Sensitivity analysis of the main methods (ΔC*, ΦCTO, TTD) used to infer the anthropogenic carbon (Cant) along the RAPID line (26°N North Atlantic latitudinal transect) on 2010.

Since the beginning of the industrial revolution, human activities have represented a source for atmospheric CO₂ (C_ant), leading the increase of this gas from 280 to 400 ppm. Currently, the global ocean is sequestrating a quarter of this emission (~2.2 PgC/yr), particularly in key areas, such as the North Atlantic. However, a direct C_ant measurement is not possible owing to its small fraction compared to the natural background. Several methods have been, then, evolved (ΔC*, ΦCTO, TTD) to derive this variable from other carbon-related measurements (alkalinity, DIC, nutrients) or using the transient tracers (CFCs and sulphur hexafluoride (SF₆)). All of these methods rely on assumptions (e.g.: oceanic steady state, constant Redfield’s ratios) that introduce an uncertainty estimated as 20%. Here, a double approach based on the one at a time (OAT) and the variance-based sensitivity (VBSA) analyses has been used to assess this uncertainty and compare it among the three methods. Overall, the back-calculations (ΔC*, ΦCTO) result driven by alkalinity, DIC, temperature and oxygen with low inter parameter interactions. Their influence on the C_ant distribution displays an inversion for temperature lower than 9°C, although the ΦCTO compensates this challenge using an optimum multi parameters analysis below 5°C. By contrast, the TTD approach shows constancy in the predictive parameter influence along the water column, a strong dependency on the CO₂ inferred from the transit time distributions with high inter parameter interactions. To quantify the resulting variation on the C_ant estimates, each predictive factor influence has been correlated with the respective analysis precision and the maximum influence determined as a percentage variation on the C_ant distribution. Results display that only the DIC, alkalinity and nutrients have an influence on the back-calculations comparable with the method uncertainty, highlighting the TTD approach as the most promising in the C_ant studies.