New Estimates of Global and Hemispheric Ocean-Atmosphere CO2 flux 1992-2018. and Their Uncertainty

We use the SOCAT surface ocean fCO₂ data base, with corrections for near-surface temperature gradients and the skin effects on temperature and salinity, to make new estimates of ocean-atmosphere CO₂ fluxes. The corrections increase uptake fluxes by about 0.4 PgC yr^-1 compared to the majority of recent estimates. To quantify uncertainties due to the uneven data coverage, we treat the data by three different methods, each applied to three different schemes for dividing the dataset into biomes, so a total of nine different realizations. The methods used are fitting curves to time series of averages, which requires no data other than the surface fCO₂, and two methods (multiple linear regression and a feed-forward neural network) that seek regressions of fCO₂ onto sea surface temperature, salinity, mixed layer depth and atmospheric CO₂ mixing ratio. The neural net method provides the best fit of data at a given point, but all methods converge when calculating the flux over sufficiently large regions and when there is sufficient data, e.g. in the northern hemisphere and southern hemisphere after about 2000. The uncertainty on the northern hemisphere sink inherent in the interpolation is +/-10% on a flux into the northern hemisphere oceans that is today in excess of 1 PgC, representing a tight constraint on the global carbon cycle. Before 2000 in the southern hemisphere, results become increasingly method dependent and vary widely, with uncertainties 35-40% of the total sink in that hemisphere. This illustrates that we lack essential observations to constrain the southern hemisphere carbon sink over time.