The ESA-NASA International Workshop on AI Foundation Model for Earth Observations (EO) was held at ESA-ESRIN in Frascati, Italy from May 5–7. This was a dynamic gathering focused on recent advances in foundation models for EO (EO FMs). Clark CGA’s team participated in the workshop, presented results of our research on explainable AI methods for EO FMs, and moderated multiple sessions during the week.

The workshop brought together experts working on the development and application of these models. NASA’s team presented its Prithvi family of models for EO and weather applications, while ESA and IBM introduced TerraMind, a new model featuring an innovative pretext task that showed promising results on Sentinel satellite products.

A major theme across the sessions was the direct use, fusion, and interpretation of EO FM embedding outputs. Presenters explored how to design models that effectively address the multi-modal nature of remote sensing by leveraging data from diverse imaging sensors, particularly optical, thermal, and SAR. The workshop also highlighted growing efforts to create massive datasets to support self-supervised training, semantic search, and inference directly within learned embedding spaces.

“It was a pleasure to collaborate with our NASA and ESA colleagues to organize this flagship workshop, and join other experts to discuss the latest advancements in this field,” says Dr. Hamed Alemohammad, Director of Clark CGA, who presented at the workshop. “The workshop provided a forum to both engage in technical discussions with other experts, and explore joint opportunities to bring the impact of advancements in EO FMs to solve societal problems”.

“Beyond networking opportunities, it was exciting to see how the field of EO FMs is evolving: there is a clear shift toward more sophisticated handling of multi-modal remote sensing data and a growing interest in how to effectively leverage the embedding space of foundation models,” says Dr. Sam Khallaghi, Postdoctoral Researcher at Clark CGA who attended the workshop. “These developments offered new perspectives that will directly inform the self-supervised learning section of a book chapter I’m currently working on.”