ECMWF prototypes AIFS Marine, one model for atmosphere, waves, sea ice and surface ocean

ECMWF/AIFS

ECMWF has published a prototype, AIFS Marine, that extends its AI forecasting model beyond the atmosphere to ocean waves, sea ice and surface-ocean fields in a single network. Its reported gain is about one extra day of medium-range skill for nearly all tested marine variables. The surface ocean and sea ice are not yet operational, and the post gives no date for them.

• Surface-ocean and sea-ice variables added: sea-surface temperature, salinity and height, surface currents, sea-ice concentration, volume and velocity.
• Wave variables: significant wave height, mean period and direction.
• Reported results: about one extra day of medium-range skill for nearly all marine variables; significant wave-height errors about 10% lower against satellite observations; large-scale atmospheric skill broadly unchanged.
• Known issue: sea-surface height develops a negative bias as the forecast lead time grows.
• Trained on ERA5 atmospheric data, a dedicated wave hindcast and the ORAS6 ocean reanalysis, using the open-source Anemoi framework.