The Climate Extremes Theory and Data (CeTD) Group
We study extreme weather, climate change, geophysical turbulence, AI weather and climate prediction, and AI+science through the lens of multi-scale nonlinear dynamics. We integrate tools and concepts from nonlinear and climate dynamics, applied and computational math, and AI to gain a deeper theoretical understanding of these phenomena and to develop novel interpretable and generalizable frameworks to predict them across time and spatial scales. We are also interested in interdisciplinary collaborations that enable direct translation of fundamental advances in AI for weather and climate modeling to address critical societal needs, particularly through our involvement with UChicago’s AI for Climate (AICE) Initiative and Human-centered Weather Forecasts (HCF) Initiative.
Media coverage of our work:
AI and the physics of weather/climate: Nature; Big Brains podcast; PNAS podcast; Fox Weather; SIAM News; CBS News; MaRS; Fox 32; Communications of the ACM; The Chicago Maroon
Human-centered weather forecasting: Nature Climate Change; The Economist; WSJ; Le Figaro; The Hindu; El Pais
news
| Aug 05, 2026 | Check out the paper co-led by ENS-Paris PhD student Amaury Lancelin and UChicago postdoc Dr. Alex Wikner titled “AI-boosted rare event sampling to characterize extreme weather” published in PRL. The rarest weather is the hardest to simulate and characterize, because it almost never happens: learning about a once-in-a-thousand-year heat wave would require simulating thousands of years of weather. The paper offers a hybrid AI+math+physics solution: a fast AI weather model guides an expensive physics-based climate model, through a mathematical algorithm, to more efficiently sample times where the chance of extreme events is higher. This reproduces the statistics of heat waves with 100x less computing time, making it practical to study the most extreme (and impactful) events that have not happened yet, the “gray swans”. The paper is highlighted by APS Editors and UChicago News. |
|---|---|
| Apr 15, 2026 | Our team is one of eight selected nationally for the inaugural Moonshots program of the Laude Institute, for a project titled “Actionable AI Weather Forecasts for Developing Economies” led by statistician and computer scientist Rebecca Willett with Nobel laureate economist Michael Kremer, computer scientist Ian Foster, and me. The people who most need a good weather forecast often have the least access to one. We will combine AI weather and climate models with data from developing countries, build software that lets any country plug in its own observations, and create benchmarks so that forecasts can be compared fairly across countries. The eight teams now have six months to turn their seed grants into full proposals for a $10M multi-year Moonshot Lab, to be decided in October. The award is covered by Forbes and UChicago News. |
| Feb 10, 2026 | Check out the paper led by PhD student Karan Jakhar titled “Analytical and AI-discovered stable, accurate, and generalizable subgrid-scale closure for geophysical turbulence” published in PRL. The paper shows that AI, specifically techniques that can discover equations from data, can identify new closure models for turbulence that are accurate and stable, and work for unseen flows, but only after physics constraints are included in the discovery. The paper also shows that guided by the AI discovery, this closure could be derived analytically. The results provide an example of AI accelerating scientific discovery. The paper is highlighted by APS Editors and by UChicago News. |
| May 01, 2025 | Check out the paper led by research scientist Dr. Qiang Sun titled “Can AI weather models predict out-of-distribution gray swan tropical cyclones” published in PNAS. The paper presents controlled experiments showing that an AI weather model cannot forecast “gray swan” tropical cyclones: those stronger than any the AI model had seen in the training set (i.e., AI model cannot “extrapolate”). However, the AI model shows the remarkable ability to learn from strong storms in one ocean basin and forecast them in another (i.e., it can “translocate”). The results have important implications for the current AI weather models and climate emulators. The paper is highlighted in PNAS podcast, Gizmodo, and UChicago News. |