Primary productivity in the central Arctic Ocean has recently been reported as being much higher than earlier thought. If a significant fraction of this primary production were exported from the immediate surface region, present estimates of the carbon budget for the Arctic Ocean would have to be reassessed. Using the deficit of phosphate in the central Arctic Ocean, we show that the export production is very low, on an average less than 0.5 gC m−2 yr−1. This is at least an order of magnitude lower than the total production as measured or estimated from oxygen data, thus indicating extensive recycling of nutrients in the upper waters of the central Arctic Ocean and very little export production.
 Ice covers nearly all of the central Arctic Ocean throughout the year, while the surrounding shelf seas experience ice-free conditions during the summer. Water enters the central Arctic Ocean from the surrounding seas, enters the Atlantic water through the Fram Strait and the Barents Sea, and enters the Pacific water through the Bering Strait. Freshwater is added by river runoff and by melting of sea ice and “removed” as sea ice is formed. Much of this water enters via large continental shelves, where significant transformations occur as a result of biological and physical processes. A considerable biological export production drives a flux of carbon dioxide from the atmosphere to the ocean in the Barents Sea [e.g., Slagstad and Wassmann, 1996; Olsson et al., 1999; Fransson et al., 2001] as well as in the Bering-Chukchi Sea region [e.g., Springer and McRoy, 1993; Hansell et al., 1993].
 In the ice-covered central Arctic Ocean the magnitude of the primary production has recently been reported to be considerably higher than reported earlier [Wheeler et al., 1996; English, 1961]. If a significant fraction of this primary production were exported from the immediate surface region, present estimates of the carbon budget for the Arctic Ocean [Anderson et al., 1998a] would have to be reassessed. From a carbon budget perspective, it is essential to know the export production, i.e., the amount of organic carbon leaving the surface water annually. We use the phosphate deficit, the difference between measured phosphate concentrations in the central Arctic Ocean and the initial values in source waters, to deduce the export production and thus the fate of the dissolved inorganic carbon consumed by primary production in the photic zone. Furthermore, we include an estimate of the amount of phosphate added by vertical mixing from deeper regions to the surface photic layer. Of the chemical constituents affected in parallel to carbon by biological processes (oxygen, nitrate, and phosphate) we use phosphate, as other processes might alter nitrate and oxygen concentrations without significantly altering phosphate concentrations. For example, oxygen concentrations would be affected in the surface water by gas exchange, and nitrate concentrations are affected by denitrification in a low oxygen environment. In these instances, their classical relation to carbon would not hold. The nitrate concentration might also be influenced by nitrogen fixation in surface water where phosphate is available but nitrate is not.
2. Data and Methods
 The data used for this investigation were collected during expeditions that span much of the central Arctic Ocean (Figure 1). The data used for the evaluation of the Pacific water fraction were collected during the 1993 ARCRAD cruise [Cota et al., 1996], the 1993 Larsen cruise [Macdonald et al., 1995], the Arctic Ocean Section 1994 cruise [Swift et al., 1997], and the JOIS 1997 cruise. The data for evaluating the Atlantic water fraction were collected during the ACSYS 1996 cruise [Augstein, 1997]. We compare the measured phosphate concentration with the initial phosphate concentration, i.e., the phosphate concentration in the same water when it entered the central Arctic Ocean from the shelf seas. Source waters come from the Pacific Ocean and the Atlantic Ocean, with freshwater contributions added from sea ice meltwater and river runoff. The approach used by Jones et al.  to compute the source water fractions from the N-P relationships and salinity is applied in this work.
3. Source Waters and Initial Phosphate Concentrations
3.1. Evaluation of Source Water Fractions
 As a result of denitrification processes, relative concentrations of phosphate with respect to nitrate are significantly higher in waters of Pacific origin flowing through Bering Strait than in waters of Atlantic origin [Jones et al., 1998]. Furthermore, the sediments of shallow shelf regions of the Bering, Chukchi, and East Siberian Seas are also areas where denitrification occurs, resulting in elevated phosphate concentrations relative to nitrate in waters close to the bottom [e.g., Codispoti et al., 1991].
 Nitrate and phosphate concentrations in river runoff are quite variable, but the average concentrations in Arctic rivers, ∼3 μM of nitrate and ∼0.4 μM of phosphate [e.g., Meybeck, 1982; Gordeev et al., 1996], are close to the N-P relationship of Atlantic origin water. Nitrate and phosphate concentrations in sea ice meltwater are also variable, but reported data are in the same range as runoff [e.g., Macdonald et al., 1987]. Hence the freshwater N-P signal is close to that of Atlantic water, and the freshwater fraction is distinguished from the Atlantic fraction using salinity.
 In order to find the most appropriate equation to separate the Pacific origin fraction from the Atlantic and freshwater fraction in the central Arctic Ocean, we plot the nitrate concentration versus that of phosphate for different cruises in the regions of in-flowing Pacific and Atlantic waters (Figure 2). It should be stressed that the slopes of the equations are close to the classical N:P ratio of organic material and hence primary production will not significantly shift the data from the lines. It should be noted, however, that the N:P ratios in surface waters in the Chukchi Sea region are not an appropriate designator of Pacific source water as there is a non-zero intercept in a plot of N versus P. A linear N versus P relationship rather than N:P ratios must be used for this purpose. A similar situation holds for Atlantic source water within the Arctic Ocean, though the intercept is much smaller. The two equations best representing the nitrate-phosphate signature of the waters entering the deep central Arctic Ocean from the Atlantic and Pacific Oceans are
 The Pacific source water relationship used is the one that best fits the data from the Canada Basin JOIS 97 cruise. Several of the data from the Chukchi Sea region are found to the right of this source water line (Figure 2). We attribute this to local denitrification. Since the Canada Basin JOIS 97 section is somewhat removed from the Chukchi Sea, changes in nutrient concentrations attributed local processes in the Chukchi Sea region should be smoothed out. Hence the JOIS 97 data would be a better representation of the Pacific source water entering the central Arctic Ocean at the Chukchi Sea shelf break. All data to the right of the Pacific source line are assigned to 100% Pacific water. This deviation of the data from this line does not have any impact on the given initial phosphate concentration as seen in next section.
 The freshwater and Atlantic water fractions are computed from the equations
The fPW, fAW, and fFW are the Pacific, Atlantic, and freshwater fractions; SPW, SAW, and SFW are the salinity of the Pacific, Atlantic, and freshwater sources; and Smeas is the measured salinity of a particular sample. SFW is close to zero, with only sea ice meltwater making a minor contribution, and we thus set it to zero. With this assumption, the freshwater fraction is
 The source waters in our evaluation are those entering the photic zone of the central Arctic Ocean Basin from the shelf seas, i.e., those found at the shelf break. Thus the reference salinities do not represent pure Atlantic or Pacific source waters, but a mixture of these with fresher waters over the shelves. As we are interested in the consumption of phosphate in the surface water of the central Arctic Ocean, we have chosen to take the mean salinity for waters in the top 50 m at the Chukchi shelf break (AOS-94, stations 7–9) and the St. Anna Trough (ACSYS-96, stations 6–14), i.e., SPW = 31.44 and SAW = 33.85, respectively.
3.2. Initial Phosphate Concentrations
 The initial phosphate concentrations can be estimated using the approach described above once the fraction of the different source waters has been determined. These are the concentrations of the waters as they enter the central Arctic Ocean. We address data sets separately. For the 1994 Arctic Ocean Section and the 1996 ACSYS cruises, the data are divided into three groups: one where Pacific source water dominates (fPW > 0.9), one where Atlantic source water dominates (fAW > 0.9 and S > 33.85), and a third, largest group that is a mixture of all source waters. The phosphate source water concentrations in each of the first two groups are determined by the phosphate-salinity relationships as observed at the shelf breaks. We find a quite tight fit of the data with the following equations for the Chukchi shelf break and St. Anna Trough, respectively,
 Both equations have a negative intercept that results from export production, i.e., the consumption of phosphate, over the shelves. Equation (6) is fitted in the salinity range 30 to 31.4 corresponding to a phosphate range of ∼0.8 to ∼1.1 μM, while equation (7) is fitted in the salinity range 33.5 to 34.5, corresponding to a phosphate range of ∼0.3 to ∼0.7 μM.
 For the third group, which has contributions from all source waters, we chose concentrations of phosphate that are the mean in the top 50 m at the Chukchi shelf break and the St. Anna Trough, 1.2 and 0.43 μM, respectively. These data were collected in the summertime, but the phosphate concentrations in the surface layer were similar in the Eurasian Basin during the ACSYS-94 cruise [e.g., Anderson and Kaltin, 2001], where the residence time is many years and no evidence of recent primary production was seen. Consequently, these concentrations should be representative of the water supplying the central Arctic Ocean surface layer even though the concentrations are significantly lower than in the Pacific and Atlantic source waters as a result of productivity in the Chukchi and Barents Seas. The phosphate concentrations in the freshwater component in the central Arctic Ocean are a combination of contributions from river runoff and sea ice meltwater. The reported concentrations vary between 0.1 and 0.5 μM [Meybeck, 1982; Macdonald et al., 1989; Cauwet and Sidorov, 1996; Gordeev et al., 1996; Lara et al., 1998] with an average of 0.4 μM for both freshwater sources. An error in this value does not have a major impact on the computation of the initial phosphate concentration as the freshwater fraction is small compared to the oceanic fractions. The initial concentration of a sample having contributions from all source waters (group three) is computed by
 For the Canada Basin data (JOIS 97) a slightly different approach is taken to evaluate the initial phosphate concentration. Equation (7) is used for data with salinity between 30 and 32.75 (f PW > 0.9), while a mixing line having the end-members S = 30; [PO4] = 0.786 μM and S = 0; [PO4] = 0.4 μM is used for data with S < 30. The latter data are found in the surface waters, which have a significant addition of freshwater relative to the water found at the Chukchi shelf break. The data with S > 32.75 are found in deeper parts of the water column, and represent samples that include water of Atlantic origin. A mixing line with the end-members S = 32.75, [PO4] = 1.29 μM and S = 34.85, [PO4] = 0.78 μM represents their initial phosphate concentrations. As only a few of these data are found at depths shallower than 100 m, they will not have a significant impact on the computation of the biological export production.
4. Phosphate Deficit and Carbon Export Production
 The phosphate deficit, PO4measured − PO4initial (Figure 3), represents what has been consumed by primary production minus what is released through mineralization. Depth profiles are divided into groups representing the different ways that the initial concentrations were computed. The surface waters (shallower than about 50 m) have a deficit, while the deeper waters mostly have a surplus. Only in the Eurasian Basin, where Atlantic source water dominates, is it close to zero all the way to 200 m. The surplus is a result of the choice of Pacific source water concentration, which was selected to represent the water leaving the shelf break into the photic zone of the central Arctic Ocean. Hence the surplus below ∼50 m is a combination of lack of phosphate consumption over the shelves and/or mineralization both in the Chukchi Sea and in the deep basin. However, the surface water of the Canadian Basin is strongly stratified and it is not likely that much export production occurs below 50 m in this basin. The deficit in the surface water of the Eurasian Basin is significantly less than in the Canadian Basin (Figure 3a). The deficit in the surface waters of the Canada Basin (JOIS 97) section (Figure 3b) is slightly lower than that in the rest of the Canadian Basin (AOS 94) section.
 The phosphate deficit can be converted to carbon units using the classical C/P ratio of 106/1 [Redfield et al., 1963]. This will reflect a combination of export production (during previous seasons) and new production (during the season of the investigation). The geographical distribution of this production is plotted in sections of the upper 200 m (Figure 4). The largest production occurs in parts of the Canadian Basin AOS 94 section, (Figure 4a), while in the Canada Basin JOIS 97 section (Figure 3b) the production is lower and more homogeneous. The more variable estimate in the Canadian Basin AOS 94 section could result from patchy productivity [e.g., Gosselin et al., 1997]. It may also be caused by an uncertainty in our computations arising from the difficulty in finding a general formulation for computing initial phosphate concentrations. In the Eurasian Basin, where this computation is most straightforward, the estimate is less variable and does not show any high values.
 The data from the different areas were binned and average profiles plotted to get some quantitative estimate of the export production in the central Arctic Ocean (Figure 5). Integrating the deficits in the top 60 m of Figure 5 gives −2.9, −8.4, and −5.8 gC m−2 for the Eurasian, Canadian (AOS 94), and Canada (JOIS 97) Basins, respectively. These numbers cannot be directly compared, as they are also integrated over the residence time of the surface water at the different locations. The residence time of the Eurasian Basin surface water is around 5 years [Schlosser et al., 1994], but that of the Canada Basin surface water is significantly longer. Taking the volume of the surface water of the Canadian Basin (170 × 1012 m3) and dividing by the water flowing into the surface water from the Pacific (0.36 Sv according to Anderson et al. [1998a]), we get a turnover time of about 15 years. The circulation pattern of the surface waters indicated by nutrient relationships [Jones et al., 1998] suggests that the surface water over the central Canada Basin has a longer residence time than that over the Makarov Basin. The difference in the integrated carbon deficit over the Canada Basin compared to the rest of the Canadian Basin would be even greater if longer Canada Basin residence times were applied. A plausible explanation for this lower integrated carbon deficit could be that the Canada Basin region has the heavier ice cover, resulting in less light input to the surface water and thus lower primary production.
 Including the turnover times of the different surface waters gives the export production over the central Arctic Ocean in the order of only 0.5 gC m−2 yr−1. However, the above approach does not include nutrients added from below by vertical mixing. Wallace et al.  estimated the vertical mixing coefficient (Kz) to be 2 × 10−6 m2 s−1. Using this value, we calculate the supply of phosphate (FPO4) to the surface region by vertical mixing from below according to
 The average phosphate gradient below the photic zone is ∼0.3 × 10−5 mol m−4 resulting in a flux of
 This flux would add another ∼50% to the new production computed above from the deficit of phosphate, bringing the total export production to ∼0.7 gC m−2 yr−1. Varying Kz will have a direct effect on the computed new production. Rudels et al.  estimated a diffusion coefficient of 1.1 × 10−6 m2 s−1 in the Eurasian Basin, while D'Asaro and Morison  report values ranging from 1.5 × 10−6 m2 s−1 to 7 × 10−6 m2 s−1 over the Nansen Abyssal Plain. In regions of “rough” topography, their more typical values were 5 × 10−5 m2 s−1. Using the range reported over the abyssal plain (between 1 and 7 × 10−6 m2 s−1) the addition of phosphate from below results in an increased new production estimate of between 0.12 and 0.84 g C m−2 yr−1.
 Adding this range to the new production computed from the deficit of phosphate, 0.5 gC m−2 yr−1, results in a total new production of between ∼0.6 and ∼1.3 gC m−2 yr−1. This is more than an order of magnitude lower than the estimates of total annual primary production as measured by Gosselin et al.  or as estimated by Pomeroy  based on oxygen data. Our estimate is based on dividing the carbon deficit by the turnover time. Only if most their signal (∼99%) were a result of regenerated production during the single year of investigation would the results be of the same order.
5. Concluding Remarks
 We note that Moran et al.  estimated the export of particular organic matter (POC) along AOS 94 from 234Th/238U activity ratios to be 0.3–7 mmol C m−2 d−1 (mean of 3 mmol C m−2 d−1). If the growing season were 100 days, then annual export according to this technique would be 3.6 gC m−2 yr−1. There are several uncertainties in the POC/234Th ratio technique. One is that the 234Th/238U disequilibrium may not be in steady state. Another is that the POC/234Th ratio may be 2–10 times higher in the suspended particulate size class dominating the measurements [Moran et al., 1997] than in the large, rapidly sinking, particles in the Arctic Ocean, thus resulting in a corresponding overestimate of the POC export flux.
 An export production of ∼0.7 gC m−2 yr−1 corresponds to a sedimentation rate of 2.5 × 1012 gC yr−1 out of the photic zone of the Canadian Basin if production is constant throughout the entire 3500 × 109 m2 area of the basin. This corresponds to about 10% of anthropogenic carbon sequestered in the Arctic Ocean [Anderson et al., 1998b]. This amount of export production consumes only about 10% of the phosphate added to the surface water by the input through Bering Strait (1 μM in 0.36 Sv). Hence there is a potential to increase the carbon sink by an order of magnitude if environmental changes allow for greater production, giving a total potential sink of around 15 × 1012 g C yr−1 if all phosphate were to be consumed. The total potential sink for the whole of the central Arctic Ocean would also include the input to the Eurasian Basin from the Atlantic. The phosphate concentration in the Atlantic source water is lower and its contribution to the surface layer is mostly less than that from other sources. Altogether, there is a potential for a total increased biological carbon sink of around 20 × 1012 gC yr−1 if the ice cover of the central Arctic Ocean decreases or disappears as a result of global warming. This corresponds to less than 1% of the present total oceanic sequestering of anthropogenic carbon dioxide, and would thus only have a marginal negative feedback to the greenhouse effect. Furthermore, it is a very small fraction of the total global oceanic biological carbon sink, around 10,000 × 1012 gC yr−1 [Fasham et al., 2001].
 We wish to acknowledge support from the Swedish Natural Science Research Council (L. G. A.), the Canadian Panel on Energy Research and Development (E. P. J.), and the U.S. National Science Foundation and U.S. Office of Naval Research (J. H. S.).