IBM and NASA Release 2.3B Open-Source AI Model Prithvi WxC for Weather and Climate; Predicts Hurricane Francine

Sep 23, 2024, 12:41 PM

IBM and NASA have released Prithvi WxC, an open-source foundation model for weather and climate prediction. This 2.3B encoder-decoder model, trained on 160 atmospheric variables from NASA's MERRA-2 dataset, aims to provide scientists and developers with new insights into short-term weather and long-term climate conditions. Prithvi WxC, available on Hugging Face, can be fine-tuned for custom objectives, such as gravity wave flux data. The model has demonstrated the ability to rival traditional numerical weather prediction models, offering a comparison of predictions for Hurricane Ida. Additionally, a new machine-learning weather model has outperformed traditional models in predicting Hurricane Francine. This development highlights the potential of AI emulators in enhancing weather and climate research.