The model's architecture allows for versatile applications beyond the capabilities of traditional AI weather models, according to a recent paper published on arXiv titled "Prithvi WxC: Foundation Model for Weather and Climate." Some uses include improving localized forecasts, predicting severe weather, enhancing global climate simulations, and refining numerical models of physical processes. A key experiment in the paper demonstrated the model's ability to reconstruct global surface temperatures from just five percent of the original data.

This foundation model was trained using NASA's MERRA-2 dataset, which comprises 40 years of Earth observation data. Its scalable design allows for fine-tuning to meet various global, regional, and local requirements, making it a versatile tool for weather studies.

Available for download on Hugging Face, the model includes two fine-tuned versions designed for specific climate and weather applications:

1. Climate and Weather Data Downscaling: This fine-tuned model enables the creation of high-resolution forecasts from low-resolution data, such as temperature, precipitation, and surface winds, with up to 12x resolution improvements.

2. Gravity Wave Parameterization: Gravity waves, which significantly affect climate and weather processes, have been poorly represented in traditional models. This model improves the accuracy of simulations involving these waves, reducing uncertainty in forecasting climate events.

Karen St. Germain, director of NASA's Earth Science Division, emphasized the importance of tools like this model in providing actionable science, noting, "The rapid changes we're witnessing on our home planet demand this strategy to meet the urgency of the moment."

Juan Bernabe-Moreno, Director of IBM Research Europe, highlighted the flexibility of the model, stating, "We have designed our weather and climate foundation model to go beyond the limitations of single-use AI models."

Oak Ridge National Laboratory's Arjun Shankar underscored the significance of this collaboration for advancing research, explaining, "Our collaboration with IBM and NASA to support the creation of the Prithvi weather and climate foundation model was a key part of our goal to bring advanced computing and data to problems of national importance."

IBM has already begun working with Environment and Climate Change Canada (ECCC) to test the model's flexibility, focusing on short-term precipitation forecasting using real-time radar data. The team is also testing downscaling from global to regional forecasts.

This AI foundation model is part of a broader partnership between IBM Research and NASA aimed at leveraging AI to explore Earth. It is the latest addition to the Prithvi family of models, joining last year's geospatial AI foundation model, which has been widely used for analyzing disaster patterns, biodiversity, and other environmental changes.

The weather and climate foundation model and the gravity wave parameterization model can be accessed through NASA-IBM's Hugging Face page at https://huggingface.co/Prithvi-WxC, while the downscaling model is available via IBM Granite at https://huggingface.co/ibm-granite.

Related Links
NASA-IBM Hugging Face
All about the robots on Earth and beyond!