. a rule stating that render the object in a certain way if the value of an attribute does not exist (closed world assumption). Also, SPIN rules have a formalized vocabulary and can be more readily shared on the Web.
To test hypothesis 3, we use a case study of evaluating urban infrastructure’s suitability for bicycling, where we utilise SHACL constraints to represent subtle and complex semantic relations raised by multiple representations between the geospatial domain and the traffic domain. With SHCAL constraints, the knowledge concerning using which level of representations for which scenarios can be explicated and formally represented. Such formalized knowledge can guide cross-detailed-level data integration and also facilitate wide-use of such knowledge.
To test hypothesis 5, we plan to utilise the geospatial data available in the LOD cloud, and also we will generate geospatial data with multiple representations from authoritative geospatial datasets, and compare the method with state-of-the-art methods for linked data interlinking, and object matching methods in the geospatial domain. 7
To validate hypothesis 1, we developed a relative positioning approach based on linked
data and ontologies. That is, instead of absolutely positioning all the geospatial features
repetitively, we relatively position geospatial (thematic) features (objects) to
background data (i.e. geospatial data with multiple representations from Swedish mapping
agency). Ontologies were designed for storing the relative positioning information,
linked data was used to link the relatively positioned features to reference features. This
approach accomplished self-adapting web maps for better visualization performance,
which had seldom been addresses by other methods [
To validate hypothesis 2, we designed a knowledge base encapsulating ontologies
and semantic rules (SPIN rules) to represent the knowledge concerning cartographic
scale, data portrayal, and geometry source. The approach accomplished visualizing
distributed and multi-scale geospatial data in a cartographically satisfactory way, which
can be hardly implemented using current OGC technology stack [
To validate hypothesis 3, we employ the study case of evaluating the suitability of urban road network infrastructure for bicycling, in which geospatial data needs to be integrated with field-collected data. Such integration is complex, as the traffic researchers (who develop indexes for the suitability evaluation) perceive the geographic space differently than the modelling of geospatial data. The indexes treat the junctions in the road network as a whole (using a single point feature to represent a junction), this corresponds to the data modelling approach of the less detailed geospatial road network. However, the indexes need the dedicated bicycling paths information, which is only available in the most detailed geospatial road network (in the most detailed road network, the junctions are modelled with detailed structure, mainly including polylines and points). Such cross-detailed-level and cross-domain data integration cannot be implemented merely using ontologies, thus we impose SHACL constraints for this type of data integration. The constraints ensure the semantic correctness of utilising data from different detailed levels.
In addition, we investigate some popular and well-known RDF stores, i.e. RDF4J, Jena, Stardog, Virtuoso and GraphDB for their geospatial query capacity, particularly focusing on GeoSPARQL-compliance and query performance. This is important as it gives insights concerning where to deploy the proposed approach. The assessment and benchmarking are conducted in two scenarios. In the first scenario, geospatial data comprises a part of a large scale data infrastructure and is integrated with other types of data. In the second scenario, we benchmark the RDF stores in a dedicated SDI environment with purely geospatial data. The results show that GeoSPARQL-compliance has considerably developed with reasonable query efficiency, while query correctness still remains a challenge, as different stores sometimes return different results for the same query. 8
To date, we have collected positive evidence showing that Semantic Web technologies have great and yet not fully unlocked potential benefits for geospatial data integration and visualization. The supervision team of this PhD project is mainly from the geospatial domain, thus we are familiar with the real need and challenges that are potentially could be better solved with Semantic Web technologies, and we have tight connections with the authorities that are interested in employing Semantic Web technologies for geospatial data, e.g. Swedish mapping agency, Swedish Geological Survey, Swedish Traffic Administration, etc. This project is also conducted closely cooperating with experts from other domains that are in need of geospatial data and knowledge, e.g. traffic researchers. Furthermore, we have close collaboration with Semantic Web researchers with theoretical and technical advice. In summary, this project has good connections and background knowledge to investigate the research questions. Also, the values revealed from the case studies embody real-world usefulness of Semantic Web technologies, and this will potentially draw more extensive attention from various domains.
Despite the promising results, there are still several challenges, e.g. the geospatial data interlinking on the Web still remains a challenging and sometimes expensive task, while it is imperative for unlocking the values of Semantic Web for geospatial data. We plan to address this issue in the next step. This work will benefit from both the advancements in the Semantic Web (e.g. knowledge graph embedding technique), and the outcomes of geospatial feature matching that has been studied for decades.
This PhD project is under the supervision of Prof. Lars Harrie and Dr. Ali Mansourian at Lund University.