Uptake and Storage of Carbon Dioxide in the Ocean : The Global C 02 Survey

Introduction Human activity is rapidly changing the composition of the earth's atmosphere, contributing to warming from excess carbon dioxide (CO2) along with other trace gases such as water vapor, chlorofluorocarbons, methane and nitrous oxide. These anthropogenic "greenhouse gases" play a critical role in controlling the earth's climate because they increase the infrared opacity of the atmosphere, causing the surface of the planet to warm. The release of CO2 from fossil fuel consumption or the burning of forests for farming or pashare contributes approximately 7 petagrams of carbon (1 Pg C = 1 x 10 '5 g C) to the atmosphere each year. Approximately 3 Pg C of this "anthropogenic COd' accumulates in the atmosphere annually, and the remaining 4 Pg C is stored in the terrestrial biosphere and the ocean. Where and how land and ocean regions vary in their uptake of CO2 from year to year is the subject of much scientific research and debate. Future decisions on regulating emissions of greenhouse gases should be based on more accurate models of the global cycling of carbon and the regional sources and sinks for anthropogenic CO2, models that have been adequately tested against a well-designed system of measurements. The construction of a believable present-day carbon budget is essential for the reliable prediction of changes in atmospheric CO2 and global temperatures from available emissions scenarios. The ocean plays a critical role in the global carbon cycle as a vast reservoir that exchanges carbon rapidly with the atmosphere, and takes up a substantial portion of anthropogenically-released carbon from the atmosphere. A significant impetus for carbon cycle research over the past several decades has been to achieve a better understanding of the ocean's role as a sink for anthropogenic CO2. There are only three global reservoirs with exchange rates fast enough to vary significantly on the scale of decades to centuries: the atmosphere, the terrestrial biosphere and the ocean. Approximately 93% of the carbon is located in the ocean, which is able to hold much more carbon than the other reservoirs because most of the CO2 that diffuses into the oceans reacts with seawater to form carbonic acid and its dissociation products, bicarbonate and carbonate ions (Figure 1). Our present understanding of the temporal and spatial distribution of net CO2 flux into or out of the ocean is derived from a combination of field data, which is limited by sparse temporal and spatial coverage, and model results, which are validated by comparisons with the observed distributions of tracers, including natural carbon-14 (14C), and anthropogenic chlorofluorocarbons, tritium (3H) and bomb 14C. The latter two radioactive tracers were introduced into the atmosphere-ocean system by atomic testing in the mid 20 ~ century. With additional data from the recent global survey of CO2 in the ocean (1991-1998), carried out cooperatively as part of the Joint Global Ocean Flux Study (JGOFS) and the World Ocean Circulation Experiment (WOCE) Hydrographic Program, it is now possible to characterize in a quantitative way the regional uptake and release of CO2 and its transport in the ocean. In this paper, we summarize our present understanding of the exchange of CO2 across the air-sea


Introduction
Human activity is rapidly changing the composition of the earth's atmosphere, contributing to warming from excess carbon dioxide (CO2) along with other trace gases such as water vapor, chlorofluorocarbons, methane and nitrous oxide.These anthropogenic "greenhouse gases" play a critical role in controlling the earth's climate because they increase the infrared opacity of the atmosphere, causing the surface of the planet to warm.The release of CO2 from fossil fuel consumption or the burning of forests for farming or pashare contributes approximately 7 petagrams of carbon (1 Pg C = 1 x 10 '5 g C) to the atmosphere each year.Approximately 3 Pg C of this "anthropogenic COd' accumulates in the atmosphere annually, and the remaining 4 Pg C is stored in the terrestrial biosphere and the ocean.
Where and how land and ocean regions vary in their uptake of CO2 from year to year is the subject of much scientific research and debate.Future decisions on regulating emissions of greenhouse gases should be based on more accurate models of the global cycling of carbon and the regional sources and sinks for anthropogenic CO2, models that have been adequately tested against a well-designed system of measurements.The construction of a believable present-day carbon budget is essential for the reliable prediction of changes in atmospheric CO2 and global temperatures from available emissions scenarios.
The ocean plays a critical role in the global carbon cycle as a vast reservoir that exchanges carbon rapidly with the atmosphere, and takes up a substantial portion of anthropogenically-released carbon from the atmosphere.A significant impetus for carbon cycle research over the past several decades has been to achieve a better understanding of the ocean's role as a sink for anthropogenic CO2.There are only three global reservoirs with exchange rates fast enough to vary significantly on the scale of decades to centuries: the atmosphere, the terrestrial biosphere and the ocean.Approximately 93% of the carbon is located in the ocean, which is able to hold much more carbon than the other reservoirs because most of the CO2 that diffuses into the oceans reacts with seawater to form carbonic acid and its dissociation products, bicarbonate and carbonate ions (Figure 1).
Our present understanding of the temporal and spatial distribution of net CO2 flux into or out of the ocean is derived from a combination of field data, which is limited by sparse temporal and spatial coverage, and model results, which are validated by comparisons with the observed distributions of tracers, including natural carbon-14 (14C), and anthropogenic chlorofluorocarbons, tritium (3H) and bomb 14C.The latter two radioactive tracers were introduced into the atmosphere-ocean system by atomic testing in the mid 20 ~ century.With additional data from the recent global survey of CO2 in the ocean (1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998), carried out cooperatively as part of the Joint Global Ocean Flux Study (JGOFS) and the World Ocean Circulation Experiment (WOCE) Hydrographic Program, it is now possible to characterize in a quantitative way the regional uptake and release of CO2 and its transport in the ocean.In this paper, we summarize our present understanding of the exchange of CO2 across the air-sea

Background
The history of large-scale CO2 observations in the ocean date back to the 1970s and 1980s.Measurements of the partial pressure of CO2 (pCO~), total dissolved inorganic carbon (DIC) and total alkalinity (AT) were made during the global Geochemical Ocean Sections (GEOSECS) expeditions between 1972 and 1978, the Transient Tracers in the Oceans (TTO) North Atlantic and Tropical Atlantic Surveys in 1981-83, the South Atlantic Ventilation Experiment (SAME) from 1988-1989, the French Southwest Indian Ocean experiment, and numerous other smaller expeditions in the Pacific and Indian Oceans in the 1980s.These studies provided marine chemists with their first view of the carbon system in the global ocean.These data were collected at a time when no common reference materials or standards were available.As a result, analytical differences between measurement groups were as large as 29 ~lrnol kg -1 for both DIC and Av which corresponds to more than 1% of the ambient values.Large adjustments had to be made for each of the data sets based on deepwater comparisons at nearby stations before individual cruise data could be compared.These differences were often nearly as large as the anthropogenic CO2 signal that investigators were trying to determine (Gruber et al., 1996).Nevertheless, these early data sets made up a component of the surface ocean pCO2 measurements for a global climatology and also provided researchers with new insights into the distribution of anthropogenic CO2 in the ocean, particularly in the Atlantic Ocean.
At the onset of the Global Survey of CO2 in the Ocean (Figure 2), several events took place in the United States and in international CO2 measurement communities that significantly improved the overall precision and accuracy of the large-scale measurements.In the United States, the CO2 measurement program was co-funded by the Department of Energy (DOE), the National Oceanic and Atmospheric Administration (NOAA) and the National Science Foundation (NSF) under the technical guidance of the U.S. CO2 Survey Science Team.This group of academic and government scientists adopted and perfected the recently developed coulometric titration method for DIC determination that had demonstrated the capability to meet the required goals for precision and accuracy.They advocated the development and distribution of certified reference materials (CRMs) for DIC, and later for A~, for international distribution under the direction of Andrew Dickson of Scripps Institution of Oceanography (see sidebar).They also supported a shore-based intercomparison experiment under the direction of Charles Keeling, also of Scripps.Through international efforts, the development of protocols for CO2 analyses were adopted for the CO2 survey.The international partnerships fostered by JGOFS resulted in several intercomparison CO2 exercises hosted by France, Japan, Germany and the United States.Through these and other international collaborative programs, the measurement quality of the CO2 survey data was well within the measurement goals of _+ 3 ~mol kg -1 and + 5 ~rnol kg -1, respectively, for DIC and AT.Several other developments significantly enhanced the quality of the CO2 data sets during this period.New methods were developed for automated underway and discrete pCO2 measurements.An extremely precise method for pH measurements based on spectrophotometry was also developed by Robert Byrne and his colleagues at the University of South Florida.These improvements ensured that the internal consistency of the carbonate system in seawater could be tested in the field whenever more than two components of the carbonate system were measured at the same location and time.This allowed several investigators to test the overall quality of the global CO2 data set based upon CO2 system thermodynamics.Laboratories all around the world contributed to a very large and internally consistent global ocean CO2 data set determined at roughly 100,000 sample locations in the Atlantic, Pacific, Indian and Southern oceans (Figure 2).The data from the CO2 survey are available through the Carbon Dioxide Information and Analysis Center (CDIAC) at Oak Ridge National Laboratory as Numeric Data Packages and on the World Wide Web (http://cdiac.esd.ornl.gov/home.html).Taro Takahashi and his collaborators have also amassed a large database of surface ocean pCO2 measurements, spanning more than 30 years, into a pCO2 climatology for the global ocean (Takahashi et al., in press).These data have been used to determine the global and regional fluxes for CO2 in the ocean.

C02 Exchange Across the Air-Sea Interface
In seawater, CO2 molecules are present in three major forms: the undissociated species in water, [CO2]aq, and two ionic species, [HCO3-] and [COd-] (Figure 1).The concentration of [CO~]aq depends upon Andrew G. Dickson Scripps Institution of Oceanography Universily of California, San Diego, La Jolla California USA High-quality measurements of carbon dioxide (CO2) in the ocean have been an integral part of JGOFS.Despite their importance for understanding the oceanic carbon cycle, measurements made by different groups were rarely comparable in the past.A significant contribution of U.S. JGOFS has been to produce and distribute reference materials for oceanic CO2 measurements.These materials are stable substances for which one or more properties are established sufficiently well to calibrate a chemical analyzer or to validate a measurement process.
Our laboratory at Scripps Institution of Oceanography (SIO), established in !989 with U.S. JGOFS support from the National Science Foundation (NSF), has prepared over 50 separate batches of reference material and has distributed more than 25,000 bottles of this material to scientists in 33 laboratories in the U.S. and 58 facilities in 24 other countries.The reference materials have been used both as a basis for collabora~ tive studies and as a means of quality control for at-sea measurements.Although most JGOFS field studies are over, we are still distributing more than 2,000 bottles per year and demand is again growing.
To prepare the reference material, we sterilize a batch of seawater, equilibrate it to a virtually constant partial pressure of CO2 and deliver it for bottling.For each batch, surface seawater collected on ships of opportunity and stored in our laboratory is pumped into a holding tank using filters and a sterilizing unit to reduce contamination.When the holding tank is nearly full, mercuric chloride is added as a biocide.The seawater is then recirculated for a few days to ensure complete mixing and enable some gas exchange with filtered air that is pumped through the head-space of the tank.Finally, aliquots of the seawater are pumped through an ultraviolet sterilizing unit and a O. 1 tsm filter and into clean 500-mL glass bottles.These are sealed with grease and labeled.
Random samples from each batch of reference material are analyzed over a period of 2-3 months for both total dissolved inorganic carbon (G) and total alkalinity (AT), and the results are used to certify the batch before it is distributed.From the start, we used a high-quality method for the determination of CT in which a weighed amount of seawater is acidified and the CO2 extracted under vacuum, purified and determined manometrically.These analyses are carried out in the laboratory of C.D. Keeling at SIO using equipment originally developed for the calibration of gases for atmospheric CO2 measurements.
By 1996, we had also developed an accurate method for the measurement of AT using a two-stage potentiometric, open-cell titration with coulometrically analyzed hydrochloric acid.Once this latter method was being employed routinely to certify new batches of reference material, we used it retrospectively to analyze archived samples from earlier batches.
The uncertainties of these analyses used for certification are + 1.5 pmol kg ~ in CT and + 2 mol pkg ~ in A., and our reference materials have been shown to be stable for more than 3 years.We are now working on providing pH values on future reference materials, as well as values for 6~3C.As part of U.S. JGOFS, the Department of Energy (DOE) supported measurements of !1 ocean CO2 system parameters on sections of the WOCE Hydrographic Programme one-time survey.The CO2 Survey Science Team adopted the use of our reference materials as soon as they were available in early 1991 and continued to use them on subsequent cruises.Measurements made on reference materials while at sea were used to ascertain data quality on these expeditions and are thought to have contributed substantially to the overall high quality of the resultant data set.
A further indication that the use of reference materials has improved oceanographic data quality can be seen by examining the degree of agreement between measurements for deep water masses obtained where two separate cruises intersect.For cruises where reference materials were available, measurements of CT in deep water now typically agree to within 2 pmol kgL This is in sharp contrast to the problems encountered over the years with earlier ocean carbon data sets, where adjustments of as much as 15-20 pmol kg ~ were fairly common.The high-quality data sets now available provide a resource for synthesis and modeling that makes it possible to put together a coherent global view of the oceanic carbon cycle.
the temperature and chemical composition of seawater.The amount of [CO2]aq is proportional to the partial pressure of CO2 exerted by seawater.The difference between the pCO2 in surface seawater and that in the overlying air represents the thermodynamic driving potential for the CO2 transfer across the sea surface.The pCO2 in surface seawater is known to vary geographically and seasonally over a range between about 150/,latin and 750 ~atm, or about 60% below and 100% above the current atmospheric pCO2 level of about 370 /satm.Since the variation of pCO2 in the surface ocean is much greater than the atmospheric pCO2 seasonal variability of about 20/Jatm in remote uncontaminated marine air, the direction and magnitude of the sea-air CO2 transfer flux are regulated primarily by changes in the oceanic pCO2.The average pCO= of the global ocean is about 7/~atm lower than the atmosphere, which is the primary driving force for uptake by the ocean (see Figure 6 in Karl et al., this issue).
The pCO2 in mixed-layer waters that exchange CO2 directly with the atmosphere is affected primarily by temperature, DIC levels and AT.While the water temperature is regulated by physical processes, including solar energy input, sea-air heat exchanges and mixedlayer thickness, the DIC and A T are primarily controlled by the biological processes of photosynthesis and respiration and by upwelling of subsurface waters rich in respired CO= and nutrients.In a parcel of seawater with constant chemical composition, pCO2 would increase by a factor of 4 when the water is warmed from polar temperatures of about -1.9 °C to equatorial temperatures of about 30 °C.On the other hand, the DIC in the surface ocean varies from an average value of 2150 ~rnol kg -1 in polar regions to 1850 ~rnol kg-' in the tropics as a result of biological processes.This change should reduce pCO2 by a factor of 4. On a global scale, therefore, the magnitude of the effect of biological drawdown on surface water pCO2 is similar in magnitude to the effect of temperature, but the two effects are often compensating.Accordingly; the distribution of pCO2 in surface waters in space and time, and therefore the oceanic uptake and release of CO=, is governed by a balance between the changes in seawater temperature, net biological utilization of CO2 and the upwelling flux of subsurface waters rich in CO2.

Database and Methods
Surface-water pCO2 has been determined with a high precision (+_2/Jatm) using underway equilibrator-CO2 analyzer systems over the global ocean since the International Geophysical Year of 1956-59.As a result of recent major oceanographic programs, including the global CO2 survey and other international field studies, the database for surface-water pCO2 observations has been improved to about 1 million measurements with several million accompanying measurements of SST, salinity and other necessary parameters such as barometric pressure and atmospheric CO2 concentrations.Based upon these observations, a global, monthly climatological distribution of surface-water pCO2 in the ocean was created for a reference year 1995, chosen because it was the median year of pCO2 observations in the database.The database and the computational method used for interpolation of the data in space and time will be briefly described below.
For the construction of climatological distribution maps, observations made in different years need to be corrected to a single reference year (1995), based on several assumptions explained below (see also Takahashi et al., in press).Surface waters in the subtropical gyres mix vertically at slow rates with subsurface waters because of strong stratification at the base of the mixed layer.As a result, they are in contact with the atmosphere and can exchange COs for a long time.Consequently, the pCO2 in these warm waters follows the increasing trend of atmospheric CO2 concentrations, as observed by Inoue et al. (1995) in the western North Pacific, by Feely et al. (1999) in the equatorial Pacific and by Bates (2001) near Bermuda in the western North Atlantic.Accordingly, the pCO2 measured in a given month and year is corrected to the same month in the reference year 1995 using the following atmospheric COs concentration data for the planetary boundary layer: the GLOBALVIEW-CO2 database (2000) for observations made after 1979 and the Mauna Loa data of Keeling and Whorf (2000) for observations before 1979 (reported in CDIAC NDP-001, revision 7).
In contrast to the waters of the subtropical gyres, surface waters in high-latitude regions are mixed convectively with deep waters during fall and winter, and their CO2 properties tend to remain unchanged from year to year.They reflect those of the deep waters, in which the effect of increased atmospheric COs over the time span of the observations is diluted to undetectable levels (Takahashi et al., in press).Thus no correction is necessary for the year of measurements.

Distribution Maps for Climatological Mean Sea-air pC02 Difference
Figure 3 shows the distribution of climatological mean sea-air pCO2 difference (ApCO2) during February (Figure 3a) and August (Figure 3b) for the reference year 1995.The yellow-red colors indicate oceanic areas where there is a net release of COs to the atmosphere, and the blue-purple colors indicate regions where there is a net uptake of CO2.The equatorial Pacific is a strong source of CO2 to the atmosphere throughout the year as a result of the upwelling and vertical mixing of deep waters in the central and eastern regions of the equatorial zone.The intensity of the oceanic release of CO2 decreases westward in spite of warmer temperatures to the west.High levels of COs are released in parts of the northwestern subarctic Pacific during the northern winter and the Arabian Sea in the Indian Ocean during August.Strong convective mixing that brings up deep waters rich in COs produces the net release of COs in the subarctic Pacific.The effect of increased DIC concentration surpasses the cooling effect on pCO2 in seawater during winter.The high pCO2 in the Arabian Sea water is a result of strong upwelling in response to the southwest monsoon.High pCO2 values in these areas are reduced by the intense primary production that follows the periods of upwelllng.
The temperate regions of the North Pacific and Atlantic oceans take up a moderate amount of COs (blue) during the northern winter (Figure 3a) and release a moderate amount (yellow-green) during the northern summer (Figure 3b).This pattern is the result primarily of seasonal temperature changes.Similar seasonal changes are observed in the southern temperate oceans.Intense regions of CO2 uptake (blue-purple) are seen in the high-latitude northern ocean in summer (Figure 3b) and in the high-latitude South Atlantic and Southern oceans near Antarctica in austral summer (Figure 3a).The uptake is linked to high biological utilization of CO2 in thin mixed layers.As the seasons progress, vertical mixing of deep waters eliminates the uptake of COs.
These observations point out that the Z~CO2 in high-latitude oceans is governed primarily by deepwater upwelling in winter and biological uptake in spring and summer, whereas in the temperate and subtropical oceans, the ApCO2 is governed primarily by water temperature.The seawater ApCO2 is highest during winter in subpolar and polar waters, whereas it is highest during summer in the temperate regions.Thus the seasonal variation of ApCO2 and therefore the shift between net uptake and release of CO2 in subpolar and polar regions is about 6 months out of phase with that in the temperate regions.
The ApCO2 maps are combined with the solubility (s) in seawater and the kinetic forcing function, the gas transfer velocity (k), to produce the flux: (1) The gas transfer velocity is controlled by near-surface turbulence in the liquid boundary layer.Laboratory studies in wind-wave tanks have shown that k is a strong but non-unique function of wind speed.The results from various wind-wave tank investigations and field studies indicate that factors such as fetch, wave direction, atmospheric boundary layer stability and bubble entrainment influence the rate of gas transfer.Also, surfactants can inhibit gas exchange through their damping effect on waves.Since effects other than wind speed have not been well quantified, the processes controlling gas transfer have been parameterized solely with wind speed, in large part because k is strongly dependent on wind, and global and regional wind-speed data are readily available.
Several of the frequently used relationships for the estimation of gas transfer velocity as a function of wind speed are shown in Figure 4 to illustrate their different dependencies.For the Liss and Merlivat (1986) relationship, the slope and intercept of the lower segment was determined from an analytical solution of transfer across a smooth boundary.For the intermediate wind regime, the middle segment was obtained from a field study in a small lake, and results from a wind-wave tank study were used for the high wind regime after applying some adjustments.This relationship is often considered the lower bound of gas transfer-wind speed relationships.
The quadratic relationship of Wanninkhof (1992) was constructed to follow the general shape of curves derived in wind-wave tanks but adjusted so that the global mean transfer velocity corresponds with the long-term global average gas transfer velocity determined from the invasion of bomb 14C into the ocean.Because the bomb 1~C is also used as a diagnostic or tun-  ing parameter in global ocean biogeochemical circulation models, this parameterization yields internally consistent results when used with these models, making it one of the more favored parameterizations.
Using the same longterm global 14C constraint but basing the general shape of the curve on recent CO2 flux observations over the North Atlantic determined using the covariance technique, Wanninkhof and McGillis (1999) proposed a significantly stronger (cubic) dependence with wind speed.This relationship shows a weaker dependence on wind for wind speeds less than 10 ms -1 and a significantly stronger dependence at higher wind speeds.However, the relationship is not well constrained at high wind speeds because of the large scatter in the scarce observations.Both the U 2 and U 3 relationships fit within the data envelope of the study, but the U 3 relationship provides a significantly better fit.Nightingale et al. (2000) determined a gas exchange-wind speed relationship based on the results of a series of experiments utilizing deliberately injected sulfur hexafluoride (SF6), 3He and non-volatile tracers performed in the last decade.
The global oceanic CO2 uptake using different wind speed/gas transfer velocity parameterizations differs by a factor of three (Table 1).The wide range of global CO2 fluxes for the different relationships illustrates the large range of results and assumptions that are used to produce these relationships.Aside from differences in global oceanic CO2 uptake, there are also significant regional differences.Figure 5 shows that the relationship of W&M-99 yields systematically lower evasion rates in the equatorial region and higher uptake rates at high latitudes compared with W-92, leading to significantly larger global CO2 uptake estimates.
In addition to the non-unique dependence of gas exchange on wind speed, which causes a large spread in global air-sea CO2 flux estimates, there are several other factors contributing to biases in the results.Global wind-speed data obtained from shipboard observations, satellites and data assimilation techniques show significant differences on regional and global scales.Because of the non-linearity of the relationships between gas exchange and wind speed, significant biases are introduced in methods of averaging ] I dYeS --, o £olP .1.

_, )
the product of gas transfer velocity and wind speed.
The common approach of averaging the ApCO~ and k separately over monthly periods, determining the flux from the product and ignoring the cross product leads to a bias that is about 0.2 to 0.8 Pg.C yr -1 lower in the global uptake estimate.This bias shows a regional variation that is dependent on the distribution and magnitude of winds.This issue has been partly rectified in some of the relationships in which a global wind-speed distribution is used to create separate relationships between gas transfer and wind speed for short-term (a day or less) and long-term (a month or more) periods.Since wind-speed distributions are regionally dependent and vary on time scales of hours, this approach is far from perfect.The groundwork of efforts laid over the past decade and recently improved technologies make the quantification of regional and global CO2 fluxes a more tractable problem now.Satellites equipped with scatterometers that are used to determine wind speed offer daily global coverage.Moreover, these instruments measure sea-surface roughness that is directly related to gas transfer.This remotely sensed information, along with regional statistics of wind-speed variability on time scales shorter than a day, offers the real possibility that more accurate gas transfer velocities will be obtained.Efforts are underway to increase the coverage of pCO2 through more frequent measurements and data assimilation techniques, again utilizing remote sensing of parameters such as sea-surface temperature and wind speed.Better quantification of the fluxes will lead to better boundary conditions for models and improved forecasts of atmospheric CO2 concentrations.

Distribution of the Climatological Mean Net . . Sea-air C02 Flux
To illustrate the sensitivity of the gas transfer velocity and thus the sea-air CO2 flux to wind speed, we have estimated the regional and global net sea-air CO2 fluxes using two different formulations for the CO2 gas transfer coefficient across the sea-air interface: the quadratic ~ o c-U 2 dependence of W-92 and the cubic U 3 dependence of W&M-99.In addition, we have demonstrated the effects of wind-speed fields on the computed sea-air CO2 flux using the National Center for Environmental Prediction (NCEP)-41 mean monthly wind speed and the NCEP-. .1995 mean monthly wind speed distributions over 4 ° x 40 60 5 ° pixel areas.
In Table 2 the fluxes computed using the W-92 and the NCEP/National Center for Atmospheric Research (NCAR) 41-year mean wind are listed in the first row for each grouping in column one (for latitudinal bands, oceanic regions and regional flux).The column "Errors in Flux" located at the extreme right of Table 2 lists the deviations from the mean flux that have been determined by adding or subtracting one standard deviation of the wind speed (about _+ 2 m sec -1 on the global average) from the mean monthly wind speed in each pixel area.These changes in wind speeds affect the regional and global flux values by about _ 25%.The fluxes computed using the single year mean wind speed data for 1995 are listed in the second line in each column one grouping in the table.
The global ocean uptake estimated using the W-92 and the NCEP 41-yr mean wind speeds is -2.2 + 0.4 Pg C yr -1.This is consistent with the ocean uptake flux of -2.0 _+ 0.6 Pg C yr 4 during the 1990s (Keeling et al., 1996;Battle et al., 2000) estimated from observed changes in the atmospheric CO~ and oxygen variations.
The wind speeds for 1995 are much lower than the 41-year mean in the northern hemisphere and higher over the Southern Ocean.Accordingly, the northern ocean uptake of CO2 is weaker than the climatological mean, and the Southern Ocean uptake is stronger.The global mean ocean uptake flux of -1.8 Pg C yr -~ using the NCEP-1995 winds is about 18% below the climatological mean of -2.2 Pg C yr 1, but it is within the +_25% error estimated from the standard deviation of the 41-yr mean wind speed data.
When the cubic wind speed dependence (W&M-99) is used, the CO2 fluxes in higher latitude areas with strong winds are increased by about 50%, as are the errors associated with wind speed variability.The global ocean uptake flux computed with the 41-year mean wind speed data and the NCEP-1995 wind data is -3.7 Pg C yr -1 and -3.0 Pg C yr -1 respectively, an increase of about 70% over the fluxes computed from the W-92 dependence.These flux values are significantly greater than the flux based on atmospheric CO2 and oxygen data (Keeling et al., 1996;Battle et al., 2000).However, the relative magnitudes of CO2 uptake by ocean basins (shown in % in the regional flux grouping in the last four rows of  This has been computed using the mean monthly distribution of sea-air pC02 difference, the climatological NCEP 41-year mean wind speed and the wind-speed dependence of the C02 gas transfer velocity of Wanninkhof (1992).The yellow-red colors indicate a region characterized by a net release of C02 to the atmosphere, and the blue-purple colors indicate a region with a net uptake of COJrom the atmosphere.This map yields an annual oceanic uptake flux for C02 of 2.2 + 0.4 Pg C yr -1.

Distribution of Anthropogenic C02 in the Oceans
To understand the role of the oceans as a sink for anthropogenic CO~, it is important to determine the distribution of carbon species in the ocean interior and the processes affecting the transport and storage of CO2 taken up from the atmosphere.Figure 7 shows the typical north-south distribution of DIC in the Atlantic, Indian, and Pacific oceans prior to the introduction of anthropogenic CO~.In general, DIC is about 10-15% higher in deep waters than at the surface.Concentrations are also generally lower in the Atlantic than the Indian ocean, with the highest concentrations found in the older deep waters of the North Pacific.The two basic mechanisms that control the distribution of carbon in the oceans are the solubility and biological pumps.
The solubility pump is driven by two interrelated factors.First, CO~ is more than twice as soluble in cold polar waters than in warm equatorial waters.As westem surface boundary currents transport water from the tropics to the poles, the waters are cooled and absorb more CO2 from the atmosphere.Second, the high-latitude zones are also regions where intermediate and bottom waters are formed.As these waters cool, they become denser and sink into the ocean interior, taking with them the CO2 accumulated at the surface.
The primary production of marine phytoplankton transforms CO~ and nutrients from seawater into organic material.Although most of the CO2 taken up by phytoplankton is recycled near the surface, a substantial fraction, perhaps 30%, sinks into the deeper waters before being converted back into CO2 by marine bacteria.Only about 0.1% reaches the seafloor to be buried in the sediments.The CO2 that is recycled at depth is slowly transported over long distances by the largescale thermohaline circulation.DIC slowly accumulates   Gruber (1998).
in the deep waters as they travel from the Atlantic to the Indian and Pacific oceans.Using a 3-D global carbon model, Sarmiento et al. (1995) estimated that the natural solubility pump is responsible for about 20% of the vertical gradient in DIC; the remaining 80% originates from the biological pump.
The approaches for estimating anthropogenic CO2 in the oceans have taken many turns over the past decade.Siegenthaler and Sarmiento (1993) summarized early approaches for estimating the anthropogenic sink in the oceans, including ocean models of various complexity, atmospheric measurements and transport models used together with pCOo-measurements and estimates based on changes in oceanic ~BC and oxygen mass balance.They noted the wide range of ocean uptake estimates (1.6 -2.3 Pg C yr 1) and concluded that the larger uptake estimates from the models were the most reliable.
The first approaches for using measurements to isolate anthropogenic C02 from the large, natural DIC signal were independently proposed by Brewer (1978) and Chen and Millero (1979).Both these approaches were based on the premise that the anthropogenic DIC concentration could be isolated from the measured DIC by subtracting the contributions of the biological pump and the physical processes, including the pre-industrial source water values and the solubility pump.Gruber et al. (1996) improved the earlier approaches by developing the AC* method.This method is based on the premise that the anthropogenic COo concentration (C ~') can be isolated from measured DIC values (C m) by subtracting the contribution of the biological pumps (ACb'°), the DIC the waters would have in equilibrium with a preindustrial atmospheric CO: concentration of 280 ppm (C~q:~°), and a term that corrects for the fact that surface waters are not always in equilibrium with the atmosphere (ACa' ~°~9: The three terms to the right of the first equal sign make up AC*, which can be explicitly calculated for each sample.The fact that AC* is a quasi-conservative tracer helps remedy some of the mixing concerns arising from the earlier techniques (Sabine and Feely, 2001).The AC a' ~°" term is evaluated over small isopycnal intervals using a water-mass age tracer such as CFCs.
We have evaluated anthropogenic COa for the Atlantic, Indian, and Pacific oceans using the AC* approach.Figure 8 shows representative sections of anthropogenic CO2 for each of the ocean basins.Surface values range from about 45 to 60 ~mol kg ~.The deepest penetrations are observed in areas of deep water formation, such as the North Atlantic, and intermediate water formation, such as 40-50°S.Integrated water column inventories of anthropogenic CO2 exceed 60 moles m ~ in the North Atlantic (Figure 9).Areas where older waters are upwelled, like the high-latitude waters around Antarctica and Equatorial Pacific waters, show relatively shallow penetration.Consequently, anthropogenic CO2 inventories are all less than 40 moles m ~ in these regions (Figure 9).Data-based estimates indicate that the oceans have taken up approximately 105 _+ 8 Pg C since the beginning of the industrial era.Current global carbon models generally agree with the total inventory estimates, but discrepancies still exist in the regional distribution of the anthropogenic inventories.Some of these discrepancies stem from deficiencies in the modeled circulation and water mass formation.There are also a number of assumptions in the data-based approaches regarding the use of constant stoichiometric ratios and time-invariant air-sea disequilibria that may be inadequate in some regions.These are all areas of current research.Anthropogenic estimates should continue to converge as both the models and the data-based approaches are improved with time.

Conclusions
As CO2 continues to increase in the atmosphere, it is important to continue the work begun with the Global Survey of CO: in the Ocean.Because CO2 is an acid gas, the uptake of anthropogenic CO2 consumes carbonate ions and lowers the pH of the ocean.The carbonate ion concentration of surface seawater in equilibrium with the atmosphere will decrease by about 30% and the hydrogen ion concentration will increase by about 70% with a doubling of atmospheric CO2 from pre-industrial levels (280 to 560 ppm).As the carbonate ion concentration decreases, the buffering capacity of the ocean and its ability to absorb more COa from the atmosphere is diminished.Over the long term (millennial time scales) the ocean has the potential to absorb as much as 85% of the anthropogenic CO2 that is released into the atmosphere.Because the lifetime of fossil fuel CO2 in the atmosphere ranges from decades to centuries, mankind's reliance on fossil fuel for heat and energy will continue to have a significant effect on the chemistry of the earth's atmosphere and oceans and therefore on our climate for many centuries to millennia.
Plans are being formulated in several countries, including the United States, to establish a set of repeat sections to document the increasing anthropogenic inventories in the oceans.Most of these sections will follow the lines occupied during the WOCE Hydrographic Programme on which JGOFS investigators made CO2 survey measurements.The current synthesis effort will provide an important baseline for assessment of future changes in the carbon system.The spatially extensive information from the repeat sections, together with the temporal records from the time-series stations and the spatial and temporal records available from automated surface pCO~ measurements on ships of opportunity, will greatly improve our understanding of the ocean carbon system and provide better constraints on potential changes in the future, l~   Gruber (1998).

Figure 2 .
Figure 2. The Global Survey of C02 in the Ocean: cruise tracks and stations occupied between 1991 and 1998.

Figure 3 .
Figure 3. Distribution of climatologicaI mean sea-air pC02 difference (ApCO2) for the reference year 1995 representing non-El Niho conditions in February (a) and August (b).These maps are based on about 940,000 measurements of surface water pCOJrom 1958 through 2000.The pink lines indicate the edges of ice fields.The yellow-red colors indicate regions with a net release of C02 into the atmosphere, and the blue-purple colors indicate regions with a net uptake of COJrom the atmosphere.The mean monthly atmospheric pC02 value in each pixel in 1995, (pCO2)air, is computed using (pCO2)air = (C02)air x (Pb -pH20).(C02)air is the monthly mean atmospheric C02 concentration (mole fraction of C02 in dry air)from the GLOBALVIEW database(2000); Pb is the climatological mean barometric pressure at sea level from the Atlas of Surface Marine Data(1994); and the water vapor pressure, pH20, is computed using the mixed layer water temperature and salinity from the World Ocean Database (1998) of NODC/NOAA.The sea-air pCO2 difference values in the reference year 1995 have been computed by subtracting the mean monthly atmospheric pC02 value from the mean monthly surface ocean water pC02 value in each pixeI.

Figure 6 .
Figure6.Distribution of the climatological mean annual sea-air C02 flux (moles C02 m -2 yr-g for the reference year 1995 representing non-El Niho conditions.This has been computed using the mean monthly distribution of sea-air pC02 difference, the climatological NCEP 41-year mean wind speed and the wind-speed dependence of the C02 gas transfer velocity ofWanninkhof (1992).The yellow-red colors indicate a region characterized by a net release of C02 to the atmosphere, and the blue-purple colors indicate a region with a net uptake of COJrom the atmosphere.This map yields an annual oceanic uptake flux for C02 of 2.2 + 0.4 Pg C yr -1.

Figure 7 .
Figure 7. Zonal mean pre-industrial distributions of dissolved inorganic carbon (in units of /~mol kg9 along north-south transects in the Atlantic, Indian and Pacific oceans.The Pacific and Indian Ocean data are from the Global C02 Survey (this study), and the Atlantic Ocean data are from Gruber (1998).

Figure 8 .
Figure 8. Zonal mean distributions of estimated anthropogenic C02 concentrations (in units of Mmol kg -1) along north-south transects in the Atlantic, Indian and Pacific oceans.The Pacific and Indian Ocean data are from the Global C02 Survey (this study), and the Atlantic Ocean data are from Gruber (1998).

For these calculations the monthly ApCO2 climatology of Takahashi et al. (in press) was used
accordin¢l to: F = k (Sc/660) ~/2 s ~pCO2, Sc is the Schmldt number, which is determined for each pixel from climatoloqical SST.The solubility s was determined from standard relationships with SST and salinity.Un'fess noted, the monthly mean NCEP wind speeds for 1995 were used.b:Using the NCEP 41-year average monthly wind speed product rather than that of 1995.¢ : Using the NCEP &hour wind product.In these cases, the instantaneous wind speed formulations of W-92 and W&M-99 were used.velocity,k,as a function of wind speed.The relationships were developed from wind-wave tank experiments, oceanic observations, global constraints and basic theory.The different forms of the relationships are summarized in Table1.U~o is wind speed at 10 m above the sea surface.