New Platform CREDIT Simplifies AI Integration in Weather Research, Empowering Scientists and Students Alike

Artificial intelligence (AI) is revolutionizing weather forecasting by offering faster and more efficient models compared to traditional methods. However, despite these advantages, AI models face significant limitations and accessibility issues for the broader research community. To address this, the U.S. National Science Foundation National Center for Atmospheric Research (NSF NCAR) has introduced the Community Research Earth Digital Intelligence Twin (CREDIT), a platform designed to simplify the use and training of AI weather models. Challenges of Traditional Weather Models Traditional weather models rely on complex equations representing atmospheric physics, from small-scale storm dynamics to large-scale air mass movements. While highly accurate, they are computationally intensive, making high-resolution studies costly and time-consuming. Doubling the model’s resolution requires a 16-fold increase in computing power. Additionally, traditional models struggle with phenomena whose underlying physics are poorly understood, such as the rapid intensification of hurricanes or the formation of hail. Introduction of CREDIT CREDIT aims to democratize the use of AI in weather research by providing: A Library of AI Models: Researchers can choose from a variety of pre-built AI models, reducing the need for extensive coding and equation-solving knowledge. Pre-Prepared Datasets: High-quality datasets are available for training the models, streamlining the data preparation process. Access to High-Performance Computing: The platform leverages NSF NCAR’s computing resources, enabling researchers to run models even if they lack powerful hardware. David John Gagne, leading NSF NCAR's machine learning efforts, emphasizes that CREDIT is designed to lower the barriers to entry, allowing a broader range of users, from experienced researchers to novices, to harness the potential of AI. WXFormer: A New AI Weather Model As part of the CREDIT development, the NSF NCAR team created WXFormer, an AI weather model tailored for atmospheric research. WXFormer addresses common issues in AI weather modeling, such as error growth and the provision of hourly forecasts rather than the typical six-hour intervals. The model's performance was evaluated against the High Resolution Integrated Forecast System (HRES-IFS), a leading traditional weather model developed by the European Center for Medium-Range Weather Forecasts. Test Results During tests, WXFormer and another AI model, FuXi, were tasked with predicting the track and intensity of Hurricane Laura, a Category 4 storm that struck western Louisiana in 2020. WXFormer excelled in predicting the storm’s intensity at a five-day lead time but had track errors. An hourly forecasting version of WXFormer provided a more accurate track but underestimated the storm’s strength. FuXi similarly predicted a weak storm and struggled with track accuracy. In contrast, HRES-IFS also underestimated the storm’s intensity and had track errors in the opposite direction. Overall, both WXFormer and FuXi performed competitively with HRES-IFS, highlighting the potential of AI models in weather forecasting. Ongoing Developments NSF NCAR is actively enhancing CREDIT to improve its user accessibility and scalability. A newer release of the software includes updates that make it more user-friendly. The team is also developing a "CAMulator" AI model to emulate the Community Atmosphere Model (CAM), a key component of the Community Earth System Model. This will facilitate the integration of AI into global Earth system simulations. In tandem with CREDIT, NSF NCAR is building an integrated data commons to enhance the availability and quality of datasets for AI model training. High-quality data is crucial for AI, and this initiative will provide a solid foundation for future developments in the field. Industry Insights Industry experts commend NSF NCAR for its pioneering efforts in making AI more accessible to the atmospheric research community. They see CREDIT as a significant step toward leveraging AI's potential to solve complex weather-related problems, ultimately improving forecast accuracy and reducing computational costs. The collaborative nature of the platform is expected to foster innovation and collective improvement in AI weather modeling. NSF NCAR, known for its contributions to open-source weather models and comprehensive user support, is poised to continue leading advancements in Earth system science research. The center’s commitment to reducing technical barriers and promoting community involvement ensures that CREDIT will evolve and become a valuable resource for researchers worldwide. In summary, CREDIT represents a promising convergence of AI and traditional meteorological techniques, potentially transforming how weather research is conducted and enhancing our ability to predict critical weather events.