Five More Years: NSF Renews Funding for AI-Powered Earth System Modeling Center

The National Science Foundation renewed its funding for the AI-powered Earth system modeling center LEAP, aimed at improving Earth system projections through the integration of Earth system modeling and data science.

Efforts by policymakers, communities, and institutions worldwide to adapt to changing extreme events largely face a thorny problem: the projections of current models are simply too imprecise and computationally expensive to run. For example, estimates for events such as the frequency of droughts or an increase in extreme rainfall vary widely across models.

In other words, most stakeholders will find the picture fuzzy as they ponder where to adjust to more prolonged droughts or more intense storms. These investment decisions, which are estimated to cost hundreds of billions of dollars a year worldwide, currently lack robust and reliable models that can accurately predict extreme weather events. While Earth system data is abundant, only a small fraction of it is usable in current models, because of their limited capacity to ingest large datasets.

Learning the Earth with Artificial Intelligence and Physics, LEAP, was launched in 2021 with a $25 million grant from the National Science Foundation to fill this gap.

"With the increased intensity and duration of extreme events, LEAP's work is more critical than ever,” said Pierre Gentine, LEAP’s director and Maurice Ewing and J. Lamar Worzel Professor of Geophysics at Columbia Engineering, professor of earth and environmental sciences at Columbia, and a professor at Columbia Climate School. 

“Our team is using even more data with AI to improve Earth system models to improve our capacity to predict extreme events, such as droughts, heat waves, storms, and hopefully in the future other events such as wildfires and the associated smoke that we have been witnessing more and more over the last summers.”

Over the past five years, the center funded 26 research projects covering all Earth system components: the ocean, land, atmosphere, and cryosphere. The researchers developed and deployed LEAP Pangeo, an open, cloud-based data and compute platform that can be used globally for Earth system science research. The team also developed ClimSim, the largest dataset designed for hybrid machine learning and physics research, further proving LEAP’s core proposition that AI and physics-based Earth modeling can be fused to glean previously inaccessible insights about the Earth’s atmosphere.

Examples of additional impactful team projects include using the latest techniques from Generative AI, used for image and video generation, to develop novel ways to reconstruct the fine-scale structure of tropical cyclones or to better simulate very rare extreme events. The center has also developed an AI-agent pipeline to faithfully translate parts of an Earth system model written in an outdated code (in Fortran) to a modern AI-ready format (JAX), significantly improving the efficiency of the model calibration processes that used to rely on computationally expensive trial-and-error.

The center’s education and training efforts spanned workforce training, bootcamps, and an education initiative that reached more than 70 New York City public school teachers. LEAP’s public engagement work included co-hosting a hackathon at the American Museum of Natural History aimed at bridging the gap between available climate data and neighborhood-based resilience planning, drawing more than 75 hackers, 15 community group members, and 14 mentors. The center has also engaged hundreds of participants in its weekly Climate Data Science lectures.

Looking ahead, the LEAP team will build on the foundations of the first five years, lean deeper into the capabilities of generative AI to improve model reliability and usefulness, as well as agentic AI for scientific discovery, and further expand public and stakeholder engagement to make this research even more relevant to communities, researchers, officials, and policymakers.

Subscribe to LEAP’s newsletter to learn more about future events.