Quantifying Errors in Observationally Based Estimates of Ocean Carbon Sink Variability

Abstract

The ocean significantly modulates atmospheric CO 2 , having absorbed approximately 38% of industrial-age fossil carbon emissions (Friedlingstein et al., 2020). Under high emission scenarios, the ocean sink is projected to grow and become the primary sink for anthropogenic carbon emissions over the next several centuries (Randerson et al., 2015). Under low emission scenarios, such as those that would limit global warming to 2°C, the ocean carbon sink will decline rapidly as the near-surface waters that hold the bulk of anthropogenic carbon (Gruber et al., 2019) come into equilibrium with the atmosphere (Cox, 2019;Jones et al., 2016). As the long-term response to the changing atmospheric pCO 2 unfolds, the ocean sink will continue to be modified on seasonal to decadal timescales by climate variability and change. Ultimately, our ability to accurately monitor the fate of anthropogenic carbon in the Earth system requires a quantification of the spatially resolved variability of the ocean carbon sink on timescales from seasonal to multidecadal. To achieve this goal, global maps of surface ocean pCO 2 are required, from which air-sea CO 2 exchange can be derived.The direction of the air-sea CO 2 flux is set by the gradient in pCO 2 across the air-sea interface with additional controls from the gas transfer velocity and CO 2 solubility setting the magnitude. Satellites cannot directly measure surface ocean pCO 2 ; therefore, hindcast simulations with ocean models (Friedlingstein et al., 2020) and observation-based gap-filling techniques (Rödenbeck et al., 2015) are integral to providing a global

SeaFlux: harmonization of air–sea CO<sub>2</sub> fluxes from surface <i>p</i>CO<sub>2</sub> data products using a standardized approach

“…( 1): k w is the gas transfer velocity (further discussed in Sect. 2.3); sol is the solubility of CO 2 in seawater (in units of mol m −3 µatm −1 ), calculated using the formulation by Weiss (1974), near-surface EN4 salinity (Good et al, 2013), NOAA Optimum Interpolation Sea Surface Temperature V2 (OISSTv2) (Reynolds et al, 2002), and European Centre for Medium-Range Weather Forecasts (ECMWF) ERA5 sea level pressure (Hersbach et al, 2020); ice is the sea-ice fraction from NOAA OISSTv2 (Reynolds et al, 2002); pCO 2 is the partial pressure of oceanic CO 2 (in µatm) for each observation-based product after filling, as discussed in Sect. 2.1; and pCO atm 2 is the dry air mixing ratio of atmospheric CO 2 (xCO 2 ) from the ESRL surface marine boundary layer CO 2 product available at https://www.…”

The sensitivity of pCO<sub>2</sub> reconstructions in the Southern Ocean to sampling scales: a semi-idealized model sampling and reconstruction approach

“…(Landschützer et al, 2013(Landschützer et al, , 2016Gregor et al, 2019;Denvil-Sommer et al, 2019). The choice of the GBDT approach is motivated by its achievement of state-of-the-art performances in many ML tasks (Ke et al, 2017), and also the success of GBDT's previous approaches (Gregor et al, 2019;Gloege et al, 2021;Gregor and Gruber, 2021). We use the Scikit-learn and LightGBM Python packages for our implementation of FNN and GBM, respectively.…”

“…The reconstruction power of the surface ocean 𝑝CO ! of the full experimental domain are thus estimated using a series of four statistical metrics that include the mean bias error (MBE), mean absolute error (MAE), root mean square error (RMSE), and Pearson's correlation coefficient (𝑟) to measure the tendency or strength of estimates and observations to vary together (Stow et al, 2009) or, more technically, to quantify the level at which reconstruction captures the phasing observed in the model truth (Gloege et al, 2021).…”

“…observations are thus intrinsically biased by the sampling limitations in space and time for the past several decades covering only ~2% of all the monthly 1º observational grid points (Sabine et al, 2013;Bakker et al, 2016). Mainly due to its remoteness and harsh weather especially during winter, it has been increasingly shown that the SO is the ocean region that contributes the most to these uncertainties in the contemporary estimates of mean annual CO2 uptake (Ritter et al, 2017;Gregor et al, 2019;Bushinsky et al, 2019;Gloege et al, 2021). For instance, sparse observations in largely inaccessible SO areas, particularly during wintertime, have been the biggest barrier in constraining the seasonal and inter-annual variability of regional and global contemporary ocean-atmosphere CO2 exchange (Monteiro et al, 2015;Rödenbeck et al, 2015;Bakker et al, 2016;Ritter et al, 2017).…”

“…These advances have allowed the density of the Southern Ocean surface CO2 observing networks to increase. Consequently, the problem of general sparseness in observations and particularly the sampling biases (Monteiro et al, 2015;Gloege et al, 2021) was partially addressed but not resolved by the ocean CO2 in-situ observations community (Bushinsky et al, 2019;Sutton et al, 2021). For example, under-sampling in winter by ships has been addressed by the 10-day resolution SOCCOM profiling floats and/or Lagrangian platforms that are carried zonally by water currents (Majkut et al, 2014;Williams et al, 2017;Gray et al, 2018;Bushinsky et al, 2019).…”