Constraints on the variability of the oceanic CO2 sink from observations and theory

The ocean is an important sink for carbon, absorbing 27% of the anthropogenic CO₂ emitted to the atmosphere from human activities. The variability of this oceanic CO₂ sink has come under scrutiny recently because data-products that are based on surface ocean CO₂ observations (fCO₂) and various statistical methods estimate twice the variability produced by process-based ocean models. These data-based products (pCO₂ products) also suggest a strong growth of the oceanic CO₂ sink in the past decade not produced by ocean models. Here we develop a hybrid approach that makes use of both fCO₂ observations and an ocean model through a novel two-step optimisation approach. First, the model parameters are optimised to best reproduce fCO₂ observations over the 1970-2019 period. Second, for each year with sufficient observations, the CO₂ flux that minimises the distance to fCO₂ observations is used as best estimate. This approach preserves the coherence of the physical and biogeochemical processes as represented in ocean models, while calibrating the CO₂ fluxes to observations as done in pCO₂ products. This approach confirms the ocean model weaknesses for producing coherent values of pCO₂ in the high-latitude regions. The interannual variability in ocean CO₂ sink after the two-step optimisation remains broadly consistent with the variability produced by the process model after the first optimisation step, but the amplitude of the variability is larger and closer to the one suggested by pCO₂ products. The two-step otpimisation also improves the model’s representation of some documented decadal trends in the oceanic CO₂ sink: its stagnation in 1990s and its reinvigoration in the 2000s. However, the optimisation approach does not support a strong increase of the oceanic CO₂ sink in the past decade as reported by some pCO₂ products. The finding is confirmed with the separate optimisation by latitude bands, where even in the Northern band (> 30°N) where the density of observations is maximum, the two-step optimisation returns a modest trend in the CO₂ sink. In this high-latitude region, the fCO₂ observations also suggest a low increase of the oceanic CO₂ sink in the past decade, while they supported the previous increasing trend in the 2000s. Therefore, although the two-optimisation approach presented here partly reconciles results from process-based ocean models and data-based pCO₂ products, it also suggests that the current generation of pCO₂ products could be sensitive to numerical artefacts that artificially enhance the recent trend.