Machine Learning

Contents

Machine Learning#

CAMulator is a machine-learned emulator of the Community Atmosphere Model version 6 (CAM6) — the atmospheric component of CESM. Trained on CAM6 output, it autoregressively predicts the next atmospheric state given prescribed sea surface temperatures (SST) and incoming solar radiation, while explicitly conserving dry air mass, moisture, and total atmospheric energy. CAMulator accurately reproduces the CAM6 climatology and key modes of variability including ENSO, the North Atlantic Oscillation, and the Pacific-North American pattern, at 350x the speed of the full model, making it a powerful tool for large ensemble experimentation and rapid exploration of atmospheric responses to climate perturbations.

Learning Goals#

  • Understand what an atmosphere model emulator is and how it differs from a traditional physics-based model

  • Run a global atmospheric simulation using CAMulator with prescribed boundary conditions (SST, CO₂, sea ice, solar insolation)

  • Compare a control simulation against perturbed simulations (+2K SST warming and an idealized ENSO SST pattern) to isolate the atmospheric response

  • Interpret dynamical atmospheric responses to SST perturbations using model output diagnostics

  • Gain hands-on experience with the CREDIT package used to train and run CAMulator

Contents