The Prithvi geospatial AI foundation model, a collaborative effort between NASA and IBM, has been uploaded and demonstrated in orbit, becoming the first geospatial foundation model deployed on two in-orbit platforms—the South Australian Kanyini satellite and the Thales Alenia Space IMAGIN-e payload on the International Space Station.

NASA’s Prithvi AI model is now the first geospatial foundation model deployed and operated in orbit, enabling real-time environmental analysis from space.

Prithvi was chosen in part because it is open-source, allowing rapid experimentation and avoiding from-scratch training, as project leader Dr. Andrew Du noted.

Open-source foundation models like Prithvi accelerate scientific and technological development by enabling broad access and collaboration, according to NASA officials including Kevin Murphy.

In-orbit validation included flood detection around Lake Norman, North Carolina after Hurricane Helene, demonstrating the model’s versatility beyond fire detection.

Prithvi’s architecture supports task adaptation with minimal data transfer, requiring only small decoder packages for new applications, a crucial advantage given satellite bandwidth limits.

The deployment highlights potential tasks such as mapping flood plains, monitoring disasters, and predicting crop yields, with future prospects including natural-language–like interaction with satellites and onboard data interrogation.

A foundation model is trained on vast unlabeled data and can be fine-tuned with smaller labeled datasets; deploying such models in orbit enables flexible, onboard data analysis with minimal bandwidth for updates.

Prithvi was trained on 13 years of Earth observation data from Harmonized Landsat and Sentinel-2, enabling strong generalization across diverse Earth observation tasks, including flood and cloud detection across different orbits.

Prithvi complements NASA’s broader open-source AI initiative, with other projects like Surya and plans for models spanning planetary science, astrophysics, and biology/physical sciences.

Lead researcher Andrew Du emphasizes collaborative, accessible AI development and Prithvi’s strong generalization across Earth observation tasks.

The in-orbit deployment was led by NASA’s IMPACT team at the Marshall Space Flight Center, with contributions from IBM and Australia’s Adelaide University/SmartSat collaboration.