Project Grant 2616257

The National Science Foundation Office of Integrative Activities awarded The Pennsylvania State University $300,000 on August 1, 2026, under the Geosciences program (CFDA 47.050) for collaborative research developing physics-informed artificial intelligence emulators to improve subseasonal-to-seasonal weather prediction through better representation of stratosphere-troposphere coupling.

The project addresses a recognized limitation in current AI weather and climate emulators: their poor representation of the stratosphere and its connection to surface weather, which constrains forecast skill for high-impact events such as severe winter cold snaps and heavy snowfall across North America and Eurasia occurring two weeks to two months ahead. The research will build physics-informed AI emulators incorporating fundamental dynamical constraints into machine learning loss functions to achieve more robust and consistent representation of middle and lower atmosphere interaction. The work builds on existing weather emulator platforms (Graphcast and ACE2) and will clarify when stratospheric information provides additional predictability, identifying forecasts of opportunity for extreme surface weather events.

The award supports research and educational activities at Penn State's University Park campus in State College, Pennsylvania, through July 31, 2029. Deliverables include development of new AI modeling tools, cross-training of students at the intersection of geoscience and computer science, curriculum integration, and release of open-source software pipelines intended to accelerate improvements in weather forecasting and climate modeling across the scientific community.