ClimForGe: A Diffusion-based Forcing–Response Climate Emulator on Daily Timescales (Papers Track)
Jack Kai Lim (UC San Diego); Salva Rühling Cachay (UC San Diego); Duncan Watson-Parris (UC San Diego)
Abstract
Climate models are indispensable tools for projecting and understanding climate change. Unfortunately, their computational demands severely limit the exploration of diverse climate scenarios and the characterization of extreme events, hindering informed policy decisions. While computationally efficient climate model emulators offer a potential solution, they typically only provide monthly or annual statistics. This paper introduces ClimForGe, a diffusion-based stochastic climate emulator trained on CESM2 capable of efficiently sampling global, daily-scale weather changes under realistic climate forcings. We demonstrate that our emulator accurately reproduces both daily snapshots and long-term statistical properties of temperature and precipitation, offering a powerful tool for rapid exploration and characterization of extreme events in a changing climate.