Rethinking Interoperability for AI-Driven Earth Intelligence: what AI-ready means for biodiversity from space

Slides of the talk given at ESA Frontiers of Science, 30 September 2026 (19 slides, references and backup slides). AI changes how and when we combine Earth Observation with in-situ data: an AI agent can convert, regrid and interpret data at every query, but every query pays again, in cost and in errors that vary from run to run. The talk asks what should be built into the data once, at the source, and what can be left to AI each time. It walks through five layers of interoperability (technical, semantic, organisational, legal and knowledge), using biodiversity as the hardest test case: a common grid (HEALPix on the WGS84 ellipsoid, GRID4EARTH), declared meaning (EOSC Interoperability Framework, I-ADOPT, CF conventions), who converts, keeps and pays, machine-actionable access for sensitive species locations (ODRL with encryption), and findings and needs recorded as nanopublications, published and queried through the Science Live platform (platform.sciencelive4all.org). Case studies: ESA BIOMASS, CCI Biomass, Fire CCI and GBIF in the Beni (Bolivian Amazon); Copernicus Marine wave products at marine Natura 2000 sites; weather radar and bird migration. It proposes a definition of AI-ready data, packaged as FAIR Digital Objects, and a loop in which recorded gaps and needs inform the next in-situ networks and missions. Code and data for every figure: doi:10.5281/zenodo.23002390 (annefou.github.io/esa-frontiers-figures).