Collaborative Research: CAIG: Unraveling the Long-Range Predictability and Environmental Dependency of High-impact Cyclones with Artificial Intelligence Tools

University Of Washington, Seattle WA

Investigators

Abstract

Cyclones occur in both the tropics (as hurricanes) and middle latitudes (as, e.g., Nor'easters), posing significant threats to lives and economies. While weather forecasts have improved overall, the evolution and hazards of some of these high-impact storms remain poorly forecast and understood, partly because different types of storms are often studied in isolation. However, real-world storms can transform and interact with other weather systems, as seen with Hurricanes Sandy (2012) and Helene (2024). This project addresses this challenge by using advanced Artificial Intelligence (AI) to study the full life-cycle of cyclones, viewing them not as separate categories but as an interconnected spectrum of weather systems. The ultimate goal is to uncover the early warning signs of dangerous storms, leading to more accurate and trustworthy long-range predictions. This will give communities more time to prepare, potentially saving lives and reducing economic damage. The project will also train a new generation of scientists at the cutting edge of atmospheric science and AI, and by making its tools and findings openly available, it will help accelerate improvements in forecasting for all. This project aims to advance fundamental understanding of the long-range predictability and environmental dependency of cyclones using new observational datasets and AI tools. The research leverages an interdisciplinary approach to: 1) Delineate the predictability limits of high-impact cyclones using newly developed AI weather forecasting models and identify the initial conditions that control their development; 2) Adapt AI algorithms based on robust representation learning to automatically flag atmospheric precursors that influence long-range forecasts; 3) Conduct and co-develop physical and physics-informed AI simulations to test hypotheses about the factors controlling cyclone evolution; and 4) Use AI and statistical models to probe the contributions of ocean, sea ice, and land properties to the predictability of cyclone activity. This work will generate new insights into cyclone dynamics and provide a robust framework for improving weather and climate prediction systems. 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.