Item

Abstract

The ocean’s Meridional Overturning Circulation (MOC) plays a key role in the climate system, and thus monitoring its evolution is a scientific priority. However, monitoring arrays are limited to just a few latitudes in the Atlantic Ocean. Here we explore the possibility of inferring the MOC from globally-available satellite measurements via machine learning (ML) techniques, using the ECCOV4 state estimate as a test bed. The methodological advantages of the present approach include the use purely of available satellite data, its applicability to multiple basins within a single ML framework, and the ML model’s simplicity. The ML model exhibits high skill in reconstructing the overturning cells in the Southern Ocean, Indo-Pacific Ocean, and the Atlantic Ocean. In particular, the approach achieves a higher skill in predicting the Southern Ocean abyssal MOC and the AMOC at 26.5N than has previously been achieved via dynamically-based approaches. We quantify the skill of our ML-based MOC reconstructions as a function of latitude in each ocean basin, and as a function of the time scale of MOC variability. We further test which combinations of satellite-observable variables are optimal, and explore how spatial coarsening of the input variables influences the ML model skill. For example, we find via ML interpretability techniques that high reconstruction skill in the Southern Ocean is mainly due to bottom pressure variability at a few prominent bathymetric ridges. Finally, we discuss the potential for reconstructing MOC strength from real satellite measurements.