The unveiling of the watsonx.ai geospatial foundation model is a ground-breaking development in AI technology for climate science from IBM and Hugging Face. This geospatial model — which is hosted on Hugging Face and uses NASA satellite data — is the largest of its kind and represents an innovative step towards open-source AI models created in collaboration with NASA.
Jeff Boudier, the head of product and growth at Hugging Face, emphasized the importance of collaboration and knowledge sharing in growing AI. Open-source AI, which works in concert with model and dataset releases, is a crucial element in ensuring widespread access to the benefits of AI.
Due to the constantly changing environmental dynamics, climate science is faced with continual difficulties that need for access to real-time data. Scientists find it challenging to examine large datasets properly, despite the quantity of data. According to NASA, new missions will produce 250 000 terabytes of data by 2024.
In order to address this issue, IBM and NASA signed a Space Act Agreement earlier this year with the goal of developing an AI base model specifically for geospatial data.
By making the geospatial foundation model openly accessible on the Hugging Face platform, both entities aim to stimulate collaboration and lead advancements in climate and earth science.
IBM Research AI’s Vice President Sriram Raghavan said, “The importance of open-source technologies in accelerating critical discoveries, such as those related to climate change, has never been clearer. By merging IBM’s efforts to create versatile, reusable AI systems with NASA’s repository of Earth-satellite data and making it available on the premier open-source AI platform, Hugging Face, we can harness collaborative power to implement swifter and more impactful solutions that benefit our planet.”
Results from the geospatial model, which IBM and NASA jointly built using Harmonized Landsat Sentinel-2 satellite data (HLS) covering a year throughout the continental United States, have been encouraging. While using only half of the labeled data, it showed a 15% improvement over existing approaches.
The programme can be modified for a variety of tasks, including tracking deforestation, forecasting crop yields, and spotting greenhouse emissions.
The IBM and NASA partnership in developing this AI model is in line with NASA’s ten-year Open-Source Science Initiative, which supports a more open and diverse scientific community. 2023 has been declared the Year of Open Science by NASA and other federal organisations as a way to recognise the benefits of freely sharing data, knowledge, and ideas.