In a joint effort, NASA and IBM have unveiled Prithvi WxC, an open-source AI model designed to improve weather forecast accuracy while cutting down on the computational power needed. This development is additionally backed by the Oak Ridge National Laboratory under the US Department of Energy.
Model Mechanics and Training Process
The Prithvi WxC model comprises 2.3 billion parameters and is trained on NASA's MERRA-2 dataset, spanning 40 years of Earth observations. Despite its modest size, it predicts global surface temperatures using just a fraction of the data, effectively capturing complex weather patterns like hurricanes and atmospheric rivers.
Built as a foundational model, Prithvi WxC can be customized for numerous purposes, from short-term weather forecasting to long-range climatic projections. IBM Research Europe points out in a blog post that the model overcomes the limitations typically faced by large AI models confined to specific datasets.
AI Weather Forecasting Approaches Compared
Tech giants like Google and Nvidia are also exploring AI in meteorology. However, IBM and NASA's model is set apart by its efficiency and adaptability. Google's efforts involve merging machine learning methods with physical models for greater forecasting precision, whereas Nvidia focuses on refining its Earth-2 models. In contrast, Prithvi WxC stands out due to its distinctive approach.
Available on Hugging Face, Prithvi WxC includes two specialized iterations. One improves downscaling of climate data for high-resolution forecasting, and another optimizes gravity wave parameterization, which enhances numerical models' accuracy in representing these phenomena.
With its open-source release, Prithvi WxC is ready for global adaptation, offering researchers the flexibility to modify it for specific needs, including severe weather alert refinement and climate simulation improvement. According to Karen St Germain from NASA's Earth Science Division, this tool promises to support more informed decision-making concerning weather and climate trends.
Leveraging AI in the Climate Prediction Industry
Weather forecasting is an area where Microsoft is becoming a leader, thanks to its AI models. Microsoft Start weather development team has been recognized as the “World's most accurate global forecast provider” by ForecastWatch in March 2023.
The team's groundbreaking research, detailed in a paper on the arXiv site at Cornell University, showcases a medium-range forecast model that significantly improves upon the current system utilized by the European Centre for Medium-range Weather Forecasts (ECMWF). In June 2024, Microsoft announced further enhancements for its forecasting.
Also that same month, Microsoft introduced Aurora, an AI model that focuses on atmospheric predictions. Aurora is a 3D foundation model capable of predicting various weather conditions with high accuracy and speed. Trained on vast datasets, it outperforms traditional models in many areas, including air pollution and weather forecasting.
Google too is making strides in this area. In July, the company introduced NeuralGCM, which blends AI and physics to improve weather forecasting. In collaboration with the European Centre for Medium-Range Weather Forecasts (ECMWF), NeuralGCM is a hybrid model that leverages artificial intelligence and traditional forecasting methodologies. The approach is intended to provide more precise long-term climate monitoring and prediction of events such as cyclones compared to conventional systems.
