New Delhi: A team of Australian researchers have successfully uploaded and demonstrated the open-source Prithvi Geospatial artificial intelligence foundation
model on board two in-orbit platforms, making it the first geospatial foundation model to be deployed in orbit. NASA and IBM trained the model on 13 years’ worth of data. A compressed version of the model was uploaded to the Kanyini Satellite of the South Australian government, as well as the Thales Alenia Space’s ISS Mounted Accessible Global Imaging Nod-e (IMAGIN-e) payload on board the International Space Station (ISS). The performance of the model on flood and cloud detection was evaluated across two different orbiting platforms and computing environments.
The researchers chose Prithvi because of its strong generalisation across a variety of Earth observation tasks, as well as the fact that it was open source. A foundation model is an AI model that is trained on a vast corpus of unlabeled data, allowing the model to detect patterns in the data that humans cannot notice on their own. The model can then be fine-tuned for specific applications using a much smaller amount of labeled data. By open sourcing the Prithvi family of geospatial AI models, NASA and IBM are accelerating scientific and technological developments.
The development of Prithvi
The Prithvi family of models was developed by data scientists from IBM and NASA’s IMPACT AI team. The geospatial model was trained on the Harmonised Landsat and Sentine-2 dataset, that compiles over a decade of global geospatial data from NASA’s Landsat and ESA’s Sentinel-2 satellites. Prithvi can be adapted for mapping flood plains, monitoring disasters and predicting crop yields. Satellites today first relay the data captured to the ground, where they are processed for actionable intelligence. Processing data in orbit can allow for rapid dissemination of actionable intelligence for time-critical tasks such as floods, landslides or forest fire alerts.
