NASA estimates that its Earth science missions will generate approximately 250,000 terabytes of data in 2024. To help climate scientists and researchers efficiently analyze these vast amounts of satellite data, IBM, Hugging Face, and NASA have joined forces to create an open-source geospatial foundation model. This model will pave the way for a new generation of climate and Earth science AIs capable of tracking deforestation, predicting crop yields, and monitoring greenhouse gas emissions.
IBM utilized its Watsonx.ai as the foundational model, training it on a year’s worth of NASA’s Harmonized Landsat Sentinel-2 satellite data (HLS). This data, collected by the European Space Agency’s Sentinel-2 satellites, provides high-resolution optical imagery across 13 spectral bands over land and coastal areas.
Hugging Face is hosting the model on its open-source AI platform. IBM's team fine-tuned the model using labeled data for flood and burn scar mapping, resulting in a 15% performance improvement over existing solutions while using only half as much data.
Sriram Raghavan, VP of IBM Research AI, emphasized the importance of open-source technologies in addressing pressing challenges like climate change. "By merging IBM’s flexible, reusable AI systems with NASA’s extensive Earth-satellite data and offering it on the Hugging Face platform, we can harness collaboration to develop faster, more effective solutions that benefit our planet."