rasdaman newsletter 06/2026
How Model Fencing Powers AI-Cubes
At EGU26 in Vienna this year, Peter Baumann delivered a course on how the combination of AI, datacubes and open standards can simplify, secure and accelerate Geo AI.
Baumann presented the latest developments in combining AI, datacubes and open standards. The focus was on making Big Earth Data easier, safer, and more powerful to use — particularly for those who want to analyse complex Earth Observation data without delving deeply into technical pre-processing, metadata management, or model integration.
In his session titled 'AI on Spatio-Temporal Data: Why Does It Have to Be So Complicated?” Baumann discussed the main challenges of applying machine learning models to spatio-temporal data. While many models are available today, their practical application can be challenging: data must be properly prepared, metadata is not always adequately described, and the limitations of a model can be unclear.
This is where the concepts of AI-Cubes and Model Fencing come into play. AI-Cubes allow machine learning (ML) inference to be embedded directly into WCPS Datacube queries. Model Fencing, developed in the FAIRgeo project, helps the server recognise when a model is being used outside its reliable 'comfort zone'. Having server-side checks is of critical importance when AI inference is deeply embedded in automated workflows, without any human monitoring correctness. Also guarding against wrong use when moving towards zero-coding non-expert use of geo analytics requires extra server-side safeguarding. The presentation used examples to demonstrate how standards-based, zero-coding Agentic AI can become more realistic and trustworthy in the future.
Additionally, the poster presentation entitled 'Federated AI-Cubes: Towards Democratising Big Earth Datacube Analytics', attracted interest. It showcased the potential of Federated AI-Cubes in scalable and reproducible workflows within the field of Earth system sciences.
Federated AI-Cubes play a pivotal role in the Skyfed project, connecting cloud-, edge-, and satellite-based information spaces. The presentations emphasised that, in the future, trustworthy, federated and reproducible Earth data infrastructures will bring research and applications closer together. Both the SkyFed and FAIRgeo (Fencing AI for Enhanced Reliability in Geo Services) projects are supported by the European Regional Development Fund (ERDF) and the state of Bremen.
contact: Prof. Dr. Peter Baumann