Meaningfulness of permanent El Niño attractor under SST-only coupling with a 6-hour atmospheric timestep

Determine whether the convergence to a permanent El Niño state observed when coupling a sea-surface-temperature-only ocean component to a default Spherical Fourier Neural Operator atmosphere with a 6-hour atmospheric timestep represents a physically meaningful signal of the learned coupled emulator or is instead an artifact of empirical machine learning noise.

Background

Early coupling experiments described by the authors produced a model that converged to a permanent El Niño state within six months when using a 6-hour atmospheric timestep with an SST-only ocean. They note that it is unknown whether this behavior reflects a genuine dynamical signal or is merely due to empirical ML noise, highlighting the need for systematic intercomparisons and tests.