Global ocean carbon uptake: magnitude, variability and trends

The globally integrated sea–air anthropogenic carbon dioxide (CO 2 ) flux from 1990 to 2009 is determined from models and data-based approaches as part of the Regional Carbon Cycle Assessment and Processes (RECCAP) project. Numerical methods include ocean inverse models, atmospheric inverse models, and ocean general circulation models with parameterized biogeochemistry (OBGCMs). The median value of different approaches shows good agreement in average uptake. The best estimate of anthropogenic CO 2 uptake for the time period based on a compilation of approaches is −2.0 Pg C yr −1 . The interannual variability in the sea–air flux is largely driven by large-scale climate re-organizations and is estimated at 0.2 Pg C yr −1 for the two decades with some systematic differences between approaches. The largest differences between approaches are seen in the decadal trends. The trends range from −0.13 (Pg C yr −1 ) decade −1 to −0.50 (Pg C yr −1 ) decade −1 for the two decades under investigation. The OBGCMs and the data-based sea–air CO 2 flux estimates show appreciably smaller decadal trends than estimates based on changes in carbon inventory suggesting that methods capable of resolving shorter timescales are showing a slowing of the rate of ocean CO 2 uptake. RECCAP model outputs for five decades show similar differences in trends between approaches.

where the kinetic term, k, is called the gas transfer velocity and incorporates all processes that control the kinetics of the gas transfer across the sea-air interface, while the term (C w − C a ) is the concentration gradient of the gas in the liquid boundary layer that is on the order of 100 micron thick. For sea-air CO 2 fluxes the equation is commonly written in terms of the partial pressure (or fugacity) difference across the interface ac-5 cording to: where the convention is that a net flux into the ocean is expressed as negative value. pCO 2w is the partial pressure of CO 2 of surface water, pCO 2a is the partial pressure of CO 2 in air, K 0 is the solubility of CO 2 , and ∆pCO 2 is the partial pressure gradient (∆pCO 2 = pCO 2w − pCO 2a ). The CO 2 levels in air are reported as a mixing ratio or mole fraction, X CO 2a which must be converted to a partial pressure through pCO 2a = X CO 2a (P −p H 2 O ), where P is the ambient pressure and p H 2 O is the saturation pressure of water vapor. An observational approach to estimating the sea-air flux of CO 2 is to make mea- 15 surements of ∆pCO 2 from ships and moorings. Other approaches used to infer global sea-air CO 2 fluxes rely on models, total inorganic carbon measurements in the ocean interior and/or atmospheric data. Of these methods, those relying on simulations with Ocean General Circulation Models (OGCMs) with parameterization of biogeochemical processes calculate the CO 2 flux using Eq. (2). In this case, pCO 2w is computed 20 from the modeled state variables of the carbonate system; total alkalinity (TAlk) and total dissolved inorganic carbon (DIC). The resolution of global biogeochemical models used in RECCAP is on the order 2 by 2 degree with output provided at monthly scales (e.g., Canadell et al., 2012). Oceanic inverse models constrain the regional and global fluxes from interior ocean circulation and ocean interior data based on measurements 25 of DIC and other tracers (Mikaloff Fletcher et al., 2006Gruber et al., 2009), with the advantage of estimating both the natural and anthropogenic CO 2 flux components on decadal scales. The inverse estimates are independent of the estimates 10965 Introduction based on ∆pCO 2 (Eq. 2). Khatiwala et al. (2009Khatiwala et al. ( , 2012 provide estimates of changes in anthropogenic CO 2 in the ocean interior using a Green function with transient tracers that yields the anthropogenic CO 2 uptake estimates at regional scales. Atmospheric inversions use atmospheric transport models and measured atmospheric CO 2 levels to assess sources and sinks of contemporary CO 2 , i.e., the sum of natural and anthro-5 pogenic CO 2 . Faster atmospheric transport and mixing leads to coarser spatial resolution but higher temporal resolution compared to ocean inversions with about a dozen ocean regions (Jacobson et al., 2007). Trends in the atmospheric ratio of O 2 / N 2 along with atmospheric CO 2 levels can be used to separate terrestrial CO 2 uptake from that of oceanic uptake due to reservoir specific fractionation (Manning and Keeling, 2006). 10 Scales of ocean uptake estimates are hemispheric and monthly depending on the level of sophistication of interpretation. Data based estimates of variability and trends in ocean CO 2 uptake are limited by the short record of observations. High quality measurements in the surface water and air commenced in the early 1960's but at limited scope ( Fig. 1). Global physical forcing 15 fields such as wind speeds are available for the last 5 decades, but the older estimates are inconsistent with current measurements due to large changes in observing and interpolation methods.
In this overview the focus is on the last 20 yr (1990-2009) with the recognition that the first decade has data limitations. The emphasis is on the ∆pCO 2 observations and 20 empirical approaches that are based heavily on observations of surface ocean temperature. These are compared with models and observations of anthropogenic CO 2 and other tracers in the ocean interior. The background section is a summary of measured variability and trends as well as current global-integrated carbon cycle inventories and fluxes. The methodology describes an empirical approach to estimate a 20-yr time se- 25 ries of global-integrated sea-air fluxes from the global pCO 2 climatology, sea surface temperature (SST) and wind anomalies over the past two decades, and a procedure to incorporate the effect of increasing atmospheric CO 2 levels. It discusses the different models and other approaches used in the analysis. In the discussion we will focus Introduction on the anthropogenic flux and detail adjustments between the different modeling and observational approaches to get consistent estimates. In the last section the estimates are reconciled for a consistent global anthropogenic CO 2 uptake. It concludes with a summary of the median anthropogenic CO 2 uptake, and the sub-annual and interannual variability of the net sea-air CO 2 fluxes. 5 2 Background

Atmospheric CO 2 variability and trends
The pCO 2 of air (pCO 2a ) is well constrained from measurement of the mole fraction of CO 2 , X CO 2 , in air at about 80 global flask sampling stations worldwide (Conway et al., 1994). Seasonal changes of approximately 10 ppm over the Northern hemisphere 10 ocean are driven by the photosynthesis and respiration cycle of the terrestrial biosphere. Much smaller seasonal changes are observed in the Southern Hemisphere due to the lack of land cover. The atmospheric CO 2 measurements point towards rapid zonal atmospheric mixing (weeks to month) and impedance in the tropics causing slower north-south inter-hemispheric exchanges on the order of a year (Denning Introduction

Oceanic pCO 2w variability and trends
In contrast to the atmospheric pCO 2 , the pCO 2 in the surface ocean (pCO 2w ) is spatially and temporally more variable, and therefore requires several orders of magnitude more data to map variation (Figs. 1 and 2). Seasonal and interannual changes can be 100 µatm or more. The spatial decorrelation length scales are on the order 5 of 100's of km (Li et al., 2005) compared to 1000's of km in the marine atmosphere. The greater variability and challenges in making measurements of pCO 2w means that for large parts of the ocean there are insufficient observations to obtain direct estimates of ∆pCO 2 everywhere (Fig. 2). Only select regions such as the equatorial Pacific, and time-series stations in the sub-tropical North Atlantic (ESTOC and BATS) and sub-tropical North Pacific (HOT) have sufficient measurements and robust interpolation schemes to discern decadal variability and trends based on observations alone. Two approaches have been pursued to overcome this limitation. The most commonly used approach is to collate the data to form a monthly global climatology of ∆pCO 2 (Takahashi et al., 2009), henceforth referred to as T-09. The T-09 climatology is con-15 structed from approximately 3 million data points obtained over the last 40 yr, assuming that the ∆pCO 2 does not vary on multi-year timescales. The climatology is on coarse (4 • latitude × 5 • longitude) resolution, and data have been interpolated over the annual cycle and in space using the mean surface flow fields from an OGCM. The second approach is to interpolate the data in time and space using ancillary observations, such 20 as SST, mixed layer depth and other (remotely sensed) surface parameters. No global estimate is yet available based on this approach, but self-organizing maps (Telszewski et al., 2009) and multi-parameter regressions (Schuster, McKinley et al., 2012) have been used to determine regional ∆pCO 2 fields and fluxes.
For variability and trends over the past two decades for the global ocean we are 25 limited to numerical models and a scheme that utilizes the monthly pCO 2 climatology of T-09 and local-scale empirical relationships of pCO 2w against SST. These empirical Introduction

Global ocean CO 2 uptake estimates
Various "best estimates" of ocean anthropogenic CO 2 uptake have been reported based on a variety of data and methods. The 4th IPCC Assessment (Table 1; Denman 5 et al., 2007) provide estimates for the 1980s and 1990s on the basis of measurements of ∆pCO 2 (Takahashi et al., 2009), ocean inversions of oceanic data (Mikaloff Fletcher et al., 2006), atmospheric inversions, carbon isotopic mass balances (Gruber and Keeling, 2001;Quay et al., 2003) and atmospheric O 2 / N 2 measurements (Bender et al., 2005;Manning and Keeling, 2006), while that for the 2000-2005 period is based on model results.
More recently, long-term trends of ocean uptake of CO 2 determined by models have been reported in the literature (e.g., Sarmiento et al., 2010) and regularly updated as part of the Global Carbon Project (GCP) (Le Quéré et al., 2009;www.globalcarbonproject.org/). The estimates consider mass balances 15 and fluxes among all major labile reservoirs (ocean, atmosphere, and terrestrial biosphere). The annual model-based estimate for the net air-sea CO 2 flux since 1960 is provided in Fig. 3. These results suggest that despite the fact that the ocean sink has increased significantly over the past 50 yr, the increase is slower than the increase in fossil fuel emissions. Thus the percent of fossil fuel emissions absorbed by the oceans 20 had steadily declined.
The treatment of the different air-sea CO 2 flux components in modeling approaches represent a substantial challenge when comparing different flux estimates. Our aim here is to focus on the anthropogenic component of the CO 2 flux, which we compute from estimates of the contemporary CO 2 flux by subtracting the natural CO 2 flux com- 25 ponent. The latter can be estimated if we assume that ocean circulation and biological activity has remained roughly constant over the last 250 yr. In this case, this flux, when integrated over the globe, cancels to zero except for the river-carbon induced 10969 Introduction outgassing flux of CO 2 . Although this assumption provides a good first estimate, we need to recognize that interannual variability and trends in circulation and biogeochemistry cause temporal fluctuations in the natural CO 2 flux components. Most of these deviations are believed to occur on interannual time-scales, such that decadal averages of the natural CO 2 flux component are less than ±0.3 Pg C yr −1 (Lovenduski et al., 2008). The decadal trends in circulation and biogeochemistry are less well known and could cause changes in the decadal trends of sea-air CO 2 fluxes. The contribution of the river-carbon induced outgassing flux of CO 2 amounts to about +0.45 Pg C yr −1 (Jacobson et al., 2007) and needs to be accounted for when comparing model simulation results with estimates based on ∆pCO 2 measurements and Eq. 2.

10
The river efflux is believed to be relatively constant through time, so that a constant offset of 0.5 Pg C yr −1 is applied to the net sea-air CO 2 fluxes based on ∆pCO 2 to obtain the anthropogenic CO 2 flux. Additional adjustments are to compare different estimates and include surface area of the ocean used, sea-ice, and coastal carbon input.

15
The Takahashi et al. (2009) (T-09) ∆pCO 2 climatology is used as an observational benchmark for the net contemporary or net sea-air CO 2 fluxes, and it is the basis for our empirical approach to estimate interannual variability. However, even with over 3 million data points and its coarse resolution, the T-09 pCO 2w climatology is data limited. For much of the ocean, particularly the Southern Hemisphere, the seasonal 20 cycle cannot be fully resolved from measurements alone. As shown in Fig. 2, only in the Northern Hemisphere are there sufficient monthly observations to create a full climatological year. A propagation of errors suggests an uncertainty in the global fluxes from the climatology of 50 %. However, the estimated fluxes are in better than 50 % agreement with independent mass balance and model estimates (e.g., Gruber et al., 25 2009). The adjustments and breakdown of errors are listed in Table 2 along with an updated estimate.
The uncertainty estimates in Table 2  uncertainty in global sea-air CO 2 fluxes is associated with the ∆pCO 2 estimate. This mirrors the conclusion for a regional estimate by Watson et al. (2009) that the large de-correlation length scales of hundreds of kilometers and the large number of measurements in each grid cell increase the certainty in ∆pCO 2 appreciably. However, the uncertainty estimate in ∆pCO 2 does not fully account for the dearth of measure-5 ments in many parts of the ocean. The uncertainty in the gas transfer velocity, k, is based on the range of common parameterizations presented in the literature. Recent syntheses suggest that globally the uncertainty in gas transfer is in the range of 10 to 20 % (Ho et al., 2011). Differences in global wind products are substantial, but this is partially compensated for by normalizing gas transfer-wind speed relationships to 10 match global ocean bomb-14 C inventories (Sweeney et al., 2007;Naegler, 2009). As described above, the uncertainty estimates are not rigorous but rather based on best knowledge. Moreover, some of uncertainties are likely to be systematic and cannot be propagated in the simple fashion shown in Table 2. The largest uncertainty in the global CO 2 flux climatology of T-09 is attributed to 15 the assumption that the surface seawater pCO 2 increases at the same rate as the atmospheric CO 2 levels of ≈ 1.5 ppm yr −1 for the past four decades. The uncertainty estimate of ±0.5 Pg C yr −1 is derived from assuming an uncertainty of ±0.5 µatm yr −1 in the oceanic CO 2 increase accounting for the data distribution in time. Therefore, the assumption that the ∆pCO 2 remains invariant is critical for the climatology. For regional 20 shorter term assessments, where data are not normalized to a common time reference, this uncertainty does not come into play. The rate of CO 2 gas exchange is much faster than exchange of CO 2 between the mixed layer and the waters below (Broecker and Peng, 1982). On a global scale the surface ocean CO 2 levels should keep up with the atmosphere albeit with a lag of about  (Broecker and Peng, 1982). This estimate is similar to a more comprehensive representation in OGCMs when the atmospheric CO 2 is increasing at a rate as observed Introduction  Quéré et al. (2010) show that for several regions of the ocean the ∆pCO 2 does not remain constant over periods of up to three decades. This is attributed to circulation changes, and 5 possibly changes in the biological cycle. The Southern Ocean (Le Lovenduski et al., 2008;Lenton and Matear, 2007;Lenton et al., 2012) and Eastern North Atlantic (Schuster et al., 2007;Omar and Olsen, 2006;Metzl et al., 2010) show that surface water pCO 2w has increased faster than atmospheric CO 2 , thereby decreasing the CO 2 sink. In the Southern Ocean where the changes are attributed to more upwelling of deep-water the changes are believed to be sustained. Increased upwelling is attributed to increases in zonal wind stress caused by large-scale reorganizations of the southern hemisphere climate system in response to global warming and stratospheric ozone loss (Thompson and Solomon, 2002). In other parts of the ocean, where the changes are attributed to more ephemeral causes, no systematic 15 long-term changes in uptake are anticipated, as of yet, other than those caused by increasing atmospheric CO 2 levels.
Winds have a major impact on sea-air CO 2 fluxes through their influence on k (Eq. 2). Long-term global wind records suggest an increase with time (Young et al., 2011) The global wind speed records are either based on atmospheric assimilations com-20 monly used in weather forecasts, ship and buoy based observations, remotely sensed winds, or a combination thereof. Determining accurate trends is challenging because of changes in procedures and inputs. Assimilation model outputs (e.g., NCEP, ECMWF) are re-analyzed, often with a major objective to eliminate procedural biases. The reanalysis products show appreciable global and regional differences in magnitude and 25 variability (Wallcraft et al., 2009). For the RECCAP analysis the cross-calibrated multiplatform (CCMP) winds are used as they address many of the shortcomings of other products for the determination of sea-air CO 2 fluxes. The product is well documented and consistent for the entire time record (Atlas et al., 2011). The CCMP product, which BGD 9, 2012 Global ocean carbon uptake: magnitude, variability and trends R. Wanninkhof et al. covers the time period from 1 January 1990 through 31 December 2009, shows appreciable trends in wind speed over time both regionally and globally. Figure 4 shows the trends of the second moment of the winds, U 2 , used in the analyses (see Eq. 3).
At 90 % significance level the trends show decreases in U 2 in the Subtropical North Pacific and increases in the Southern Ocean and Equatorial Pacific. The increasing 5 winds have a direct effect on the sea-air CO 2 flux through an impact on the gas transfer velocity but also an indirect effect on ∆pCO 2 from changes in ocean circulation and mixed layer dynamics.

Methods
Here we provide details on the bulk flux equation (Eq. 2) input parameters, which are 10 key for surface ocean data-based methods and OGCMs that provide fluxes based on ∆pCO 2 . The procedure to determine a 20-yr time record of fluxes from SST anomalies is provided. The OGCMs, atmospheric and ocean inverse models, and estimates based on atmosphere O 2 / N 2 are described briefly. 15 The global CO 2 flux estimate reported in T-09 used the NCEP-II assimilated wind speed product. The NCEP-II product is inconsistent in magnitude and wind speed pattern over the ocean compared to other products ( ments of the winds were determined from the 6-hourly observations at the spatial resolution of 0.25 • from the data at available at http://podaac.jpl.nasa.gov/DATA CATALOG/ ccmpinfo.html. Using these second moments and an inverse procedure to optimize the inventory of bomb-14 C in the ocean (Sweeney et al., 2007), the coefficient for the gas transfer

Gas transfer velocities and wind speeds
where k is the gas transfer velocity (cm h −1 ), " " denote temporal averages, and U 2 in (m s −1 ) 2 is the time-mean of the second moment of the wind speed at 10-m height.
The coefficient a was adjusted such that the bomb-14 C inventory increase in the ocean 5 corresponded with the atmospheric 14 C history. The optimal coefficient for the gas transfer velocity parameterization is: where Sc is the Schmidt number of CO 2 in seawater at a given SST. The relationship of Sc for CO 2 with SST in seawater is: , 1992). Other relationships between k and wind speed have been proposed but for the prevailing wind speed range from 4-12 m s −1 a quadratic appears appropriate for global scale analyses (Wanninkhof et al., 2009).

Updated gas transfer velocity parameterization and impact on global fluxes
The CCMP wind product averaged over the 4 • × 5 • grid used by T-09 yields a global 15 average, 20-yr mean, wind speed U of 7.6 m s −1 and a second moment, U 2 of 69.1 (m s −1 ) 2 . Using Eq. (4)

Method for estimating decadal variability and trends in ∆pCO 2 using a data based approach
There are limited observational data on global trends and variability in ocean sea-air CO 2 fluxes. An empirical approach first presented in Lee et al. (1998) and improved in Park et al. (2010a), henceforth referred to as P-10, provides an assessment of in-5 terannual variability from seasonal correlations of pCO 2w and SST that are used with the measured interannual SST variability. This approach is applied to the T-09 climatology as follows: The monthly mean sea-air CO 2 flux for each 4 • × 5 • grid cell for an individual year other than the climatological year 2000 is estimated from the global pCO 2w climatology, and pCO 2w anomalies determined from sub-annual pCO 2w -SST 10 relationships. The sub-annual pCO 2w -SST relationships are derived from one to four linear fits of pCO 2w and SST for each of the 1759 4 • × 5 • grid cells in the T-09 climatology. The number of sub-annual segments chosen to delineate the sub-annual trends is kept to the minimum sufficient to characterize the relationship between pCO 2w and SST for each location. The monthly U 2 is from CCMP, and the monthly mean 15 SST is from the NOAA Optimum Interpolation (OI) SST product (Reynolds et al., 2007; (1979-1989, 1990-mid-1998, and mid-1998-2008)  (PDO). Using these relationships the interannual variability based on the annual values for this region is 0.07 Pg C yr −1 (1σ) for the 1990-2009 time period.
The empirical method of P-10 to assess interannual variability is implicitly tied to the ∆pCO 2 climatology referenced to year 2000 such that is cannot reproduce trends tied to increasing atmospheric CO 2 levels. Changes in biogeochemistry of surface seawa-5 ter associated with SST changes will be reflected as long as the same mechanisms that relate sub-annual pCO 2w change to SST control the interannual pCO 2w -SST relationships. A weak increasing trend in pCO 2w , associated with surface ocean warming is estimated by the P-10 method over the 20-yr period. This leads to a reduction in net global ocean CO 2 uptake (see below). 10 The P-10 method implicitly assumes that pCO 2w increases at the same rate as atmospheric CO 2 levels, that is the ∆pCO 2 remains constant over time. The "CO 2 -only" run of NCAR CCSM-3 model (Doney et al., 2009a, b) for the period of 1987-2006 is used to determine local deviations in this trend. This model output is produced using a repeat annual cycle of physical forcing and rising atmospheric CO 2 . In each 4 • × 5 • 15 grid cell, the trend in pCO 2w is computed by a linear regression with de-seasonalized monthly values using a harmonic function. Approximately 75 % of the grid cells have statistically significant positive or negative trends in ∆pCO 2 (p < 0.05) over the past two decades. For the remaining 25 % of the grid cells with no significant trends, we assume that pCO 2w increases at the same rate as atmospheric CO 2 . From these trends 20 the "CO 2 -only" fluxes are determined for each grid cell and are added to CO 2 fluxes from the empirical model to estimate the total flux over 1990-2009. Global maps of the trends of the CO 2 -only output, using a different model but with similar results, can be found in Fig. 4b of Le Quéré et al. (2010). 25 The trends and variability observed in the models result are caused by changing physical forcing and the resulting changes in circulation and biogeochemistry. A list of the OGCMs used in RECCAP is shown in in the discussion are provided at the RECCAP website, with model details given in Canadell et al. (2012). Of the 9 model runs provided in RECCAP, several are from the same model but with different gas exchange or wind forcing. Only the LSC and UEA models include the input of riverine carbon but the input does not appear to contribute to subsequent outgassing. Therefore, the flux estimates from all models used here are 5 the anthropogenic CO 2 component and the natural CO 2 flux component, which without riverine carbon outgassing is globally nearly balanced (≈ 0).

Other estimates of global sea-air CO 2 fluxes
Several other global estimates are compared that rely on changes in atmospheric or oceanic anthropogenic CO 2 inventories. These include results from eleven atmospheric inverse models provided by RECCAP, atmospheric O 2 / N 2 , and ocean inventory changes using transient tracers and a Green function analysis. Atmospheric inversions are based on the interpretation of atmospheric CO 2 gradients, but given the under constrained nature of this inversion, they need prior information for the sea-air CO 2 fluxes, for which CO 2 flux climatologies such as that of T-09 are used as priors. 15 As a result, these estimates are net fluxes and include all flux components, i.e., natural (with river outgassing) and anthropogenic. As they use the observed sea-air flux as a prior they are not a truly independent estimate. In contrast, the O 2 / N 2 approach provides a strong independent constraint for the anthropogenic CO 2 flux component only. Estimates from 1989-2003 can be found in Manning and Keeling (2006); while 20 results from 2000-2010 are presented in Ishidoya et al. (2012). The anthropogenic CO 2 fluxes inferred from changing oceanic inventories are further detailed in Khatiwala et al. (2012). Briefly, they include an empirical Green function approach (Khatiwala et al., 2009(Khatiwala et al., , 2012 and a model-based approach, commonly referred to as the ocean inversion project (OIP) (Mikaloff Fletcher et al., 2006;Gruber et al., 2009). These ap- 25 proaches are based on the assumption of a steady-state ocean circulation and therefore only resolve the smoothly evolving changes in the oceanic uptake of anthropogenic CO 2 driven by the increase in atmospheric CO 2 , i.e., they do not resolve interannual and sub-annual variability.

Global sea-air CO 2 fluxes
A tabular summary of the decadal mean anthropogenic CO 2 uptake centered on the 5 year 2000 and based on several observation based techniques and models is provided in Table 4 with further detail below. Also shown are the interannual variability (IAV), subannual variability (SAV), and trends of the net sea-air CO 2 flux. The interannual and subannual variability is dominated by the natural CO 2 flux, and the trend is primarily caused by the anthropogenic component with modulation by the natural cycle.

Anthropogenic sea-air CO 2 flux estimate based on surface water measurements
The empirical estimates in Table 4 follow the procedures and assumptions of T-09 and P-10. The main differences are the wind speed product used, the use of the second moment of the wind speed, and the coefficient in the gas transfer parameterization. 15 The fluxes are extrapolated to include the coastal regions. Despite using the same constraints and consistent assumptions as T-09, the difference of 0.19 Pg C yr −1 ( Table 2) arises because of spatial differences in wind speed products and cross-correlations between direction of flux and wind (Wanninkhof et al., 2009). The T-09 global anthropogenic CO 2 flux estimate does not include coastal re-20 gions. Fluxes in coastal area are highly variable but net fluxes on the whole appear similar to that of adjacent ocean areas, with some regional exceptions, particularly along Eastern upwelling zones that show large effluxes near the coast, and riverine dominated shelves that show influxes (Chavez et al., 2007;Liu et al., 2010;Cai, 2011). difference in areas as provided in T-09 and the total area as used in the OIP (Supplement A), These corrections yield an anthropogenic CO 2 flux of −2.0 Pg C yr −1 that is the same as the −2.0 Pg C yr −1 presented in T-09. However, as outlined in Table 2 the lower uptake in the open ocean is compensated for by incorporating the coastal areas.
The reduced uncertainty of 0.6 Pg C yr −1 compared to 0.8 Pg C yr −1 in T-09 is due to 5 improvements in the estimate of gas transfer velocities and winds.

Anthropogenic CO 2 uptake estimates based on models and atmospheric observations
Several modeling and atmospheric observing approaches report smaller uncertainties than those relying on measured surface CO 2 levels and bulk flux approaches. An important consideration when determining uncertainties based on multi-model comparisons is that models are often similar, which can lead to an unrealistic good correspondence between them. Graphical summaries of the annual anthropogenic sea-air CO 2 fluxes for the OGCMs and ocean inverse estimates, and the atmospheric inversions are presented in Figs. 6 and 7, respectively. A summary of the medians of different methods 15 is shown in Fig. 8.

Ocean inversion estimates
Ocean inversions rely on inorganic carbon measurements in the ocean interior and can only provide an average anthropogenic sea-air CO 2 flux, interpolated to present day, based on cumulative anthropogenic CO 2 uptake since the preindustrial era. A compre-20 hensive ocean inversion estimate of CO 2 uptake based on a suite of ten general ocean circulation models is presented in Gruber et al. (2009). This inversion value is independent from the bulk sea-air CO 2 flux estimate but contains other potential sources of error associated with the partitioning of anthropogenic from natural dissolved inorganic carbon as well as the potential for biases due to inaccurate estimates of ocean Introduction determined from the spread of model results. The uncertainty in ocean transport and mixing is the largest source of uncertainty for the ocean inversion. The average anthropogenic CO 2 uptake by the inversions provided in RECCAP is an update of Mikaloff Fletcher et al. (2006) and Gruber et al. (2009) 20 A comprehensive synthesis of model performance on a previous generation of carbon cycle OGCMs was provided as part of the OCMIP (Ocean Carbon-Cycle Model Intercomparison Project) in the early 2000s (see e.g., Doney et al., 2004). A more recent subset of models was used to determine the ocean sink for 1959-2008 (Le Quéré et al., 2009;Sarmiento et al., 2010). Initialization, forcing and biological representation differs for the models (Canadell et al., 2012). The models were run in hindcast mode, i.e., they were driven by atmospheric forcing assimilation products such as NCEP or ECWMF, and prescribed atmospheric CO 2 concentration using an observation based product such as GLOBALVIEW-CO2 (2011). Two things stand out in the time series of annual CO 2 fluxes from OGCMs (Fig. 6). The models with a heritage traceable to the NCAR community ocean circulation model with biogeochemistry (BEC, ETH k15 , and ETH k19 ) show about 0.5 Pg C yr −1 less up-5 take than the others. The UEA models are adjusted to the best observationally-based estimate of ocean uptake in the 1990s and therefore show a mean of anthropogenic CO 2 uptake about −2.2 Pg C yr −1 . The second significant difference is the response of the models to major climate reorganizations. The 1997/1998 El Niño was one of the largest on record (see e.g., http://www.esrl.noaa.gov/psd/enso/mei/) with an expected 10 net increase in ocean uptake due to reduced outgassing of CO 2 in the equatorial Pacific (Feely et al., 2006). The models show greatly varying responses in CO 2 fluxes in phasing, magnitude and duration. The NCAR models show a peak-to-peak change of 0.2 Pg C yr −1 while the UEA models show peak-to-peak changes of up to 1.0 Pg C yr −1 .

Ocean General Circulation Models with biogeochemistry (OGCMs)
The Bergen model shows no global response to the El Niño while the CSI shows an For the estimate of the median anthropogenic CO 2 uptake of the OGCMs, we excluded model outputs that were largely similar and only differed in their forcing. That is, only one of the three UEA model outputs (UEA NCEP ) and one of the two ETH (ETH k15 ) 20 model outputs were used. The solid red line in Fig. 6 is the annual median of LSC, UEA NCEP , CSI, BER, BEC and ETH k15 runs. The interannual variability of the median estimate of 0.16 (1σ) Pg C yr −1 is damped compared the mean interannual variability of the individual models of 0.25 (1σ) Pg C yr −1 indicating that the variability in the individual models is not coherent. The median estimates of the 6 models shows significant 25 trend of −0.14 ± 0.02 Pg C yr −1 decade −1 (p < 0.01) in anthropogenic CO 2 uptake. This trend agrees with the trends of the individual models. Introduction

Atmospheric inverse models
Ocean anthropogenic uptake determined from atmospheric inverse models shows a similar range as the uptake in the OGCMs with a median value of −2.1 Pg C yr −1 (Table 4; Fig. 8). This is similar to the median of the OGCMs and the flux is 0.3 Pg C yr −1 less than that of the ocean inversion model. The interannual variability is considerably 5 greater (0.40 Pg C yr −1 , 1σ of detrended median values of de-seasonalized monthly anomalies).
As with the OGCMs, the response of the fluxes from the atmospheric inversions to the 1998 El Niño varies appreciably from no impact to about 1.0 Pg C yr −1 . However, the atmospheric inversions that give a response show the maximum uptake a year earlier 10 (1997). The atmospheric inversions separate the seasonal and interannual CO 2 flux of the ocean from that of the terrestrial biosphere which show sub-annual variability in CO 2 uptake and release that is up to orders of magnitude greater. Therefore the uncertainty of the separation between ocean and terrestrial carbon fluxes dictates the variability in the ocean. The time-averaged global-integrated anthropogenic CO 2 flux 15 from the atmospheric inversions is not independent since the ∆pCO 2 fields used in the CO 2 flux climatology are used as a prior within the inversion. No estimate of the decadal trend is provided as the model runs from which the median estimate is determined differ in time span and time period. 20 The trends and variability in fluxes as determined from the bulk flux equation (Eq. 2) are driven by changes in wind and ∆pCO 2 . The trends and variability are discussed in the context of the empirical approach in Park et al. (2010a). The impact of changing winds is different for OGCMs as they are dynamic in nature and higher winds will not only change the rate of gas exchange, but also the ∆pCO 2 field. This can result in Introduction  Lovenduski et al., 2008). In contrast, applying higher winds to a static ∆pCO 2 field such as the T-09 climatology will result in increased uptake.

Sub-annual (seasonal) variability: seasonal and regional patterns
Interannual variability and trends in CO 2 fluxes are masked by large sub-annual (seasonal) changes on regional scales in ∆pCO 2 and in wind. Thus, the detection of inter-5 annual and decadal changes requires a long time-series of measurements. Moreover, variability will differ for different ocean regions as detailed in the chapters of individual basins. Here we provide a zonal analysis of mean and spatial-temporal variability for the global ocean of wind, ∆pCO 2 , and the sea-air CO 2 flux density, i.e., the flux per unit area. Figure 9 shows the 20-yr zonal mean and standard deviation of zonal • wind speeds decrease slightly. The variance pattern of winds, expressed as the standard deviation in the 20-yr mean in Fig. 9a, differs appreciably between the northern and southern hemisphere. There is appreciably more variability in winds in the mid-to high-latitudes (> 30 • ) in the Northern Hemisphere than in the Southern Hemisphere. 20 Variability is at a minimum in the high-pressure regimes in the sub-tropics, and at the equator. The annual mean ∆pCO 2 from the T-09 climatology (Fig. 9b) shows maximum values just south of the equator and a decline to the north and south. The in the Northern Hemisphere is appreciably greater than in the Southern Hemisphere, as is the case with the wind speeds as well. The flux density, computed from 20-yr wind speed record and the climatological pCO 2 , mirrors the trends in ∆pCO 2 but with amplification at mid-latitudes due to higher winds. The largest sink per unit area is in the Northern Hemisphere, but the substan-5 tially larger surface area in the Southern Hemisphere causes a much greater spatially integrated mass exchange (see, Table 6 in T-09). Because the variance in wind speed and ∆pCO 2 have the same pattern, the flux density at mid-to high-latitudes in the Northern Hemisphere show twice the variance in their sub-annual patterns compared to the Southern Hemisphere. 10 The greater sub-annual CO 2 flux variability in the Northern Hemisphere compared to the Southern Hemisphere is likely because of the significantly greater landmasses in the north. The land-ocean contrast, along with continental input and bathymetry causes greater variability in weather systems, (micro) nutrient input and seasonal temperature variations in the North, all which contribute to higher variability in sea-air CO 2 flux 15 density.
The T-09 climatology excludes the coastal ocean (approximately < 200 km from land) such that much of the area with particularly high variability in ∆pCO 2 and fluxes are excluded (e.g., Borges et al., 2005;Chavez et al., 2007;Liu et al., 2010;Cai, 2011). Several open ocean regions are anomalous compared to the zonal view as described 20 in the ocean basin chapters. These include high sub-annual variability in the seasonal upwelling region in the Arabian Sea caused by the monsoonal winds. High standard deviations from the annual mean of the sea-air CO 2 fluxes are also found in the eastern tropical South Pacific, again attributed to variations in upwelling along the western boundary, and in the southwest Atlantic near the Malvinas confluence area that is im-25 pacted by variation in currents, shallow bathymetry and associated plankton blooms. Low sub-annual variability is seen in the subtropical and sub-polar region of the South-East Pacific and Eastern tropical and subtropical South Pacific.

Interannual variability
The patterns of sub-annual variability differ from the interannual variability for parts of the oceans as shown in Supplement B. There is high sub-annual and interannual variability in the North Atlantic; in contrast, the Equatorial Pacific shows relatively small sub-annual variability but large interannual variability caused by the ENSO that op-

15
The pCO 2w tracks atmospheric pCO 2a increases on longer timescales because the surface ocean is closely coupled with the lower atmosphere through rapid sea-air gas exchange and relatively slower exchange between the surface mixed layer and waters below. Longer-term, secular trends in ∆pCO 2 can arise from changes in circulation and upwelling patterns, and associated changes in biological productivity. The longer-term 20 trends are impacted by the rapid rise in atmospheric CO 2 , finite uptake capacity of the surface mixed layer, and decreasing buffer capacity. In a fractional sense, these factors lead to a diminished uptake of anthropogenic CO 2 by the ocean while in absolute sense the amount of fossil fuel derived CO 2 taken up by the ocean increases (Fig. 3). This robust result is apparent in most models. 25 The trend in the global annual flux for the empirical approach, not accounting for atmospheric CO 2 increase, is 0.07 ± 0.06 (Pg C yr −1 ) decade −1 (Figs. 10a and 11d). This decreasing uptake, in the absence of an atmospheric CO 2 increase, suggests that a strengthening of ocean CO 2 sources or weakening of CO 2 sinks, due to changes of ∆pCO 2 or wind speed. The second moment of the wind speed increases by 0.32 ± 0.04 (m s −1 ) 2 per year, which translates into about a 0.2 m s −1 increase per decade (Fig. 11a). Global SST as determined from the NOAA OI SST product 5 (Reynolds et al., 2007) shows an increasing trend of 0.08 ± 0.03 ( • C yr −1 ) decade −1 (Fig. 11b). The trends in SST impact the ∆pCO 2 and wind will impact gas transfer in the P-10 approach. An analyses holding the SST or wind constant in the P-10 approach indicate that the changes in the wind and SST can act synergistically or antagonistically on the trend in fluxes depending on region. For example, at high Northern latitudes the 10 trends in winds and SST cause a larger CO 2 sink over the past two decades; in the Equatorial Pacific the impact of winds and SST cause a larger CO 2 source; while in the Southern Ocean the change in ∆pCO 2 due to changing SST will decrease the sink while the winds work in an opposing fashion. The global trends in sea-air CO 2 flux are strongly influenced by significant interan-15 nual variability in ∆pCO 2 (Fig. 11c). In particular, the large El Niños in 1992/1993 and 1998 decreased the equatorial ∆pCO 2 and thus has the net effect of increasing the global-integrated net CO 2 flux into the ocean. This suggests that climate reorganizations on multi-annual timescales have an appreciable impact on the trends over the 2-decade time period.

Summary of global anthropogenic sea-air CO 2 fluxes and trends in models
Most models show trends of increasing anthropogenic CO 2 uptake from 1990 through 2009 (or 2007). The median for the empirical approach, OGCMs, and atmospheric inversions all show a consistent change in the anthropogenic CO 2 uptake of −0.13 to −0.15 (Pg C yr −1 ) decade −1 over the time period, that is the same 25 as that inferred from the approach in P-10 when including the atmospheric CO 2 increases. Khatiwala et al. (2009Khatiwala et al. ( , 2012  Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | −0.35 (Pg C yr −1 ) decade −1 based on inventory changes in anthropogenic CO 2 using a Green function formulated from transient tracers. By construction, the Green function approach uptake is smooth, monotonic over time, and proportional to the atmospheric CO 2 increase (Fig. 8).
The median anthropogenic sea-air CO 2 fluxes for the different models for 1990-5 2009 provided in Table 4 show a good agreement despite different approaches and constraints. The median values are centered on, or determined for, 2000, and they are within their estimated uncertainties. The approaches differ appreciably in their interannual variability (IAV), sub-annual variability (SAV) and trend. The decadal trends differ by a factor of four. Approaches that implicitly or explicitly incorporate contempo-10 rary changes in physics and biogeochemistry show much smaller trends than those that use ocean interior data under assumptions of a linear response of the ocean uptake to the atmospheric perturbation, and assumed constant ocean circulation. It is not possible to conclusively determine the best trend estimate given the methodological imperfections of each approach.

15
The relatively short time scale to determine the trends and large, and variable interannual and sub-annual variability can appreciably bias the decadal trends determined. The IAV, SAV are appreciable compared to the trends and therefore the method of inferring the trends from the OGCMs, and atmospheric and ocean inverse medians can also impact the results. Here, IAV values are calculated as a standard deviation of the 20 12 monthly median values from de-seasonalized monthly anomalies. SAV values are estimated as a standard deviation of 12 monthly mean values.
While median uptakes for different methods agree, there are appreciable differences within different types of models. Here we compare the OGCMs. Some the model differences are due to different forcing, transport or representation of the ocean carbon There are also appreciable differences in the interannual and sub-annual variability for the OGCM outputs (Table 5). Sub-annual variability is a standard deviation of 20-yr mean monthly values. Interannual variability is a standard deviation of the detrended 5 values of de-seasonalized monthly anomalies.