Search

Machine learning methods (support vector regression and random forest regression) were used to map gridded estimates of ΔpCO2 in the Southern Ocean from SOCAT v3 data. A low (1°  × monthly) and high (0.25° × 16-day) resolution implementation of each of these methods as well as the SOM-FFN method of Landschützer et al. (2014) were added to a five member ensemble. The ensemble mean ΔpCO2 was used to calculate FCO2 (air-sea CO2 flux). Data was separated into nine domains defined by basin (Indian, Pacific and Atlantic) and biomes defined by Fay and McKinley (2014). The regional approach showed large zonal asymmetry in ΔpCO2 and FCO2 estimates. Importantly, there was a seasonal decoupling of the modes summer and winter interannual variability. Winter trends had a larger 10 year mode of variability compared to summer trends, which had a shorter 4-6 year mode. To understand this variability of FCO2, we separately assessed changes in summer and winter ΔpCO2 and the drivers thereof. The dominant winter changes were driven by wind stress variability. Summer variability correlated well with chlorophyll-a variability where the latter had high concentrations. In regions of low chlorophyll-a concentrations, wind stress and sea surface temperature were lower order drivers of ΔpCO2. [ABSTRACT FROM AUTHOR]