Back in March I attended the latest workshop in the Location Powers series organised by the Open Geospatial Consortium, OGC, with which W3C has an excellent and fruitful relationship. The theme for the day was Geospatial Linked Data and speakers included BDE partner Manolis Koubarakis of the University of Athens who talked about his work on using satellite data to monitor the effects of wild fires in Europe. It is that work that created the Strabon triple store that is one of the components in the BDE Integrator platform and that is used in the Societal Challenge 7 pilot on using satellite imagery to detect change on the ground.
What struck me particularly throughout the day was the ubiquity of the problems faced when dealing with large amounts of data: the focus always has to be on efficiency and performance. In the BDE Integrator platform, a semantic layer allows a SPARQL query to be executed across all the available data, irrespective of format, as the Ontario feature effectively creates a virtual graph at query time. This means that only the specific pieces of data needed to answer the query are examined, offering substantial performance increase over the more obvious method of converting all data to triples (or any other single format) at ingestion. Manolis described a very similar approach using Ontop and its spatial extension.
There was also a lively discussion about the need for being able to handle RDF triples not as immutable facts but as statements that are true for a defined period of time and/or with a probability of being true. That’s the kind of extra dimension needed for machine learning, AI and analytics engines – like the SANSA stack available in the BDE Integrator Platform.