loading page

Application of the AI2 Climate Emulator to E3SMv2's global atmosphere model, with a focus on precipitation fidelity
  • +9
  • James PC Duncan,
  • Elynn Wu,
  • Jean-Christophe Golaz,
  • Peter Martin Caldwell,
  • Oliver Watt-Meyer,
  • Spencer Koncius Clark,
  • Jeremy McGibbon,
  • Gideon Dresdner,
  • Karthik Kashinath,
  • Boris Bonev,
  • Michael S Pritchard,
  • Christopher S. Bretherton
James PC Duncan
University of California, Berkeley

Corresponding Author:[email protected]

Author Profile
Elynn Wu
Allen Institute for Artificial Intelligence
Author Profile
Jean-Christophe Golaz
Lawrence Livermore National Laboratory (DOE)
Author Profile
Peter Martin Caldwell
Lawrence Livermore National Laboratory (DOE)
Author Profile
Oliver Watt-Meyer
Allen Institute for Artificial Intelligence
Author Profile
Spencer Koncius Clark
Allen Institute for Artificial Intelligence / NOAA-GFDL
Author Profile
Jeremy McGibbon
Allen Institute for Artificial Intelligence
Author Profile
Gideon Dresdner
Allen Institute for Artificial Intelligence
Author Profile
Karthik Kashinath
NVIDIA
Author Profile
Boris Bonev
NVIDIA
Author Profile
Michael S Pritchard
University of California, Irvine and NVIDIA
Author Profile
Christopher S. Bretherton
Allen Institute for Artificial Intelligence
Author Profile

Abstract

Can the current successes of global machine learning-based weather simulators be generalized beyond two-week forecasts to stable and accurate multiyear runs? The recently developed AI2 Climate Emulator (ACE) suggests this is feasible, based upon 10-year simulations with a network trained on output from a physics-based global atmosphere model using a grid spacing of approximately 110 km and forced by a repeating annual cycle of sea-surface temperature. Here we show that ACE, without modification, can be trained to emulate another major atmospheric model, EAMv2, run at a comparable grid spacing for at least ten years with similarly small climate biases -- a necessary prerequisite to wider applicability. With a comprehensive analysis from multiple temporal, spatial, and frequency domain perspectives, we show that ACE faithfully represents the spatiotemporal structure of EAMv2 precipitation and related variables. Finally, we show that a pretrained ACE network is able to adapt to a new global climate model simulation dataset with 10x fewer training steps than when starting from random initialization, all while still maintaining low levels of climate bias. Further analysis of these fine-tuning experiments reveal ACE's attractive ability to interpolate between distinct global climate models.