NASA Logo

NTRS

NTRS - NASA Technical Reports Server

Press Enter or click the Search button to begin your search.

Back to Results
From Pretraining to Practice: Highlights From the 2026 ESA-NASA International Workshop on AI Foundation Models for Earth ObservationThe second ESA-NASA International Workshop on AI Foundation Models for Earth Observation brought together the Earth observation and AI communities from academic, commercial, and governmental organizations to assess the rapidly evolving role of foundation models in geospatial science applications. Hosted by NASA Marshall Space Flight Center in May of 2026, the workshop built on the inaugural meeting by shifting focus from early alignment and prototyping toward how geospatial foundation models should be pretrained, evaluated, adapted, trusted, deployed, and governed.

Across the workshop, a clear theme emerged: the community has moved beyond questioning whether large models can be pretrained on satellite data. Instead, discussions centered on what pretraining should teach, what information current models fail to capture, how model outputs can be trusted, and when foundation model approaches provide practical value over conventional methods. During sessions, attendees repeatedly emphasized that no single pretraining recipe, model architecture, benchmark, or deployment strategy fits all downstream tasks, making data quality, curation, uncertainty, evaluation design, operational constraints, and user-centered validation just as critical to scale.

The workshop also highlighted a transition from standalone foundation models toward broader geospatial AI systems. Open-source tools, scalable serving approaches, embedding products, onboard inference, operational decision-support applications, and agentic AI workflows are beginning to connect these models with data archives, benchmarks, scientific literature, code, and human expertise. This report summarizes pivotal discussions from the workshop and outlines future directions for benchmarking, embeddings, adaptation, operationalization, trust, agentic AI, and the unique role of public agencies in supporting an open, reliable geospatial AI ecosystem.
Document ID
20260007806
Acquisition Source
Marshall Space Flight Center
Document Type
Conference Paper
Authors
Hamed Alemohammad
(Clark University Worcester, United States)
Mina Burns
(Oregon State University Corvallis, United States)
Ruben Cartuyvels
(European Space Agency (ESA))
Nidhi Jha
(University of Alabama in Huntsville Huntsville, United States)
Robert E Kennedy
(Oregon State University Corvallis, United States)
Nicolas Longepe
(European Space Agency (ESA))
Valerio Marsocci
(European Space Agency (ESA))
Rahul Ramachandran
(Marshall Space Flight Center Redstone Arsenal, United States)
Sujit Roy
(University of Alabama in Huntsville Huntsville, United States)
Campbell Watson
(IBM Research - Thomas J. Watson Research Center Yorktown Heights, United States)
Isabelle Wittmann
(IBM Research - Zurich Rüschlikon, Switzerland)
Date Acquired
August 13, 2026
Publication Date
August 27, 2026
Publication Information
Subject Category
Cybernetics, Artificial Intelligence and Robotics
Meeting Information
Meeting: 2nd ESA-NASA Workshop on AI Foundation Model for Earth Observation
Location: Huntsville, AL
Country: US
Start Date: May 19, 2026
End Date: May 22, 2026
Sponsors: National Aeronautics and Space Administration
Funding Number(s)
CONTRACT_GRANT: 80MSFC22M0004
Distribution Limits
Public
Copyright
Portions of document may include copyright protected material.
Technical Review
Single Expert
Keywords
Earth observation
foundation models
geospatial artificial intelligence
benchmarking
embeddings
agentic AI
operational deployment
Earth system science
No Preview Available