Freshwater Sources and their Variability through Salinity-δ18O Relationships: A Machine Learning Solution to a Water Mass Problem

While the Southern Ocean is freshening, the sources of this freshening and their variability remain uncertain. Freshwater enters the ocean as Meteoric Water (MW; precipitation, river runoff, glacial discharge) and Sea Ice Meltwater (SIM). These inputs can be quantified using seawater salinity and stable oxygen isotopes in seawater, δ¹⁸O; however it involves the challenging task of determining the isotopic signature of MW (δ¹⁸O_MW). Here, we apply Self-Organising Map (SOM), a machine learning technique, to water mass properties to estimate the global distribution of the isotopic signature of MW (δ¹⁸O_MW) by characterizing distinct salinity-δ¹⁸O relationships from two comprehensive datasets. The inferred δ¹⁸O_MW is then used in a 3-endmember mixing model to estimate MW and SIM contributions to global ocean freshwater content. Our results show the large scale distribution of MW and SIM, as well as giving insights into their role in mass transformation and interannual variability. We highlight the MW content in Ice Shelf Water and Antarctic Bottom Water linked to glacial melt, which is concurrent with brine content derived from sea ice formation. Our results also show that AABW has freshened since the 1990’s due to a reduction of sea ice formation (less brine production) rather than an increase in glacial melt, and suggest the emergence of anthropogenic forced signals in seawater δ¹⁸O.