Global Climatology of Submesoscale Activity Using Machine Learning

Submesoscale eddies are oceanic structures that occur on horizontal scales from 0.1-10 km, vertical scales from 0.01-1 km, and last from hours to several days. They are characterised by a Rossby number of Ro = ζ/f ~ O(1), where surface vertical vorticity ζ is similar to Coriolis frequency f. Submesoscale eddies are important in setting the stratification in the ocean surface mixed layer, mediating air-sea exchanges, and transporting energy between large and small scale motions. However, the study of submesoscale eddies on a global scale has been hindered by a shortage of global, long-term datasets. To fill this gap, we train and apply an unsupervised machine learning method adapted from the Profile Classification Model (PCM) to density profiles collected by Argo floats over global ocean from 2000-2021, producing the first global observational climatology of submesoscale activity. The adapted PCM identifies regions with high submesoscale activity using solely the density profiles and without any additional information on the velocity, location, or horizontal density gradients. The climatology shows that submesoscale activity peaks in spring in both hemispheres and lags behind the maxima of mixed layer depth by one month, suggesting that submesoscale eddies play important role in re-stratifying the mixed layer. Hotspots of submesoscale activity can be found in the Norwegian Sea and the Drake Passage in spring. This observational reconstruction of submesoscale activity enables the study of submesoscale distribution, seasonality, and inter-annual variation on a global scale.