Will Chapman (CU Boulder): AICE Speaker Series

Abstract: AI-driven weather models have advanced rapidly over the past five years and now rival or exceed physics-based forecasting systems across many prediction tasks. At the same time, AI climate emulators are beginning to reproduce both weather variability and long-term climate statistics at a fraction of the computational cost. Here, I will discuss the training, implementation, and capabilities of CAMulator, an emulator of the National Center for Atmospheric Research’s Community Atmosphere Model version 6 (CAM6), designed to simulate atmospheric variability from days to centuries. CAMulator evolves the full three-dimensional prognostic atmospheric state and can generate approximately 480 simulated years per day on a single GPU.

We will discuss the training and capabilities of CAMulator and then shift to the representation of atmospheric rivers across the North Pacific as a process-oriented test of emulator fidelity. CAMulator reproduces their large-scale climatology, moisture transport, precipitation patterns, and lifecycle characteristics, while also capturing many aspects of the associated synoptic-scale circulation. Examining these coherent weather systems across spatial scales provides a particularly sensitive way to identify where additional fidelity can be gained, including in the representation of precipitation extremes, frontal structure, and mesoscale variability. I will discuss how we are using these diagnostics not only to evaluate the emulator, but also to inform changes to model architecture and training, illustrating how physically grounded analysis can become part of the development cycle for the next generation of AI climate models.

Bio: Will Chapman is an Assistant Professor in the Department of Atmospheric and Oceanic Sciences at the University of Colorado Boulder. His research focuses on weather and climate predictability, with an emphasis on machine learning, climate model emulation, extreme weather, and hybrid approaches that combine data-driven and physics-based modeling. Prior to joining CU Boulder, Will was a Scientist and Advanced Study Program Postdoctoral Fellow at the National Center for Atmospheric Research (NCAR), where he worked on machine learning approaches for Earth system modeling and was involved in the Multiscale Machine Learning in Coupled Earth System Modeling (M²LInES) project. He received his Ph.D. from Scripps Institution of Oceanography, where his research focused on atmospheric predictability and machine learning for weather and climate prediction.