Note: Single-source report; awaiting corroboration.
Researchers from Adelaide University and the SmartSat Cooperative Research Center in South Australia have uploaded and tested NASA and IBM’s open-source Prithvi Geospatial AI foundation model on two orbiting platforms. This marks the first in-orbit deployment of a geospatial foundation model, enabling Earth observation tasks such as flood and cloud detection.
Prithvi was installed on South Australia’s Kanyini satellite and the Thales Alenia Space IMAGIN-e payload on the International Space Station. Tests evaluated its flood and cloud detection performance across diverse satellite computing environments.
The model was trained on a combined dataset of over 13 years of global geospatial data from NASA’s Landsat and the European Space Agency’s Sentinel-2 satellites. According to project lead Dr. Andrew Du, Prithvi’s open-source approach reduced research time and effort.
Foundation models like Prithvi use large amounts of unlabeled data to detect patterns, supporting applications such as flood, disaster, and crop monitoring. Kevin Murphy, NASA’s Chief Science Data Officer, said that open-sourcing accelerates scientific and technological progress.
Prithvi was developed collaboratively by data scientists from IBM and NASA’s IMPACT team at the Marshall Space Flight Center. The team published their findings in a preprint article, providing further details on the model’s performance in orbit.