On The Global Oxygen Anomaly and Air-Sea Flux
Hernan Garcia and Ralph Keeling
We present a new climatology of monthly air-sea oxygen fluxes throughout the ice-free surface global ocean. The climatology is based on weighted linear least-squares regressions using heat flux monthly anomalies as a template for spatial and temporal interpolation of historical O2 data. The seasonal oceanic variations show that the equatorial belt (20°S to 20°N) is characterized by relatively small air-sea fluxes when compared to the middle to high latitudes (40°-70°). The largest and lowest seasonal fluxes occur during summer and winter in both hemispheres. By means of an atmospheric transport model, we show that the model simulated amplitude and phasing of the variations in atmospheric O2/N2 ratios due to seasonal air-sea exchanges is in good agreement with observations at base-line stations in the Pacific Ocean; better than previous climatologies. The contribution of each major oceanic basin to the atmospheric observations is described. The global ocean seasonal net oxygen outgassing (SNO) is estimated to be about 0.87 Pmol O2 (1 Pmol = 1015 mol). The extra-tropical regions (³30° latitude) account for most (>80%) of the SNO in both hemispheres. The seasonal net thermal (SNOT) and biological (SNOB) outgasing components of the flux are examined in relation to latitudinal bands, basin-wide, and hemispheric contributions. The Southern Hemisphere's SNOB (~0.26 Pmol) and SNOT (~0.29 Pmol) values are larger than the Northern Hemisphere's SNOB (~0.15 Pmol) and SNOT (~0.16 Pmol) values. We estimate a seasonal extra-tropical carbon production of 3.4-3.5 Pg C (1 Pg = 1015 g) based on hemispheric-averaged SNOB values poleward of 30° latitude and an O2/C ratio of 1.45, lower than previously estimated from air-sea O2 climatologies.
The oceanic O2 flux contribution to the atmospheric O2/N2 ratio is primarily through seasonal-scale variability caused by a range of biochemical and physical processes across the air-sea interface. Biochemical processes include sources and sinks of O2 due to marine production, respiration, and remineralization of organic matter. Physical processes include sources and sinks caused by water mass renewal or ventilation, air-sea gas exchange, solubility changes driven by seawater warming and cooling, near-surface turbulence and mixing whether induced by wind, waves, or small-scale breaking (i.e., bubble injection and spray), and aeolian inputs of labile substances. The seasonal distribution of surface O2 anomalies and air-sea fluxes are useful to study the interplay between biochemical and physical processes that affect the O2 concentration in seawater and its ultimate effect on the atmosphere. Two important aspects of studying the oceanic annual cycle of O2 are to delimit sources and sinks of photosynthetic carbon production and air-sea exchange of CO2 [i.e., Keeling et al., 1993].
Najjar and Keeling  presented a global climatology of monthly air-sea O2 fluxes based on rescaling the O2 anomaly values of Najjar and Keeling  using the sea-surface temperature (SST) climatology of Shea et al.  and adding a correction for O2 anomalies caused by air bubble injection. Keeling et al.  described the SST rescaled climatology. Briefly, Keeling et al.  averaged the SST climatology of Shea et al.  into 12° latitudinal bands for each major ocean basin and calculated the fundamental harmonic amplitude of the averaged SST data. They then computed averaged SST values where concurrent O2 anomaly values were available and computed the fundamental harmonic amplitude for this reduced SST data set in each 12° latitudinal band. They then calculated a scaling factor between the fundamental harmonic amplitudes of the averaged SST climatology and the reduced SST data. They multiplied the O2 anomaly values of the Najjar and Keeling  climatology by the scaling factor after subtracting the annual mean O2 anomaly at each grid point. The SST-rescaled O2 anomaly climatology was obtained after adding back the annual mean to each monthly value in the original O2 anomaly value climatology. The SST-rescaled climatology improved the difference between the model simulated and observed variations in the atmosphere when compared to the climatology without rescaling. But the rescaled climatology still underestimated the amplitude of the atmospheric variations by about 20% at baseline stations located poleward of 30° latitude in the Northern and Southern Hemisphere.
Here we revisit the task of computing a representative global air-sea O2 flux monthly climatology. The new climatology reduces excessive smoothing in O2 data sparse regions. We outline an alternative method to that of Najjar and Keeling [1997, 2000], and Keeling et al.  for estimating a global monthly climatology of sea-surface O2 anomaly and air-sea flux using selected historical O2 data at observed depth levels. Our approach is based on binning the O2 data by coarse spatial increments in latitude and, instead of binning the data by time of collection, we bin the O2 data by fine increments in air-sea seasonal heat flux anomalies independent of the time of year. Ocean heat flux is an objective, conservative variable. We use the DQsea data as a spatial and temporal template for interpolation of O2 anomaly and flux values throughout the ice-free oceanic domain. Seasonal heat flux anomalies were chosen for interpolating the O2 data because, as shown by Keeling et al. , we expect a relationship between seasonal heat flux and air-sea O2 flux regardless of whether the O2 exchange is mediated by biological or physical processes. We compare our results to the Keeling et al.  and the Najjar and Keeling [1997; 2000] data without air bubble injection correction. We find that the new flux climatology yields improved simulations of atmospheric O2/N2 variations.
Since the solubility of O2 in seawater is primarily temperature dependent, oceanic regions of net heating degas O2 to the atmosphere and oceanic regions of net cooling absorb atmospheric O2. In this sense, air-sea O2 flux is out of phase with temperature but in phase with heat flux. A similar relationship is expected for biologically mediated O2 air-sea exchange whenever there is a strong correlation between primary production, oxygen production, and stratification. Increased vertical mixing during fall and winter driven by surface cooling brings nutrient-rich, carbon-rich, but relatively O2-poor water (less recently ventilated waters) to the mixed layer. Surface heat loss is thus associated with oceanic O2 uptake. During spring and summer, vertical stratification reduces vertical mixing between surface (nutrient-poor, O2-rich) and deeper (nutrient-rich, O2-poor) waters. Biologically O2 production then increases due to increases in primary production that result from a combination of increased solar irradiance and available nutrients and O2 is released to the atmosphere. Surface heating is thus associated with biologically mediated O2 outgassing. As the season progresses and vertical stratification occurs, nutrient concentrations and primary production generally decrease. The seasonal trends of surface heating and cooling and biological O2 production suggest a correlation between heat flux anomalies and air-sea oxygen flux.
The primary data used in this study are discrete water samples with measured O2 concentrations, in-situ temperature (T, °C), and salinity from selected near sea-surface historical oceanographic data. The hydrographic data are from the World Ocean Circulation Experiment (WOCE), NODC archived data as of 1998, and the data compilations of Reid [1989; 1994], Garcia , and Garcia et al. . The NODC data corresponds to the world ocean database 1998. Typically, the bottle data consists of 12 or 24 depth discrete observations between the surface and the bottom. The O2 data in volumetric units were converted to mass units (΅mol kg-1) using the molar volume of O2 as a real gas and a density (r, kg m-3) corresponding to T and S [Millero and Poisson, 1981]. The historical bottle data (918489 observations) provide good coverage in the Northern Hemisphere and some parts of the Southern Hemisphere. We recognize that there is variability (i.e., diurnal, inter-annual or longer time-scale) inherent in the historical data but this potential temporal bias is not treated here. In this study, we take O2 measured data in the upper 15-m of the water column to represent the distribution of O2 in the surface layer.
The historical hydrographic data were collected over several years on different cruises. The precision of historical T, S, and pressure data are estimated to be ±0.01°C, ±0.01, ±5 dbar respectively [Saunders, 1986; Garcia, 1996]. We carried out a data quality control on the historical O2 on several steps. We first rejected all measured O2 data collected prior to 1960 because we believe that about this time more precise O2 data began to be routinely collected. The hydrographic data were then sorted into 10° latitude by 10° longitude areas. We rejected spurious observations by means of variance and range checks on the historical data contained within these areas using O2 against T and S plots. Then for each area, we calculated the spatially averaged value for T, S, and O2 and the standard error using all the data at each interpolated level of density. About 10% of the historical O2 data were excluded from this data quality step. The precision of the historical O2 data retained in this study is about ±3 ΅mol kg-1.
The initial data processing step is the calculation of oxygen anomalies (D[O2], ΅mol kg-1) at every location where there were historical O2, T, and S data. D[O2] was evaluated as the difference between the measured O2 concentration ([O2]obs) and the O2 solubility ([O2]*) in seawater, D[O2]= [O2] - [O2]* + δskin, where dskin is the skin temperature correction. The [O2]* values were computed as a function of T and S from the bottle data and sea level atmospheric pressure (Pa, at) using the O2 solubility equations of Garcia and Gordon , the atmospheric pressure equation of Benson and Krause , and the monthly atmospheric pressure climatology of Oberhuber . The precision of the O2 solubility values is about 0.1-0.2% [Garcia and Gordon, 1992]. We assume complete O2 equilibration with the atmosphere. The average atmospheric pressure corrections relative to mean sea-level atmospheric pressure (1013 mbar) are typically about ±4 ΅mol kg‑1. For dskin, we used the equation of Hasse  with a depth for the bulk in-situ temperature data from the hydrographic data of 2.5 m and the European Center for Medium-range Weather Forecasting (ECMWF) heat flux monthly climatology [Gibson et al., 1997]. The dskin corrections are typically about ±1 ΅mol kg-1. The dskin and Pa corrections combined are ~±4 ΅mol kg-1. The combined corrections are small when compared to the magnitude of the seasonal variations in D[O2]. We do not treat bubble injection corrections. There are no adequate constraints on the net effect of bubble injection on the global ocean gas flux and the net seasonal effect is likely small [Schudlich and Emerson, 1996; Keeling et al., 1998; Emerson et al. 1993].
We evaluate air-sea O2 fluxes (O2, mol m-2 month-1) using O2= ρkO2D[O2], where kO2 is the gas transfer velocity for O2 (kO2, m s-1). To calculate kO2, we use the formulation of Wanninkhof  for long-term winds [kO2=0.39u2(ScO2/660)-1/2, where u (m s-1) is wind speed and ScO2 is Schmidt number for O2]. For calculating the Schmidt number we use the relation ScO2=1638-81.83T+1.483T2-0.008004T3 [Keeling et al., 1998] using SST values from Shea et al. . For wind speeds we use ECMWF monthly values [Gibson et al., 1997]. We use the kO2 formulation of Wanninkhof  because it appears to be applicable to large-scale studies [Keeling et al. 1998; Najjar and Keeling, 2000]. We computed O2 values for all locations with retained historical O2 data.
The objective of the weighted linear least-squares regression model is to estimate O2 as a function of monthly heat flux anomalies in 10° latitudinal bands covering the ice-free global ocean surface (80°S to 80°N). For evaluating the statistical weights for the O2 data in each 10° latitudinal band, both the O2 and the monthly heat flux anomaly values were processed through a three-step data processing scheme. First, we computed monthly heat flux anomalies ([DQ]sea, Watt m-2) at each point of the ECMWF grid by subtracting the annual mean heat flux value from each monthly heat flux value. Second, a monthly [DQ]sea value corresponding to the geographical location and month of collection of every O2 data value was calculated as follows. The [DQ]sea value corresponding to each of the O2 data value was found by first locating the monthly heat flux anomaly value in the ECMWF climatology equal to the month of each O2 observation, and then we used spatial interpolation between adjacent [DQ]sea grid points to assign a [DQ]sea value to each O2 data value according to its geographical position. Third, we divided the O2 data set into 10° latitudinal bands between 80°N and 80°S without overlapping and binned the O2 values within monthly [DQ]sea increments of 50-watts m-2 in each 10° latitudinal band. The choice of 50-watts m-2 increments in [DQ]sea is a qualitative compromise between having a relatively large number of O2 data points and low resolution in [DQ]sea and a high resolution in [DQ]sea with small number of O2 data values in any particular latitudinal band. We then computed the mean and standard errors of the O2 values regardless of month of collection and geographic location within each DQsea increment. We use the mid-point value of the 50-watts m-2 DQsea increments to bin the O2 values. Values of O2 that differed by more than four standard deviations from their mean values in each DQsea increment were rejected from further analysis. This data quality control step eliminated about 5% of the O2 data. We then evaluated the mean and standard errors for the retained O2 values at every 50-watts m-2 DQsea increment within each 10° latitudinal band. Thus, all bins with data were independently evaluated according to the statistical information in each bin. The data processing described above yields the mean and standard errors of O2 values binned in each DQsea increment for all the data geographically located in each 10° latitudinal band.
We performed weighted linear least-squares regressions between DQsea as the independent variable and O2 as the dependent variable for all the data in 10° latitudinal bands of the general form O2=a0+a1[DQ]sea, where a0 and a1 are the intercept and slope regression coefficients. We assume that the variance in the regressions is solely due to the O2 data. Thus, in calculating the regression coefficients for each 10° latitudinal band, we used weights based on the standard error of the O2 data binned in each DQsea increment. The purpose of the weights in the calculation of the regression coefficients is to provide more importance to O2 values whose variance factors are smaller and less importance to O2 values whose variance factors are larger at every DQsea increment. Thus, the regression coefficients are sensitive to the statistical information in the O2 values. Appendix A and Appendix B show examples of the linear least-regressions between oxygen anomaly and heat flux and oxygen flux and heat flux for different latitudinal bands in the Northern and Southern Hemispheres. Appendix C lists the O2=a0+a1[DQ]sea regression coefficients.
Figure 1 shows the magnitude of the slope regression coefficients (a1) for each 10° latitudinal band. The coefficients show large north-south gradients within the northern and southern hemispheres and are generally asymmetrical about the equator. The a1 coefficients increase poleward reaching maximum values between 40° and 70°. The Southern Hemisphere shows higher a1 values with a somewhat narrower latitudinal peak centered near 45°S than at comparable latitudes in the Northern Hemisphere. Poleward of 70° latitude, the a1 values decrease because of increased seasonal sea-ice coverage and decreases in the wind gas exchange and wind speeds. Between 10°N and 10°S, the coefficients are near zero or even slightly negative. The equatorial waters show large D[O2] values that can be attributed to upwelling of waters highly under-saturated with respect to O2. But the surface waters exhibit relatively small seasonal changes in O2 and DQsea when compared to surface waters at middle to high latitudes. For statistical comparison, the coefficient of linear determination (r2) is greater than 0.6 poleward of 20° reaching maximum values (≥ 0.8) poleward of about 40° latitude. The r2 values are greater than 0.4 between 20°S and 20°N.
We also carried out separate regression analysis between O2 and DQsea for data located in the Atlantic, Pacific, and Indian Oceans in each 10° latitudinal band (Figure 1). The main reason for evaluating separate regression analysis in each major ocean basin is because differing sampling density between basins could bias the magnitude of the global regression coefficients towards values appropriate for the basins with the largest number of observations. We adopt subjective definitions for the geographical boundaries of the major ocean basins. The latitudinal patterns of the a1 coefficients in each basin show similar features as the global a1 coefficients, although some differences are notable (Figure 1). The North Atlantic shows the largest a1 values except north of 70°N and south of 50°S. The Indian Ocean slope coefficients show that largest positive values centered near 50°S.
What is the effect of kO2 on the computed air-sea O2 fluxes as a function of latitude? To examine the effect of kO2 alone, we carried out a weighted regression analysis between D[O2] and DQsea of the form D[O2]=b0+b1[DQ]sea, where b0 and b1 are the intercept and slope regression coefficients. We followed the same procedure described earlier for obtaining the a0 and a1 coefficients in each 10° latitudinal band. Any latitudinal differences between b1 (D[O2]/[DQ]sea) and a1 (O2/[DQ]sea) reflect the effect of kO2 differences because this is the only free parameter. Figure 2 shows the latitudinal variation of the b1 slope coefficients. The b1 coefficients increase poleward nearly monotonically (Figure 2), while the a1 coefficients increase poleward reaching maximum values between 40° and 70° and decrease poleward of 70° latitude as described earlier (Figure 1). Thus, the decrease in the a1 values observed at high and low latitudes reflect latitudinal decreases in wind speed values. Appendix D lists the D[O2]=b0+b1[DQ]sea regression coefficients.
The latitudinal distribution of the regression coefficients is subject to uncertainty resulting from sampling errors, data variability, the effect of differing sampling density between latitudinal bands and between basins, and covariance effects between O2, kO2, and DQsea. It is difficult to quantify these effects. We believe that the a1 regression coefficients are representative of the mean large-scale air-sea O2 flux variability on seasonal time scales across each 10° latitudinal band, and demonstrate this later. First, the regression coefficients obtained for the Atlantic, Pacific, and Indian Oceans closely follow the latitudinal trends of the global regression coefficients (Figures 1, 2). Second, evaluating linear regressions over 5° latitudinal bands did not yield significantly different regression coefficients that those obtained using 10° bands. Third, the latitudinal distribution of the a1 coefficients indicates real north-south trends. Fourth, as shown below, the correlation between O2 and DQsea as a function of latitude captures most of the variation in O2/N2 as seen in the atmosphere. The sensitivity of the regressions coefficients to covariance between O2, kO2, and DQsea is difficult to quantify. We believe that the most important sources of error in the calculation of O2 values are the variability in the O2 data.
To obtain a global distribution of monthly O2 values, we use the regression coefficients (a0 and a1) as linear scale functions of DQsea. We calculated O2 values at every grid point of the DQsea climatology both spatially and temporally within the 10° latitudinal bands. We used a linear interpolation scheme to obtain O2 values between the mid-point of the 10° degree latitudinal bands. We calculated seasonal anomalies in air-sea fluxes (O2)sea at every grid point according to (O2)sea=a1DQsea. The annual mean of (O2)sea computed this way is zero because DQsea also has zero annual mean. The mean annual O2 does not contribute measurably to seasonal variation in the atmospheric O2/N2 ratio [Keeling et al., 1998]. Only the seasonal component of the air-sea O2 flux is needed in our atmospheric simulation as shown below. We set the (O2)sea values to zero poleward of 80° latitude and in ice-covered regions using the SST data of Shea et al. .
Maps of the distribution of the air-sea O2 fluxes monthly anomalies show strong meridional gradients reflecting the seasonal variation of the distribution of winds (exchange velocity), seasonal heat fluxes, and biological production. For brevity, we show maps for December and June, the months with the largest seasonal fluxes (Figure 3). The seasonal pattern of O2 flux anomalies in both hemispheres is characterized by sea-to-air fluxes during summer and air-to-sea fluxes during winter. This hemispheric pattern is consistent with warming, vertical stratification, and high rates of primary production during summer, and cooling, enhanced vertical mixing, and low rates of primary production during winter throughout most of the open surface ocean. In the Northern Hemisphere fluxes generally increase polewards to about 60°N and decrease north of this latitude. In contrast, in the Southern Hemisphere the fluxes are largest over a relatively narrow latitudinal belt centered near 50°S.
We use our estimated (O2)sea values to simulate the seasonal variability in the air O2/N2 ratio by means of the TM2 atmospheric transport model [Heimann, 1995]. Briefly, the TM2 model is a 3-D advection model with a spatial resolution of about 7.5°x7.5° and 9 vertical levels. We initialize the atmospheric transport model using the (O2)sea values and the ECMWF wind climatology for the year 1986. Steady state results are typically reached after the fourth year after initialization of the TM2 model. The model results are then compared to inter-annually de-trended O2/N2 observations collected at time-series stations located mainly in the Pacific (Table 1). The oceanic component of the de-trended data is isolated correcting the data for land photosynthesis and respiration using CO2 data following Keeling et al. . To predict changes in the oceanic O2/N2 ratio, it is necessary also to account for the small seasonal N2 variation in the atmosphere due to air-sea N2 exchange (N2). The N2 values were computed as the product of ECMWF seasonal heat flux anomaly times the temperature derivative of the N2 solubility ([N2]*). The ([N2]* values were computed using the solubility equation of Weiss  as a function of SST [Shea et al., 1992]. We subtracted the annual mean value from the monthly N2 flux to get the seasonal component (N2)sea. The seasonal cycles for the simulated and observed O2/N2 cycles are fitted using a four-harmonic seasonal cycle. We adopt the per meg unit to compare the oceanic and atmospheric O2/N2 variations [Keeling et al., 1993].
We do not separately simulate the contribution of the annual-mean O2 and N2 fluxes to the atmospheric O2/N2 changes. Our interest here is the seasonal cycles in O2/N2 for which the annual-mean fluxes appear to be insignificant as shown by Keeling et al. . Figure 4a shows a comparison between the model simulated and the measured O2/N2 cycles. The agreement between the simulated and observed O2/N2 variations is very good, particularly at the middle to high-latitude time-series stations where the largest seasonal variations are found. The timing of the air cycle in both hemispheres generally follows the timing of the cycle of oceanic ingassing in winter and outgassing in summer. The simulated O2/N2 variations based on the global regression coefficients lead the atmospheric observations by slightly less than a month. Except for La Jolla, the simulated values of the peak-to-peak amplitude are within 10% of the observed amplitude. At La Jolla, the simulated amplitude is too low by about a factor 0f 0.6. The Kumukahi station has the largest phase difference, about a month. Even neglecting the biological component, some phase lag is expected based on mixed layer equilibration time for O2 of a few weeks that our model simulations neglect. Also, a phase lag in the biological component is expected due to the lagged response of photosynthesis to seasonal stratification and nutrient availability.
The contribution of the major ocean basins to the observed O2/N2 variation is shown in Figure 4b. The amplitude and phasing of the seasonal oceanic cycles at the baseline stations is dominated by the O2/N2 contributions from the Pacific and Atlantic basins. The contribution of the Indian Ocean to the Pacific O2/N2 ratio plays a smaller but not insignificant role. The results suggest that the large-scale contribution of each basin to the O2/N2 variations is most sensitive to the north-south distribution of O2 surface fluxes, and to a lesser extent, to zonal distribution because the atmosphere smoothes out east-west trends.
Simulations based on the O2 anomaly climatology of Najjar and Keeling  are shown for comparison purposes in Figure 4c. Results are shown using the Najjar and Keeling  climatology both with and without SST-rescaling. The model simulations based on the Najjar and Keeling  climatology underestimate the observed O2/N2 cycles poleward of 31° latitude on average by about 15% in the Northern Hemisphere and by 23% in the Southern Hemisphere. A particularly large discrepancy is seen at Samoa (SMO) station (14°S), a result not explained by Keeling et al. . Model simulations based on our climatology are generally in better agreement with observations, particularly at Samoa, suggesting that the Najjar and Keeling  climatology may be inaccurate in the tropical South Pacific. On the other hand, the lack of atmospheric observations at additional baseline locations in the Tropical Pacific makes it difficult to draw firm conclusions, and the discrepancy could be caused by other errors such as uncertain atmospheric transports, in which case our model simulations are in agreement fortuitously.
To assess the significance of the hemispheric (O2)sea values, we compared the effect of the (O2)sea latitudinal gradients in each hemisphere on the observed O2/N2 ratio in the atmosphere by means of the TM2 model. We calculated (O2)sea values in the southern hemisphere using the a1 regression coefficients for the northern hemisphere. We calculated also (O2)sea values in the northern hemisphere using the a1 regression coefficients for the southern hemisphere. We then initialized the TM2 model with the new set of (O2)sea values. The results of the simulations indicate that the amplitude of the atmospheric cycles at the baseline stations are too low or too high by as much as 40% depending on the base-line station. This suggests that the magnitude and latitudinal gradient of the a1 coefficients shown in Figure 1 are significant and reflect real features in amplitude and phasing of the cycles in each hemisphere. The results also indicate that there is little inter-hemispheric mixing on seasonal time scales with respect to O2.
We also assessed the sensitivity of the simulated O2/N2 variations to the choice of global and basin-wide regression coefficients. We initialized the TM2 model using (O2)sea that we obtained from the a1 regression coefficients estimated independently in the Atlantic, Pacific, and Indian basins. Using these basin a1 coefficients produced changes in the amplitude of the oceanic cycles at the baseline stations of about ±10% depending on location when compared to the fluxes obtained using the global a1 coefficients. This means that the magnitude and latitudinal gradient of the global a1 coefficients capture the essential large-scale features of the air-sea O2 flux. We cannot discard the possibility that finer spatial and temporal resolution of the a0 and a1 values might improve the simulated O2/N2 cycles. In summary, the results indicate that the regression approach provide a representative prediction of the seasonal air-sea O2 flux and the oceanic O2/N2 variations when compared to the atmospheric observations.
One approach to achieve a better quantitative agreement between the simulated and observed atmospheric O2/N2 variations is to scale the global oceanic O2 flux field by a constant, dimensionless correction or calibration factor while fixing the seasonal N2 flux. This factor can be viewed as a multiplicative correction to the air-sea O2 gas exchange coefficient. Keeling et al  optimized the gas-exchange velocity for O2 based on the Najjar and Keeling  climatology using the algorithm of Heimann and Keeling . The algorithm yields scaling factors that minimize the sum of squared residuals between the observed and simulated atmospheric O2/N2 cycles in the least-squares sense. Here we used the same algorithm to estimate scaling factors for the O2 exchange velocities based on the present climatology. We refer to this scaling factor as atmospheric calibration.
Table 3 lists atmospheric calibrations based on the present O2 flux climatology as well as correction factors computed previously by Keeling et al.  using the Najjar and Keeling  climatology. Except for the La Jolla station, all of the atmospheric calibrations are very close to 1. We find that the atmospheric calibration for simultaneous fitting of all the stations is 1.01±0.05. The atmospheric calibrations for the high-latitude stations range between 1.09±0.09 in the Northern Hemisphere (ALT, CBA, NWR) and 0.97±0.08 in the Southern Hemisphere (CGO, SPO). In contrast, Keeling et al.  obtained larger atmospheric calibrations for the same grouping of stations: 1.14±0.05 for all the stations while for the high-latitude stations the scaling factors ranged between 1.15±0.05 (ALT, CBA, NWR) and 1.23±0.06 (CGO, SPO) as shown in Table 3.
The results indicate that the regression approach provides a representative simulated atmospheric O2/N2 variation without the need for atmospheric calibrations factors. The combined use of the surface O2 anomaly data of this work and the kO2 formulation of Wanninkhof  yield simulated O2/N2 variations in good agreement with the atmospheric O2/N2 observations. The results substantiate the use of the Wanninkhof  relation for computing air-sea gas exchange rates at large spatial and monthly time scales.
We represent O2 as the sum of thermal (fT) and biological (fB) fluxes. Following Keeling et al., , we compute the fT component using,
Where Q is heat flux, cp is heat capacity of seawater (3992 Joule kg-1 °C-1), and T is temperature (°C). We evaluate fT at every grid point of the O2 climatology and calculate the seasonal thermal component (fT)sea by subtracting the annual mean from each monthly value. We evaluate the seasonal biological component (fB)sea as the difference bettwen the total seasonal flux and the thermal component using (B)sea=(O2)sea- (T)sea. Because fT assumes complete O2 equilibration with the air, its seasonal amplitude might be overestimated and (fB)sea might be underestimated. Thus, the biological component of the flux represents a conservative lower limit when there is no O2 equilibration of surface waters with the atmosphere. As described earlier, we set all fluxes equal to zero in ice-covered waters.
One index of the seasonal air-sea O2 flux exchange is seasonal net outgassing (SNO) defined by Keeling and Shertz  as the spatially and temporally integrated oxygen flux over the annual periods when the spatially integrated flux is positive (e.g., sea to air). Najjar and Keeling  adopted a slightly different definition of SNO than Keeling and Shertz  in which the annual-mean O2 flux is subtracted before integration. The difference in the definition is insignificant at the hemispheric scale, and the latter definition is perhaps more appropriate on smaller spatial scales as a measure of the contribution to seasonal variations in atmospheric O2/N2. Here we follow the Najjar and Keeling  definition of SNO and compute also both the thermal (SNOT), and biological (SNOB) seasonal net outgassing. To compute SNO, we first subtracted the annual mean flux from each monthly flux value and then we integrated over the months when the fluxes are positive. We carried this calculation on hemispheric and latitudinal bands for the global ocean and major ocean basins. We computed SNOT and SNOB in a similar manner as SNO using the fT and fB monthly climatology. For simplicity, SNO values are reported in units of 1014 mol O2.
Table 2 lists SNOT and SNOB values without atmospheric calibrations for selected latitudinal ranges in the global ocean and ocean basins. We computed the global SNOB (4.2) and SNOT (4.5) values by summing the thermal and biological hemispheric results (Table 2). The SNOB:SNOT ratio provides insight into the contribution of thermal and biological sources. The SNOB:SNOT ratio in the Southern Hemisphere is 0.9, while in the Northern Hemisphere the ratio is 1.0. This suggests that the thermal and biological hemispheric averaged contributions to SNO are roughly equal. The extra-tropical (>30°) SNOB:SNOT ratios are 1.2 in the Northern Hemisphere and 1.1 in the Southern Hemisphere indicating a slightly greater biological contribution to SNO. Computing SNOB and SNOT values between 30° and 60° latitude yielded SNOB:SNOT ratios of 1.3 and 1.2 in the Northern and Southern Hemispheres suggesting a greater biological than thermal contribution to SNO in the temperate regions. In general, different latitudinal bands have distinct biological and thermal contributions to SNO. The global pattern of SNO is similar in each major basin. The Atlantic, Pacific and Indian Basins have SNOB/SNOT ratios in each hemisphere ranging between 0.9 and 1.1. The Southern Hemisphere's SNOB and SNOT values are greater than in the Northern Hemisphere by about ~1.7 and ~1.8. The difference can be explained in part due to the greater mean ice-free ocean area of the Southern Hemisphere compared to the mean ice-free ocean area of the Northern Hemisphere (about 1.4 larger). Our calculated thermal and biological SNO values for the extra-tropical regions (≥ 30° latitude) account for about 77% of the SNOT and 92% of the SNOB hemispheric results. Our estimate essentially omits equatorial and other oceanic areas where the seasonal O2 flux variation is not well resolved.
We summed the Northern and Southern Hemispheric SNO values without atmospheric calibrations to estimate a global SNO of about 8.6 (Table 4). The Northern and Southern Hemispheres contribute 36% and 64% of the global SNO. On basin scales, the Pacific is the largest contributor (50%) to the global SNO, followed by the Atlantic (30%) and the Indian Basins (20%). A basin-wide SNO value is difficult to establish because the value depends on the choice of geographical boundaries of each basin. To gain insight into how representative our simulated oxygen fluxes are without atmospheric calibrations when compared to the atmospheric observations, we compared the calculated SNO against values estimated using the calibration coefficients listed in Table 3. We multiplied the fluxes poleward of 30° latitude by 1.09 in the Northern Hemisphere and by 0.97 in the Southern Hemisphere. The choice of calibration factors is to some extent arbitrary because the optimized fluxes are sensitive to the atmospheric observations in the Pacific stations. Nevertheless, we used these calibration factors because most of the oceanic contribution to the atmospheric O2/N2 variations at the Pacific base-line stations appears to be caused by seasonal air-sea variations poleward of 30° latitude [Keeling et al., 1998]. We show that the difference between the uncalibrated (8.6) and calibrated (8.7) sum of hemispheric SNO values is small, about 1 % (Table 4). The small difference is not surprising because the high-latitude calibration values and the averaged calibration factors for all the base-line stations combined is about 1 (Table 3).
For comparison purposes, we have also calculated SNO values from the Najjar and Keeling  climatology. Using the Najjar and Keeling  climatology without atmospheric calibrations, we obtain hemispheric SNO values of 3.22 and 5.34 and we obtain extra-tropical (>30° latitude) SNO values of 2.72 and 3.84 in the Northern (N.H.) and Southern (S.H.) Hemispheres respectively. Although hemispheric SNO estimates using our climatology (without atmospheric calibration) are quite similar (3% smaller in the N.H. and 3% larger in the S.H.) the distribution is different. Whereas the Najjar and Keeling  climatology yields 85% (N.H.) and 72% (S.H.) of the hemispheric SNO in the extra-tropics, our climatology yields 82% (N.H.) and 86% (S.H.) of hemispheric SNO in the extra-tropics. Particularly in the Southern Hemisphere, our climatology has larger seasonal fluxes at higher latitudes and lower fluxes at lower latitudes than the Najjar and Keeling  climatology. The present O2 flux climatology is probably more realistic on the basis of the comparison with the atmospheric observations at Samoa (14°S), Cape Grin (41°S), and the South Pole (90°S) (Figure 4c).
Following Keeling and Shertz , Najjar and Keeling  used the biological SNO values computed from their O2 flux climatology to estimate extra-tropical (> 20° latitude) biological new production in the mixed layer during the shoaling period. First, they calculated SNOB for each 10° latitudinal band poleward of 20°N and 20°S and excluded from their calculations the North Indian Ocean. They then added up the SNOB values for individual 10° latitudinal bands and reported a spring-summer mixed layer new production of 4.5 Pg C (without atmospheric calibrations) and 5.6 Pg C (with atmospheric calibrations) assuming an O2:C ratio of 1.45 (1 Pg= 1015 g). If we add up individual 10° latitude bands poleward of 20° (excluding the North Indian Ocean) following Najjar and Keeling , we calculated a global SNOB value equivalent to 3.71 Pg C (without atmospheric calibrations) and 3.75 Pg C (using atmospheric calibrations) based on the present O2 flux climatology and an O2:C ratio of 1.45. Our carbon production estimates are smaller than those of Najjar and Keeling . The discrepancies between the estimates include differences between the O2 anomaly climatologies, calculation of thermal fluxes, and the choice of geographic boundaries of the oceanic basins and non-ice covered oceanic regions.
The calculation of SNOB is sensitive to the area of integration as well as the phasing of the flux fields. For example, the sum of hemispheric-averaged SNOB values (without atmospheric calibrations) is equivalent to 3.45 Pg C for the present climatology and 4.25 Pg C for the Najjar and Keeling  climatology. If we use the atmospheric calibrated O2 fluxes to calculate hemispheric SNOB values, then the values increase to 3.51 Pg C for the present climatology and to 5.18 Pg C for the Najjar and Keeling  climatology.
Following Najjar and Keeling , we can compare our estimates of extra-tropical carbon production based on SNOB with satellite estimates of carbon primary production. Najjar and Keeling  computed an extra-tropical seasonal primary production of 11.8 Pg C in the mixed layer based on the primary production data of Antoine et al.  based on CZCS chlorophyll observations, and computed an f-ratio (the ratio of new to total primary production) of 0.38-0.46. Using their 11.8 Pg C estimate of seasonal primary production and our estimate of extra-tropical carbon production (3.71-3.75 Pg C), we derive an f-ratio of approximately 0.3. Some caution must be exercised in interpreting SNOB values strictly in relation to carbon new production. The assumption that the seasonal O2 outgassing approximates net community O2 production during the seasonal period of shoaling mixed layer, has been shown to yield reasonable results locally (Jenkins and Goldman, 1985; Emerson, 1987), although it is only approximately correct. Oxygen outgassing could be smaller than carbon production to the extent that some of the seasonal O2 production is stored in the mixed layer, or it could be either larger or smaller than carbon production to the extent that the air-sea O2 flux is also influenced by exchanges of O2 with deeper waters.
Here we present an improved global monthly climatology of surface O2 anomalies and air-sea fluxes based on a weighted linear least-squares approach. The regressions use seasonal heat flux anomalies for spatial and temporal interpolation of selected historical O2 data. The method gives a robust relation for estimating the large-scale mean seasonal distribution of O2 anomalies and fluxes. We concentrate primarily on the distribution of the global air-sea seasonal O2 flux.
Model simulated seasonal oceanic air-sea O2/N2 contributions to the atmosphere compare well with seasonal variations in atmospheric O2/N2 ratios at base-line stations in the Pacific Ocean. Optimization of the gas-exchange velocity for O2 by means of scaling factors shows that little adjustment is necessary between the simulated and observed O2/N2 variations. This substantiates that the present O2 anomaly climatology together with the wind exchange velocity formulation of Wanninkhof  for steady wind conditions provide a robust measure of the air-sea O2 flux. Our study suggests that the component of the air-sea O2 flux that correlates with heat flux dominates the large-scale air-sea O2 exchange on seasonal time scales. The results are useful as constraints on global seasonally resolved models.
The global seasonal net oxygen outgassing is estimated to be about 0.9 Pmol O2 (1 Pmol= 1015 mol). Evaluation of the seasonal air-sea O2 flux attributable to thermal and biological sources indicates that the two components make roughly equal contributions to the seasonal air-sea exchange on hemispheric scales. The extra-tropical biological seasonal net outgassing is equivalent to a seasonal carbon production of about 3.8 Pg C based on an O2:C ratio of 1.45.