NASA and IBM have joined forces to develop an AI foundation model for applications in weather and climate. The collaboration aims to leverage their expertise in Earth science and AI to create a model that offers significant advantages over existing technology.
While current AI models like GraphCast and Fourcastnet are available, IBM notes that they are merely emulators and not foundation models. Foundation models are the base technologies that power generative AI applications, unlike emulators which are limited to making weather predictions based on training data and do not encode the physics at the core of weather forecasting.
The goals for the new foundational model include expanded accessibility, faster inference times, greater diversity of data, and improved forecasting accuracy for climate applications. The model is expected to be able to predict meteorological phenomena, infer high-res information from low-res data, and identify conditions conducive to everything from airplane turbulence to wildfires.
This collaboration follows on the heels of another foundational model developed by NASA and IBM, which harnesses data from NASA satellites for geospatial intelligence. This model, the largest geospatial model on the open-source AI platform Hugging Face, has been used to track and visualize tree planting and growing activities in water tower areas in Kenya, as well as to analyze urban heat islands in the United Arab Emirates.
