Abstract:

AI-driven weather forecast models are now more accurate and much faster than the best physics-based models. The open-source Ai2 Climate Emulator (ACE) uses similar technology to accurately emulate both daily weather variability (including extremes) and climate of historical reanalysis or of a reference global atmospheric model. ACE runs 100x faster than a physics-based model of similar grid resolution – 1600 years per day on a 100 km grid with a single GPU.  ACE can be forced by specified sea-surface temperatures (AMIP) or coupled to a slab ocean model (SOM).  When trained on SOM-coupled reference model simulations spanning multiple climates forced by changed CO2 concentrations, ACE can accurately emulate its climate change response.  We close with some key remaining challenges to general-purpose use of ACE for climate modeling applications.