IBM has announced that it is open sourcing its watsonx.ai geospatial base model on Hugging Face -- built on NASA satellite data. This will be the largest geospatial base model on Hugging Face and the first open source AI base model built in partnership with NASA.
The announcement noted that IBM began building AI-based models for geospatial data earlier this year as part of its Space Act Agreement with NASA. Providing geospatial foundational models now through Hugging Face can advance the democratization of AI access and applications, resulting in new innovations in climate and earth sciences.
Sriram Raghavan, vice president of AI at IBM Research, said, "It has never been clearer that open source technologies will play a critical role in accelerating key discoveries such as climate change...We can harness the power of collaboration to implement faster, more impactful solutions." solutions to improve our planet.”
IBM and NASA jointly trained the model on a geospatial dataset called Harmonized Landsat Sentinel-2 satellite data (HLS). The dataset includes images of the Earth's surface taken by NASA's Landsat-8 satellite and also includes measurements from Sentinel-2, a constellation of satellites operated by the European Space Agency.
IBM says the new model is designed to help researchers identify areas of the continental U.S. that may be at risk of flooding and wildfires; to date, the model has analyzed geospatial data 15 percent faster than state-of-the-art neural networks. "With further fine-tuning, the underlying model could be redeployed for tasks such as tracking deforestation, predicting crop yields, or detecting and monitoring greenhouse gases."
The researchers are also collaborating with Clark University and plan to apply the model to areas such as time series segmentation and similarity studies. These are two popular data analysis methods that can be used not only for geospatial research but for a range of other tasks. For example, time series segmentation can be used to study the causes of stock price fluctuations.
More details can be found in the official announcement .