|
Dear Φ-lab community,
Traditional geospatial data analysis is struggling. We are drowning in “big data” that is high-volume, multi-resolution, multimodal and temporally varied – yet, critically sparse in some regions of our globe.
On one side, we have specialised AI models, each built for a single task. They perform well in narrow domains (Narrow AI) but fail to scale across this complexity.
On the other side, Geospatial Foundation Models act like the ‘swiss army knives’ of Earth observation and Earth sciences. They excel at understanding diverse data and generalising across tasks. But here’s the catch: they don’t inherently know how to solve complex problems that require the orchestration of multiple models and tools, neither think like a human would.
A solution? Agentic reasoning. These agents can check intermediate results, identify if the output from one model is insufficient, and refine the input parameters or choose a different tool to achieve a more accurate final answer – minimising the risk of errors and hallucinations.
In short, the AI models are the powerful tools, and agentic reasoning is the intelligent human-like analyst that knows when, where, and how to use those tools to solve a problem at massive speed and scale. This synergetic relationship enables both retrospective investigation and proactive planning for complex scenarios on Earth.
In Europe, there are many parallel, independent platforms used to analyse and better understand our planet. Take TerraMind, a ‘made-in-Europe’ geospatial foundation model developed in a collaboration between ESA Φ-lab and IBM Research Europe. TerraMind was found to be the one of the best-performing AI foundation models for Earth observation, as assessed by the PANGAEA benchmark.
However, we are missing a more ambitious strategy to bring AI foundation models and agentic reasoning together. As of today, there isn’t such an initiative to ‘glue’ them within the European ecosystem.
Or there wasn’t, until EVE come along. EVE (Earth Virtual Expert) is set to change the paradigm in how we understand and work with Earth observation and Earth sciences data. It is a new intelligent companion, developed in a first-of-its-kind partnership between ESA Φ-lab, Wiley and Mistral AI.
Its first version focuses exclusively on textual information, providing all the essential knowledge and context needed to serve as the glue for future integration with models, tools, and reasoning engines.
Watch this space for more news about EVE, as ESA Φ-lab is set to start the new year with lots of fresh news!
Wishing you all a very Merry Christmas and an excellent 2026,
The Φ-lab team
|
