CAIG: Deep Learning-based Stochastic Models for Large-Scale Atmospheric Variability and Extreme Events

About This Grant

Extreme weather events—such as heat waves, cold snaps, wildfires, and heavy rainfall—pose increasing risks to society. These events are often driven by shifts in powerful atmospheric jet streams, which typically flow from west to east across the midlatitudes but can sometimes meander dramatically north or south. While recent advances in AI have shown promise in improving weather forecasts, many AI models still struggle with long-term stability, limiting their effectiveness in predicting extreme events. This project seeks to address that challenge by combining physics-based atmospheric science with state-of-the-art machine learning techniques. The goal is to enhance our ability to forecast high-impact weather events, ultimately supporting more effective planning and response in sectors such as energy, transportation, water management, and public health. Additionally, the project will contribute to education and workforce development. Graduate students will receive advanced training in both physics-based and data-driven approaches to atmospheric modeling, while several undergraduate students will gain hand-on experience in applying statistical analysis and AI techniques to weather data. The models and tools developed through this research will be made publicly available, fostering collaboration across the broader scientific community. The proposed research will advance understanding of atmospheric circulation and extreme weather by integrating a hierarchy of stochastic models with large-scale atmospheric dynamics. This hierarchy will include both deep learning-based generative models and traditional linear and nonlinear inverse models. The latter will serve as rigorous benchmarks to evaluate the long-term stability and reliability of the AI-based approaches, particularly in capturing the full probability distribution of midlatitude weather systems. Using these stochastic models, the project will generate large ensembles of simulated weather events, with a focus on extreme events and their subseasonal precursors. The research will also investigate the influence of stratospheric and tropical processes on midlatitude circulation, as well as the dynamic coupling between atmospheric flow and water vapor, using both linear and nonlinear frameworks. By bridging machine learning and atmospheric dynamics, this work strives to improve the performance of AI-based forecasting systems and contribute to a more comprehensive understanding of large-scale atmospheric variability. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.