NASA's Prithvi Becomes First AI Geospatial Model Deployed in Orbit
Researchers from Adelaide University and the SmartSat Cooperative Research Center successfully uploaded NASA and IBM's Prithvi geospatial AI model to two orbiting platforms-the Kanyini satellite and a payload aboard the International Space Station. This marks the first time a foundation model trained on Earth observation data has operated in space.
The team tested Prithvi's ability to detect floods and clouds across different orbital platforms and computing environments. They published their findings in a preprint article.
What Makes This Deployment Significant
Foundation models are trained on massive amounts of unlabeled data, allowing them to detect patterns humans might miss. Once trained, they can be adapted for specific tasks using smaller datasets.
Prithvi was trained on 13 years of satellite data from NASA's Landsat and the European Space Agency's Sentinel-2 satellites. The model can map flood plains, monitor disasters, and predict crop yields.
Dr. Andrew Du, the project's lead researcher at Adelaide University, said the open-source availability of Prithvi was crucial. "If Prithvi weren't open source, I would have to train my own foundation model," Du said. "Having that model openly available saved a lot of time and effort."
Why Orbit Matters for Data Processing
Earth-observing satellites collect enormous volumes of data. Processing and analyzing that data in orbit before transmission to Earth accelerates insights.
Active satellites face constraints: bandwidth limits prevent large software updates. Most carry lightweight, specialized models. A foundation model solves this problem by handling multiple tasks within one architecture.
If researchers need the model to perform a new task after launch, they upload only a small decoder package-far less data than replacing the entire model.
Future Applications
Kevin Murphy, chief science data officer at NASA Headquarters, said the open-source approach accelerates development. "By sharing these tools with anyone who wants to use them, we accelerate scientific and technological development into the future," Murphy said.
Beyond image analysis, foundation models could eventually enable natural language interaction with satellites. Operators could ask questions about onboard data or system status and receive conversational responses.
NASA's team is developing additional foundation models. A heliophysics model called Surya launched in 2025. The team plans models for planetary science, astrophysics, and biological and physical sciences.
Learn more about AI for Science & Research or explore AI Research Courses for your professional development.
Your membership also unlocks:
