The Sen2Cube.at national semantic Earth observation data cube for Austria
<div> <p>The Sen2Cube.at is a Sentinel-2&#160;semantic&#160;Earth observation (EO)&#160;data and information cube that combines an EO data cube with an AI-based inference engine by integrating&#160;a&#160;computer-vision&#160;approach to&#160;infer new&#160;information.&#160;Our approach uses semantic enrichment of optical images and makes the data and information directly available&#160;and accessible&#160;for further use within an EO data cube.&#160;The architecture is based on an expert system, in which domain-knowledge&#160;can be&#160;encoded&#160;in&#160;semantic models (knowledgebase) and applied to the Sentinel-2 data as well as&#160;semantically enriched,&#160;data-derived information (factbase).&#160;&#160;</p> </div><div> <p>The initial semantic enrichment in the Sen2Cube.at system is general-purpose, user- and application-independent, derived directly from optical EO images as an initial step towards a scene classification map. These information layers are automatically generated from Sentinel-2 images with the SIAM software (Satellite Image Automated Mapper).&#160;SIAM is a knowledge-based and physical-model-based decision tree that produces a set of information layers in a fully automated process that is&#160;applicable&#160;worldwide&#160;and does not require any samples.&#160;A&#160;graphical inference engine allows application-specific Web-based semantic querying&#160;based on the generic information layer&#160;as&#160;a&#160;replicable and explainable&#160;approach to produce information.&#160;The graphical inference engine is a&#160;new&#160;Browser-based graphical user interface&#160;(GUI)&#160;developed in-house&#160;with a semantic querying&#160;language.&#160;Users formulate semantic models in a graphical way and can execute them on any area-of-interest and time interval, which will be evaluated by the core of the inference engine attached to the data cube.&#160;This&#160;also&#160;enables&#160;non-expert users&#160;to&#160;formulate analyses&#160;without&#160;requiring&#160;programming skills.&#160;&#160;</p> </div><div> <p>While the methodology is software-independent, the prototype is based on the Open Data Cube and&#160;additional&#160;in-house developed components&#160;in&#160;the Python&#160;programming language.&#160;Scaling is possible depending on the available infrastructure&#160;resources&#160;due&#160;to the system&#8217;s Docker-based container architecture.&#160;Through its fully automated semantic enrichment, innovative graphical querying language&#160;in the GUI&#160;for semantic querying and analysis as well as the implementation as&#160;a&#160;scalable infrastructure,&#160;this approach is suited for big data analysis of Earth observation data. It&#160;was successfully scaled to a national data cube for Austria, containing all available Sentinel-2 images from&#160;the&#160;platforms&#160;A&#160;and&#160;B.&#160;</p> </div>