ACE2-NEMO: Coupling an ML atmospheric emulator to a full-depth dynamical ocean model

Abstract

Understanding how fast atmospheric variability shapes slow climate variability and sensitivity remains a central challenge in Earth-system science. Recent advances in machine-learned (ML) atmospheric models have demonstrated remarkable skill on weather timescales, but their emergent behaviour in a fully coupled climate system remains largely unexplored. We present early results from a new hybrid modelling framework, in which the ACE2 ML atmospheric emulator is interactively coupled to the NEMO ocean model. We report on a set of 70-year coupled simulations (1950-2020 historical forcing and fixed-1950s control). These experiments represent, to our knowledge, the first multi-decadal integrations of a machine-learned atmosphere interacting with a full-depth dynamical ocean. Several historical and fixed-1950s control simulations from the fully dynamic global coupled climate model EC-Earth, which has the same ocean component used in ACE2-NEMO, are also considered for comparison. We assess the behaviour of the coupled system, with particular focus on low-frequency tropical variability and the climate response to greenhouse-gas forcing. Analysis of potentially emergent El Niño-like variability reveals realistic fast timescale air-sea coupling in the tropical Pacific, but the temporal variability is unrealistic, with very low amplitude oscillations; this appears to be due to weak atmospheric feedback in the tropical Pacific. The response to CO2 forcing shows initial agreement with EC-Earth3P, but deviates due to reduced downward short-wave radiation in ACE2. These results provide a unique test of physical realism for atmospheric emulators, and evaluate the possible role of entirely machine-learned components in next-generation Earth system models.

Figures (20)

• Figure 1: Time-averaged variables for the first ensemble member of the 70-year ACE2-NEMO-control and ECE3P-control experiments (a and b) daily precipitation (c and d) 2-metre temperature (e and f) sea surface temperature (g and h) and sea surface height.
• Figure 2: Globally averaged time series for the ACE2-NEMO-control 3-member ensemble and EC-Earth3P-control. Cosine-latitude weighting is used to account for different grid cell size (a) sea surface temperature, (b) sea ice volume, (c) sea surface height and (d) total heat flux (non-solar plus solar heat fluxes).
• Figure 3: Analysis of ENSO properties of the model simulations (a) Power spectra of the Niño 3.4 index for the different models and ERA5 reanalysis. The grey dashed line indicates an AR(1) process fitted to the ACE2-NEMO-control spectra, and the grey shaded area indicates the 2.5th-97.5th percentile range estimated by sampling 1000 AR(1) processes with the same parameters. (b) and (c) Pointwise regression of daily precipitation onto the Niño 3.4 index for the first ensemble member.
• Figure 4: Regression slopes between variables related to ENSO positive and negative feedbacks, organised by column (left column) 10-metre eastward wind regressed onto domain SST gradient (middle column) Gridpoint-wise regression of downward short-wave radiation flux on SST and (g-i) gridpoint-wise regression of latent plus sensible heat flux on SST. ACE2-NEMO plots use the first ensemble member only.
• Figure 5: Globally averaged temperatures for the ACE2-NEMO-hist, ECE3P-hist and ECE3-hist experiments, together with ERA5 reanalysis. Temperatures are expressed as changes relative to 1951. Cosine-latitude weighting is used to account for different grid cell size. (a) 2-metre temperature (b) sea surface temperature.