Recent variability of the global ocean carbon sink

Abstract

We present a new observation-based estimate of the global oceanic carbon dioxide (CO2) sink and its temporal variation on a monthly basis from 1998 through 2011 and at a spatial resolution of 1×1. This sink estimate rests upon a neural network-based mapping of global surface ocean observations of the partial pressure of CO2 (pCO2) from the Surface Ocean CO2 Atlas database. The resulting pCO2 has small biases when evaluated against independent observations in the different ocean basins, but larger randomly distributed differences exist particularly in high latitudes. The seasonal climatology of our neural network-based product agrees overall well with the Takahashi et al. (2009) climatology, although our product produces a stronger seasonal cycle at high latitudes. From our global pCO2 product, we compute a mean net global ocean (excluding the Arctic Ocean and coastal regions) CO2 uptake flux of −1.42 ± 0.53 Pg C yr−1, which is in good agreement with ocean inversion-based estimates. Our data indicate a moderate level of interannual variability in the ocean carbon sink (±0.12 Pg C yr−1, 1𝜎) from 1998 through 2011, mostly originating from the equatorial Pacific Ocean, and associated with the El Nino–Southern Oscillation. Accounting for steady state riverine and Arctic Ocean carbon fluxes our estimate further implies a mean anthropogenic CO2 uptake of −1.99 ± 0.59 Pg C yr−1 over the analysis period. From this estimate plus the most recent estimates for fossil fuel emissions and atmospheric CO2 accumulation, we infer a mean global land sink of −2.82 ± 0.85 Pg C yr−1 over the 1998 through 2011 period with strong interannual variation

Reducing uncertainties in decadal variability of the global carbon budget with multiple datasets

“…34. The correlations between the uncertainties in the two data-driven estimates of O (17,18) are estimated from the series of annual fluxes of the two products by assuming that the correlation in annual fluxes within each 5-y period is an approximation of the 5-y mean flux error correlation.…”

“…(17,18,20). The constraining data tiers (Table S2) were defined as follows: tier 1, direct carbon observations (e.g., CGR); tier 2, indirect carbon observations unambiguously related to carbon quantities (e.g., O 2 /N 2 ); tier 3, direct carbon observations with an empirical (data-driven) model used to obtain global estimates, for example, the use of geostatistics to up-scale local data into global values; tier 4, indirect carbon observations not simply related to global carbon flux quantities.…”

South African carbon observations: CO2 measurements for land, atmosphere and ocean

“…The relationship is then applied to remotely sensed data for which there are no CO 2 measurements, to improve CO 2 data coverage. This approach has shown some promising potential in the North Atlantic where data coverage is more extensive 31,32 , and has also been extended to the Southern Ocean 33 . Furthermore, the approach has more recently been refined by using artificial neural networks to highlight the importance of input parameters and self-organising maps, to illustrate the usefulness of empirical models as tools to reduce uncertainty of CO 2 estimates.…”

In situ measurements of atmospheric O2 and CO2 reveal an unexpected O2 signal over the tropical Atlantic Ocean

“…The global ocean is a major long-term sink for anthropogenic carbon dioxide (CO 2 ) [Le Quéré et al, 2016], and yet the spatial and temporal variability of the oceanic CO 2 sink has large uncertainties [Landschützer et al, 2014;Rödenbeck et al, 2015]. High precision atmospheric oxygen (O 2 ) and CO 2 data can be combined to isolate the ocean influences on atmospheric O 2 variability, by removing the influence of the terrestrial biosphere on O 2 .…”