GALCIT Colloquium

Machine learning methods are proving enabling across the climate simulation research pipeline. In the first part of this talk, I will discuss experience from academia using ML in a hybrid multi-scale simulation context to parameterize sub-grid processes of clouds, radiation and turbulence. In the second part, I will provide some emerging perspectives from my time spent at NVIDIA, on increasingly ambitious uses of ML to replace entire components of Earth System Models via autoregressive global emulation, and for generative super-resolution to replace traditional dynamical downscaling.