A Model‐Based Evaluation of the Inverse Gaussian Transit‐Time Distribution Method for Inferring Anthropogenic Carbon Storage in the Ocean

The Inverse Gaussian approximation of transit time distribution method (IG‐TTD) is widely used to infer the anthropogenic carbon (Cant) concentration in the ocean from measurements of transient tracers such as chlorofluorocarbons (CFCs) and sulfur hexafluoride (SF6). Its accuracy relies on the validity of several assumptions, notably (i) a steady state ocean circulation, (ii) a prescribed age tracer saturation history, e.g., a constant 100% saturation, (iii) a prescribed constant degree of mixing in the ocean, (iv) a constant surface ocean air‐sea CO2 disequilibrium with time, and (v) that preformed alkalinity can be sufficiently estimated by salinity or salinity and temperature. Here, these assumptions are evaluated using simulated “model‐truth” of Cant. The results give the IG‐TTD method a range of uncertainty from 7.8% to 13.6% (11.4 Pg C to 19.8 Pg C) due to above assumptions, which is about half of the uncertainty derived in previous model studies. Assumptions (ii), (iv) and (iii) are the three largest sources of uncertainties, accounting for 5.5%, 3.8% and 3.0%, respectively, while assumptions (i) and (v) only contribute about 0.6% and 0.7%. Regionally, the Southern Ocean contributes the largest uncertainty, of 7.8%, while the North Atlantic contributes about 1.3%. Our findings demonstrate that spatial‐dependency of Δ/Γ, and temporal changes in tracer saturation and air‐sea CO2 disequilibrium have strong compensating effect on the estimated Cant. The values of these parameters should be quantified to reduce the uncertainty of IG‐TTD; this is increasingly important under a changing ocean climate. Key Points Uncertainty of the IG‐TTD method is evaluated using model simulated anthropogenic carbon and transient tracers Globally a lower and upper limit of uncertainty of 7.8% and 13.6% is found, smaller than previous estimates Regional and parameter contributions to the uncertainty are quantified, and potential improvements of the method are suggested