The distribution of freshwater within the Arctic Ocean and its export from it are intimately involved in climate and climate change processes both within and outside the Arctic Ocean. River runoff in the Arctic Ocean constitutes a major part of the Arctic Ocean freshwater budget. Within the Arctic Ocean, variability in the distribution of river runoff will be reflected in the location of the cold halocline that isolates the sea ice from the warm Atlantic Layer. Outside the Arctic Ocean, such variability will impact on the salinity of North Atlantic waters (Great Salinity Anomaly) and on deep convection areas of the North Atlantic Ocean, and thereby potentially on global thermohaline circulation. Rivers entering the Arctic Ocean have high levels of total alkalinity that contribute significantly to the total alkalinity of the surface Polar Mixed Layer. We exploit total alkalinity data to trace river runoff in the surface Polar Mixed Layer and to observe variability in the river runoff distribution in the Eurasian Basin over the period 1987–2001. The river runoff front changed from a position over the Gakkel Ridge in 1987 and 1991 to over the Lomonosov Ridge in 1996, and returned to a midpoint between the two ridges in 2001. Wind field changes as characterized by the Arctic Oscillation index are considered to be a major factor in determining ice and surface water flow. We note a correlation with 4–6 years delay between changes in river runoff distribution and the Arctic Oscillation index. We show that the delay can be inferred from a geostrophic flow calculation.
 There is more evaporation than precipitation in the Atlantic Ocean. Much of this evaporated freshwater falls as rain into the Pacific Ocean and onto the Eurasian continent. In both instances, the freshwater is returned to the Atlantic Ocean via the Arctic Ocean. The relatively fresh Pacific origin water (S ∼ 32.5) flows into the Arctic Ocean through Bering Strait. River runoff flows into the Arctic Ocean mostly from the Eurasian continent with about 10% coming from North America [Aagaard and Carmack, 1989]. With the freshwater content of the inflowing Pacific water calculated relative to the inflowing Atlantic water (S ∼ 35), each source contributes a similar amount, roughly 0.1 Sv (1 Sv = 106 m3 s−1), to the total return flow of freshwater. Sea ice meltwater within the Arctic Ocean and sea ice exported from the Arctic Ocean (about 0.16 Sv [Vinje and Finnekåsa, 1986]) are added to the budget of freshwater leaving the Arctic Ocean.
 The amount and timing of the export of freshwater from the Arctic Ocean will depend on its circulation and changes in circulation within the upper layers of the Arctic Ocean. The circulation in the upper ∼600 m can be summarized in general terms. Atlantic water entering through eastern Fram Strait and the Barents Sea flows eastward following the Arctic continental margin. The flow divides in the vicinity of the Lomonosov Ridge, with a part turning north to follow the ridge and exit in a return flow through western Fram Strait. A second part crosses the Lomonosov Ridge and flows into the Canadian Basin, where it follows the continental margin under the Polar Mixed Layer (PML). Pacific water entering through Bering Strait splits in the Chukchi Sea region with one branch flowing along the coast of North America and the other flowing north eventually to partially mix with the Atlantic water [Jones et al., 1998; Shimada et al., 2001].
 The Arctic Ocean freshwater export into the North Atlantic Ocean can potentially affect global thermohaline circulation. River runoff from Eurasia becomes incorporated mostly into the PML in a region north of the Laptev Sea and seas farther east, with some additional runoff from the Mackenzie River being added into the Beaufort Sea. River runoff (as well as Pacific water and sea ice meltwater) exits the Arctic Ocean through the Canadian Arctic Archipelago and through western Fram Strait. The Arctic Ocean freshwater export to the Nordic Seas (Greenland, Iceland, and Norwegian seas) through Fram Strait [e.g., Meredith et al., 2001] follows the East Greenland Current and exits the Nordic Seas through Denmark Strait, while the freshwater export to the Labrador Sea occurs primarily through the Canadian Arctic Archipelago. This freshwater export and its timing can change the deep water formation rate in these regions by creating a stratified surface layer that could inhibit the production of water dense enough for deep convection to occur [Aagaard and Carmack, 1989]. Changes in this freshwater export affect other downstream climate-related properties of the North Atlantic Ocean through freshening such as the “Great Salinity Anomaly” [Dickson et al., 1988; Belkin et al., 1998].
 River runoff in the Arctic Ocean has been traced using oxygen-18 (O-18) isotopes [Ekwurzel et al., 2001; Schlosser et al., 2002], barium concentrations [Guay et al., 2001] and total alkalinity [Anderson et al., 1994]. All of these tracers have limitations, the main one being that the concentrations of each can be different in different rivers. However, they are all useful to follow the general flow pattern of river runoff. The barium excess was used to provide direct evidence that the fluvial discharge to the Kara and Laptev seas enters the interior of the Arctic Ocean in the vicinity of the Lomonosov and Mendeleyev Ridges [Guay et al., 2001]. In 1994 the off-shelf flow of river water was concentrated to the Mendeleyev Ridge area, as traced by O-18 [Ekwurzel et al., 2001]. Also using O-18, Schlosser et al.  showed a decrease in meteoric water (mainly river runoff) of between 38% and 50% from 1991 to 1996 over the Amundsen Basin and the Lomonosov Ridge. In this paper we use total alkalinity concentrations from five different cruises between 1987 to 2001 over the Eurasian Basin to define river runoff fronts and changes in their location.
2. Data and Methods
 The total alkalinity (AT) in Arctic rivers is fairly high, typically above 1000 μmol kg−1 [Olsson and Anderson, 1997], as a result of combined decay of organic matter (mainly carbohydrates) and dissolution of metal (mainly calcium) carbonates in the drainage basin (reaction (R1)). The relative amount of runoff can be computed in a seawater sample of known salinity and AT if the average concentration of AT in river runoff and in sea ice meltwater are known. In well-oxygenated water the relative concentration of AT is affected only in a limited way by biological processes as long as no calcium carbonate-producing plankton are present. In Arctic Ocean near-surface waters, the oxygen concentration is high, mostly >80% saturated [Koltermann and Lüthje, 1989], and the dominating plankton species, diatoms and the dinoflagellate Phaeocystis pouchetii [Vernet et al., 1998], do not form carbonate shells. During photosynthesis, proteins are formed (reaction (R2)), where nitrate and hydrogen ions are consumed in the same molar ratio. Hence the maximum change in AT by this process is equal to the maximum consumption of nitrate in the surface water, ∼5 μmol kg−1 in the central Arctic Ocean, which is close to the precision of the measurements.
 Within the Eurasian Basin the near-surface water is predominantly Atlantic water mixed with sea ice meltwater and river runoff. A schematic plot of total alkalinity versus salinity shows how the salinity and total alkalinity of the Atlantic water will change as it mixes with sea ice meltwater and river runoff (Figure 1). When sea ice is formed and brine is rejected, the sea ice maintains a bulk salinity (which in multiyear ice, ranges from 0 to 5 but is typically near 4 [Weeks and Ackley, 1986]), with AT corresponding to this salinity along the mixing line S1. (We disregard the possibility of fractionation of salts contributing to AT that could occur during the freezing process as this was not observed under normal conditions [Anderson and Jones, 1985].) When sea ice melts and no river runoff is present, the meltwater mixes with the near-surface water and changes S and AT along the same mixing line S1. When no sea ice meltwater is present, river runoff mixes with the near-surface water and changes S and AT along the mixing line R, whose intercept on the AT axis corresponds to the AT of the river runoff.
 In the near-surface water of the Arctic Ocean, all of the above processes take place, but not necessarily at all locations. In some regions with essentially no runoff, the salinity-AT properties fall along mixing line S1. In some regions the surface water is mixed with runoff and thus has a lower salinity. When sea ice is formed from such a surface water, for example, salinity ∼28, the salinity AT property of the bulk ice follows line S3; for salinity ∼31 it follows line S2. The properties in multiyear sea ice formed from these different salinities will fall within the box indicated by SI. When this sea ice melts and mixes with underlying water, the mixed water properties follow the mixing line having a seawater end-member lying on mixing line R, with the corresponding salinity.
 In general, knowing the salinity and AT of the three sources, river runoff, sea ice meltwater, and Atlantic water, one can compute their relative fractions in a sample with known salinity and AT. The mean AT concentration in the rivers added to surface water of the Eurasian Basin can be evaluated by fitting a line in a salinity-AT plot of all samples shallower than 400 m (at depth where the Atlantic layer mixes with river water) in the eastern Eurasian Basin. Doing so for data collected north of the Laptev Sea in 1996 gives a line with best fit of AT = 25.53(±0.64) × S + 1411.6(±21.7) (R = 0.9484, N = 181) (Figure 2), and a concentration of 1412 μmol kg−1 should thus be a good mean of AT in river runoff entering the Arctic Ocean from the Laptev Sea region. The salinity of the seawater end-member of this line is 34.85, corresponding to an AT concentration of 2296 μmol kg−1. To further investigate how much the sea-ice meltwater contributes to the water used for this study, we assume a salinity of 4 for multiyear sea ice, giving an AT of 263 μmol kg−1 if it is formed from seawater with S = 34.85 (and without any fractionation). Hence the relative fractions of seawater (fsw), river runoff (frro), and sea ice meltwater (fsim) can be computed from the following three equations:
 A minor error is caused in this computation by the assumption that the sea ice is formed from and is melted by water of salinity 34.85. A maximum 2% overestimate of the amount of river runoff occurs with the present seawater end-member if instead we used S = 34.2, corresponding to that of the lower halocline [Rudels et al., 1996]. The data used in the evaluation were collected in the Eurasian Basin of the Arctic Ocean during five cruises over the period 1987 to 2001 (Table 1). Sections along which stations are located are shown in Figure 3.
Table 1. Cruises From Which Data are Used in This Work
 The total alkalinity during the AOS 94, ACSYS 96, and AO-01 cruises was determined by potentiometric titration using HCl [Haraldsson et al., 1997]. The precision determined by analyzing duplicate samples was ±2 μmol kg−1. On the ARK IV/3 and IAOE 91 cruises a potentiometric titration method in a closed cell [Johansson and Wedborg, 1982] having precision of ±4 μmol kg−1 was used.
 Certified reference materials (CRM), supplied by A. Dickson, Scripps Institution of Oceanography (USA), were used to determine the accuracy. However, these were only available for the two latest cruises (ACSYS 96 and AO-01). To be able to compare data from all cruises, we corrected the data from those not having CRM by comparing the mean of the total alkalinity in the Atlantic layer (identified by the temperature maximum in the depth profile in each station) with the corresponding AO-01 data. All corrections were less than 20 μmol kg−1. There are no total alkalinity data from a large part of the ACSYS 96 section across the Eurasian Basin. For that region, total alkalinity was calculated from total dissolved inorganic carbon (CT) and pH using CO2SYS, version 01.03 [Lewis and Wallace, 1998]. The carbonate dissociation constants (K1 and K2) used were those of Roy et al.  and KSO4 determined by Dickson .
 The ratio between the means was used to correct the total alkalinity data for the ARK IV/3, IAOE 91, and AOS 94 cruises. The mean total alkalinity in the Atlantic layer for the ACSYS 96 and AO-01 cruises were also compared and showed good agreement.
3. Results and Discussion
 In plots of total alkalinity versus salinity for each of the five cruises (Figure 4), two features are obvious. First, the scatter is larger in the total alkalinity data of the 1987 and 1991 cruises, where an analytical method with less precision was used. Second, few of the samples collected in 1987 and 1996 included much of a contribution from sea ice meltwater.
 The data from 2001 have the least complex distribution (Figure 4, bottom plot), with points clearly following the mixing lines of Atlantic water mixing with river runoff and sea ice meltwater, respectively. Only at the lowest salinities (<∼31) do the data deviate toward lower AT relative to the river runoff mixing line. The absence of data between the two mixing lines at salinities above ∼31 shows that in this salinity range Atlantic water either mixes with river runoff or with sea ice meltwater. This illustrates that we have two regimes, one with and one without contributions from river runoff, and that little horizontal mixing occur between these regimes.
 The deviation toward lower AT relative to the river runoff mixing line at salinities below ∼31 in 2001 illustrates that these samples represent a mixing of sea ice meltwater with low AT and a surface water with a high fraction of river runoff. In 1991, 1994, and 2001 a number of data are aligned along the mixing line of Atlantic water and sea ice meltwater; that is, there is a negligible presence of runoff. These samples were located in or close to the Nansen Basin, as illustrated in section plots (Figure 5).
 A number of data points in most cruises are found above the mixing line between Atlantic water and river runoff, largely within the salinity range 33–34. This is in the interval where a nutrient maximum also is found, a signal of water flowing off the Chukchi shelf. This nutrient maximum has been proposed to be a result of decay of organic matter on the Chukchi shelf, to a large degree at the sediment surface [Jones and Anderson, 1986]. The excess of nitrate is low compared to phosphate and carbonate because of denitrification in low-oxygen environments. Denitrification adds to the total alkalinity in the order of 130 times the increase in phosphate. The resulting differences in the excess of AT will have a minor impact on the computed runoff fraction and are not at the surface in the region of this study. They will thus not effect the position of the computed runoff front in the PML.
 Another possible cause of errors is variability in the source water AT concentrations. Observed concentrations of dissolved inorganic carbon (slightly lower than AT as HCO3− is the main contributor to AT in runoff) in the main Arctic Rivers are 1300, 1200, 1100, 1900, and 1700 μmol kg−1 for the Ob, Yenisey, Lena, Yukon, and Mackenzie Rivers, respectively (recalculated from the work of Probst et al. ). Hence the mean concentration of AT used in our computations, 1412 μmol kg−1, should be of the right order, and our computation thus does not distinguish between rivers coming from Siberia or North America, added directly to the Arctic Ocean or to the Pacific waters that enter through the Bering Strait. A 10% difference in the river runoff source water AT concentration (more than 6 times the uncertainty in the estimate by linear regression of Figure 2) gives a linear change in river runoff contribution from zero at S = 34.85 to ±∼1% at S = 31. Lowering the sea ice melt salinity and AT to half will change the river runoff fraction by less than 0.2%. Changing the Atlantic source water AT by 23 μmol kg−1 (1%) will give a change of less than 1.5% on river runoff. All these uncertainties will shift the river runoff fraction but will not impact the location of the front and thus will not impact the coverage of the Eurasian Basin as presented below.
 The relative coverage of river runoff over the Eurasian Basin, computed as the area fraction over the deep basin along the sampling line with runoff content >5% (arrows in Figure 5), was about the same at ∼45% in 1987 and 1991. After that it decreased to ∼20% in 1994, had essentially disappeared in 1996, and then returned to ∼20% in 2001 (Figure 6). It should be noted that this representation is along the investigated section and does not necessarily represent the whole Eurasian Basin. This shift in river water from the Amundsen Basin to the Makarov Basin from 1991 to 1996 has also been shown using O-18 data [Schlosser et al., 2002].
 Even though the time series is short, it can be related to changes in the atmospheric circulation. Wind field changes as characterized by the Arctic Oscillation (AO) or North Atlantic Oscillation (NAO) indices are considered to be a major factor in determining ice and surface water flow [Proshutinsky and Johnson, 1997; Dickson et al., 2000; Maslowski et al., 2000, 2001]. The AO is characterized by lower sea level pressure (SLP) over the Arctic and stronger westerlies at subpolar latitudes when the AO index is high, and higher SLP over the Arctic and weaker westerlies when the AO index is low. It has also been shown that the regional atmospheric circulation and ice drift pattern in the Arctic Ocean is markedly different between different phases of the AO. A high AO index phase is characterized by a weaker Beaufort high situated closer to Alaska, and the transpolar ice drift is shifted toward the Canada Basin relative to a low index phase that has a stronger Beaufort high and transpolar drift centered over the Lomonosov Ridge. Of special interest here is that the ice drift over the Kara Sea and Laptev Sea is directed more eastward during phases with a high index compared to low index [Rigor et al., 2002]. Since the upper ocean is strongly forced by the prevailing winds, freshwater from the large rivers entering the Laptev and Kara Sea will take a more easterly route during high index phases, which in turn will have a large influence of the spreading of river water in the Eurasian Basin. Before 1988 the AO winter index varied around zero. This was followed by 6 years with high indexes, after which there was a return to years with indexes again around zero (Figure 6). The 1987 data show ∼45% coverage by river runoff when the AO winter index is around zero. Hence it seems that the runoff has been pushed away from the Eurasian Basin by high AO winter index, but with a lag time of 4–6 years. In 2001 the runoff has not come back to the coverage of before 1991, indicating that the retrieval might even take somewhat longer time.
 It is possible to explain the time lag from an estimate based on the geostrophic transport along the frontal region. Imagine that the Makarov Basin together with a part of the Eurasian Basin is covered with a pool of low-salinity water with a front somewhere in the middle of the Eurasian Basin and roughly parallel to the Gakkel Ridge. Suppose also that the along-frontal flow is in geostrophic balance. In a steady state, there must be a supply of low-salinity water at the upstream end of the frontal region that is likely to originate from the river runoff in the Kara and Laptev seas. When this source is deflected eastward during the positive AO phase, the source that maintains the Eurasian Basin part of the pool will shut off. The pool will then be emptied gradually, and the front will move toward the Lomonosov Ridge. The flow out of the pool can be estimated from the geostrophic flow, Q, in the frontal region using a two-layer approximation [Stigebrandt, 1987].
 Here g is the gravity acceleration, Δρ is the density difference between the upper and lower layer, H is the thickness of the upper layer, ρo is a reference density, and f is the Coriolis parameter. Using a linearized equation of state ρ = ρo(1 + βS) where β is the coefficient of salt contraction (β ≈ 8 · 10−4) and S is the salinity, it is possible to compute the flow using the salinity difference, ΔS, between the layers. Assuming that ΔS = 2 and H = 50 m, which are typical values for the salinity difference and layer thickness during 1991 [Anderson et al., 1994], gives a flow of 0.14 Sv. The pool of river runoff has left roughly half the Eurasian Basin, which equals an area of about 4.6 × 1011 m2. This gives a timescale of 5.5 years for emptying a 50-m-thick layer over this area, fitting quite well with the time lag indicated in Figure 6, although there are admittedly quite large uncertainties involved. A numerical model study by Maslowski et al.  supports in a qualitative way the suggested timescale showing significantly less freshwater in the Eurasian Basin during 1993 compared to 1983 as a result of a shift in the wind pattern. This indicates that 10 years is an upper limit of the timescale.
4. Concluding Remarks
 This work shows how total alkalinity can be used to separate the two freshwater sources, river runoff, and sea ice melt. By combining data from five crossings of the Eurasian Basin, we have been able to observe variability in river runoff distributions in the basin. In 1987 and 1991, river runoff was present over the Amundsen Basin but not in the Nansen Basin, corresponding to about 45% coverage of the Eurasian Basin. In 1994 the runoff front had moved closer to the Lomonosov Ridge, equal to ∼20% coverage, had essentially disappeared in 1996, and then returned to ∼20% in 2001. The shifts in the river runoff fronts are consistent with other observations of the presumed absence of river runoff in the central Eurasian Basin during the mid-1990s and its return [Steele and Boyd, 1998; Schauer et al., 2002; Björk et al., 2002]. It should be noted that the different cruises cross the Eurasian Basin at different locations, where especially the 1996 cruise was much farther to the east than the other cruises, especially over the Nansen Basin. Over the Lomonosov Ridge the sections are not that far spread out, but in 1996 the river runoff front was farthest from where much freshwater from rivers enters the Arctic Ocean, suggesting that the shift in the front was indeed most extreme during that year.
 The pattern of river runoff coverage shows a similar trend as the Arctic Oscillation winter index, but with a delay of 4–6 years. A 5-year delay is consistent with an estimate based on the geostrophic transport along the frontal region for the time required to empty the pool of water containing the river runoff. It is also similar to a qualitative result from the Maslowski et al.  numerical model study.
 The O-18 derived river water fractions obtained by Schlosser et al.  are 50–80% higher in the Amundsen Basin-Lomonosov Ridge region than the fractions that we have derived from AT. One difference between the two techniques is that with O-18 it is possible to distinguish between river runoff, sea ice melt, and brine that is excluded during sea ice formation, while only runoff and sea ice melt is separated with AT in this study. If the brine part is subtracted from the meteoric fraction as computed by the O-18 technique, the results become similar to our values of runoff fraction. In principle, it would be possible to compute the brine fraction by the AT technique, too, but it is a matter of the signal-to-noise ratio. Also, the question of the salinity of the surface water when the sea ice is actually formed adds to the uncertainty of that computation. What is important in this study is that the relative river runoff fraction and not the position of the computed front are affected.
 We wish to thank the Swedish Polar Research Secretariat for logistic support. This work was partially supported by grants from the Swedish Research Council and the Panel on Energy Research and Development (Canada).