Special Topics in Natural, Biological and Medical Sciences

Spring 2027 Subject: AI in Weather & Climate Simulation: From Days-to-Decades Deep-learning models can now produce global weather forecasts in seconds that rival those from traditional numerical models that have taken decades to develop. And they're getting better fast. At the same time, researchers are pushing these models beyond weather to simulate the earth system on timescales of seasons, years, and decades. You'll work hands-on with state-of-the-art AI weather and climate models, learning how they are designed and trained, as well as how to run them, evaluate them, and diagnose where they succeed and fail. Topics include working with large spatio-temporal datasets, fundamentals of data-driven time series simulation, probabilistic prediction, forecast verification and evaluation metrics, choices in deep-learning architecture and training (generative approaches, autoregressive rollout, embedding physical constraints), and current open challenges: blurring, checkpoint selection, chaos, and the problem of climate change itself pushing models out-of-distribution.