Recent research, published in Science Advances (Li et al., 2025; DOI:10.1126/sciadv.adt4195), leverages machine learning (ML) to quantify how oceanic mesoscale eddies significantly enhance the ocean's capacity to absorb atmospheric carbon dioxide (CO2).
Oceanic mesoscale eddies play a crucial but underexplored role in regulating carbon fluxes and climate change. While they redistribute heat, salt, nutrients, and other tracers, their effects on CO2 uptake remain uncertain. The challenge in this field has been the sparse observational data for sea surface partial pressure of CO2 (pCO2), which makes it difficult to track the effects of eddies. To overcome this, an observation-based Feed-Forward Neural Network (FNN) was employed. This ML model was trained to accurately reconstruct sea surface pCO2 levels using a suite of available observational predictors. By generating an ensemble of 10 FNN models, a robust and continuous data set was created that allowed to analyze the CO2 flux throughout the lifetimes of thousands of individual eddies. The key finding, powered by this ML approach, is a striking asymmetry. Anticyclonic eddies substantially enhance CO2 uptake on average, while cyclonic eddies marginally diminish it. This asymmetry yields an overall net increase in CO2 absorption by 9.98 ± 2.28 and 13.82 ± 9.94% in the Kuroshio Extension and Gulf Stream, respectively, major carbon sequestration regions. The finding suggests a potential underestimation of the ocean’s capacity for carbon sequestration because of insufficient incorporation of eddies in current observations, emphasizing the need for expanded monitoring in eddy-rich, undersampled regions. Read the full article here.