The long-term data sets of total alkalinity (TA) (1929–2002 A.D.) and δ18O (1966–2002 A.D.) are used to investigate freshwater and brine distributions in the Arctic Ocean. Fractions of sea ice meltwater and other freshwaters (OF) (precipitation, river runoff, and freshwater carried by Pacific water implied as salinity deficit) are calculated on the basis of salinity-TA and salinity-δ18O relationships. Rejected brine during sea ice growth resides in surface water in the central Arctic Ocean, while net melting is found along the surface flow of water from the Pacific and Atlantic oceans. Distribution of OF at 10 m water depth suggests that Russian runoff leaves the shelf mainly west of the Mendeleyev Ridge, enters into the deep basin, and exits from the ocean through the western part of Fram Strait. The influence of Mackenzie River water is limited in the region and in depth. Accumulation of freshwater in the Canadian Basin is caused by deep penetration of OF with brine, indicating the transport of freshwater by shelf-derived water. The major origin of shelf-derived water entering into the upper halocline layer in the Canadian Basin should be the Chukchi and East Siberian Sea shelves, and the main freshwater sources are the salinity deficit of Pacific water and/or Russian runoff. An increase in OF inventory accompanied by an increase in brine content may suggest an increase of the shelf-derived water supply into the western Canadian Basin in anticyclonic years.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The Arctic Ocean receives 3300 km3 yr−1 [Aagaard and Carmack, 1989] of freshwater by river runoff, equivalent to ∼10% of runoff from all the world's rivers. Precipitation over the ocean and the less saline Pacific water coming through the Bering Strait provide additional freshwater. Freshwater carried by Pacific inflow is hereinafter referred to as the salinity deficit of Pacific water. In addition to these freshwater inputs, formation and melting of sea ice also alter the distribution of freshwater in the Arctic Ocean. Sea ice formation effectively removes freshwater, and rejected salt enters the underlying seawater as brine. Behavior of freshwater and brine in the Arctic Ocean has a key role in the regional and also the global climate by changing surface stratification, which affects the vertical heat transport in the Arctic Ocean and also the deep convection in the Greenland, Iceland, and Labrador seas.
 Many changes in water mass properties in the Arctic Ocean have been observed in the 1990s. Freshwater content in the Eurasian Basin decreased in the 1990s corresponding to the retreat of cold halocline water [Steel and Boyd, 1998]. Schlosser et al.  also report the decrease in meteoric water (river runoff plus precipitation) content and reduction in sea ice formation in the upper waters of the Eurasian Basin from 1991 to 1996. A significant return of the cold halocline water in the Eurasian Basin was confirmed in 2001 [Björk et al., 2002]. On the contrary, in the Canadian Basin, Macdonald et al.  observed surface freshening in the 1990s. Warming of the Atlantic layer in the Arctic Ocean has also been observed in the early 1990s [e.g., McLaughlin et al., 1996; Quadfasel et al., 1991; Carmack et al., 1995]. It seems to be caused by the increased inflow of warmer Atlantic water into the Arctic Ocean [Grotefendt et al., 1998; Dickson et al., 2000]. Increased transport of Atlantic water can displace the water mass in the Arctic Ocean and also alter freshwater distribution [e.g., Häkkinen and Proshutinsky, 2004].
 Model studies have shown that these observed changes in water mass and freshwater distributions in the Arctic Ocean associate with atmospheric forcing fluctuations. Proshutinsky and Johnson  found two regimes in the circulation in the Arctic Ocean: cyclonic and anticyclonic regimes. During the cyclonic circulation regime, freshwater on the Russian continental shelves flows eastward before it enters into the central part of the ocean [Johnson and Polyakov, 2001] followed by an increase of surface freshwater content in the Canadian Basin and a decrease in the Eurasian Basin [Maslowski et al., 2001]. A model simulation by Maslowski et al.  shows an eastward shift in the sea ice circulation, freshwater distribution, and Atlantic water extent in the 1990s and a reverse tendency in the late 1990s. The regime shifted from anticyclonic to cyclonic in 1988, and in 1997 it shifted in the opposite direction [Johnson et al., 1999].
 Although there are some observational changes to compare with modeled changes, it is still under debate whether observed changes in freshwater content in various locations in the Arctic Ocean are due to change in the storage volume or the redistribution because of the fragmented information from observations. Changes in river discharge from major rivers into the Arctic have been found [Johnson et al., 1999; Yang et al., 2002; Peterson et al., 2002]. Inflow of Pacific water and its freshwater content (as salinity deficit) also changes seasonally and interannually by mixing with meteoric water or sea ice formation/melting [Roach et al., 1995]. Furthermore, of course, change in transport pathways of these waters and sea ice also has large impacts on freshwater distribution in the Arctic Ocean, such as the shift of transpolar drift, which carries sea ice and surface freshwater from Russian shelf seas to the Atlantic Ocean.
 In order to obtain clear understanding of the driving mechanisms determining freshwater distribution in the Arctic Ocean it is highly desirable to know the distribution and abundance of freshwater from each major source and their temporal variations in the entire Arctic Ocean. There have been many efforts to distinguish each freshwater source in the Arctic Ocean by using chemical tracers such as the oxygen isotope ratio of water (river runoff plus precipitation versus sea ice meltwater/brine) [e.g., Östlund and Hut, 1984; Bauch et al., 1995], nutrients (Atlantic water versus Pacific water) [e.g., Jones et al., 1998; Ekwurzel et al., 2001], and Ba (American runoff versus Russian runoff) [Guay and Falkner, 1997; Macdonald et al., 1999; Taylor et al., 2003]. These tracer studies have provided good insight on freshwater behavior in the Arctic Ocean. However, the observational data set in their studies is still too limited in space and time to reveal the freshwater behavior and the controlling mechanisms in the whole Arctic Ocean, resulting in many unresolved speculations and hypotheses.
 There is a substantial data set of total alkalinity (TA) of seawater in the Arctic Ocean since 1929. Two sets of data have been recently open to the public: the World Ocean Database and the Hydrochemical Atlas of the Arctic Ocean [Colony and Timokhov, 2001]. The latter is compiled by the Arctic and Antarctic Research Institute (AARI, Russia) with the substantial financial support of the Frontier Research Program, Independent Administrative Institution, Japan Agency for Marine-Earth Science and Technology (JAMSTEC). The main objective of this study is to use the TA data for reconstructing freshwater and brine distributions and to differentiate freshwater sources, together with a conventional method using the oxygen isotope ratio. Since the use of TA as a freshwater and brine tracer has been limited to a certain region of the Arctic Ocean [Anderson et al., 2004], first its universal applicability in the entire Arctic Ocean is checked by comparing it with the estimates from oxygen isotope ratio. Then by combining results from two chemical tracers, spatial and temporal freshwater and brine distribution in the whole Arctic Ocean is examined for the transport mechanism in the surface and subsurface layer, including the cold halocline layer, the shelf-basin exchange, and the difference between cyclonic and anticyclonic regimes.
2. Theoretical Background
 The oxygen isotope ratio of water is expressed with a δ18O, defined as a per mil deviation of H218O/H216O ratio of a sample from that of an international standard water (VSMOW). The isotope ratio in seawater has been successfully used to separate contributions of meteoric water and sea ice meltwater in parts of Arctic Ocean for a few decades [e.g., Redfield and Friedman, 1969; Vetshteyn et al., 1974; Östlund and Hut, 1984; Melling and Moore, 1995; Schlosser et al., 2002] because the arctic meteoric water is largely depleted in 18O (δ18O ≈ −18‰), whereas δ18O of sea ice meltwater is close to 0‰ due to δ18O of ∼−2‰ of Arctic surface water and fractionation during ice formation (+2.6‰). Assuming that each seawater sample is a mixture of Atlantic water (ATW), meteoric water (MW), and sea ice meltwater (SIM), fractions of each end-member can be calculated by using the mass balance equations as follows:
where f, S, and δ refer to the fraction, salinity, and δ18O, respectively. The subscripted numbers (1, 2, and 3) indicate three end-members of ATW, MW, and SIM, respectively. Sea ice formation, which removes freshwater from seawater and sheds brine into seawater, is represented by a negative fraction of SIM. There is, however, one more water source flowing into the Arctic: Pacific water. To take the Pacific water into account, Bauch et al.  used silicate, and Ekwurzel et al.  and Schlosser et al.  used PO*4 as a additional tracer in a four-component mass balance. Macdonald et al.  and Macdonald et al.  used the polar mixed layer and upper halocline water, respectively, as the saline end-member in the three-component mixing instead of the Atlantic water. In the present paper, we use the end-member of “other freshwaters” (OF) instead of MW. The OF includes river runoff, precipitation, and salinity deficit of Pacific water. The reason for including it is presented later.
 The total alkalinity (TA) between freshwater sources is also different. Arctic rivers contain relatively high TA (∼1000 μmol kg−1 [Anderson et al., 1983; Olsson and Anderson, 1997]), and TA in natural sea ice (<300 μmol kg−1 at SSIM < 6 [Anderson and Jones, 1985]) and in precipitation (close to zero) are low. Anderson et al.  have tried to use TA for calculation of river runoff fraction in the Eurasian Basin of the Arctic Ocean. Since temperature, pressure, and biological processes cannot significantly change TA in the Arctic Ocean (except in the Barents Sea, where CaCO3-producing plankton exist [e.g., Smyth et al., 2004]), changes in TA must be related to the mixing of seawater with other water masses and river runoff, as well as to sea ice formation and melting. In this context, Anderson et al.  have used TA instead of δ18O to solve equations (1)–(3) by assuming mixing of three components of ATW, river runoff, and SIM. Note that the second component in their calculation is not meteoric water but river runoff. River runoff and precipitation are almost identical in δ18O value so that they can be represented together as a “meteoric water” end-member, but they are largely different in TA value, though meteoric water supply is thought to be dominated more by river runoff than precipitation in the region [Schlosser et al., 2002].
 We have TA data since 1929 (World Ocean Database 2001 and Colony and Timokhov ), and δ18O data since 1966 [Schmidt et al., 1999] for the Arctic Ocean. In the present paper, we evaluate the use of TA as a freshwater tracer and combine estimates from TA and δ18O to cover the whole Arctic Ocean and a longer time length in order to reveal synoptic freshwater and brine distributions and their variability in the Arctic Ocean.
Figure 1 shows observation sites of the historical and newly acquired data sets used in this study. The historical δ18O data come from Schmidt et al. . Data of TA are obtained from Colony and Timokhov  and the World Ocean Database 2001. Data from World Ocean Database 1998 included by Colony and Timokhov  are replaced by updated World Ocean Database 2001 data. TA and salinity data from three cruises of R/V Mirai in 1999, 2000, and 2002 in the Chukchi Sea, Beaufort Sea, and Canada Basin are also used [JAMSTEC, 1999, 2000, 2002]. We have measured δ18O of water samples collected during Mirai cruises by a mass spectrometer connected with a CO2-H2O equilibration unit. Precision of the measurement is ±0.03‰.
 Both TA and δ18O data are available from a cruise in the Eurasian Basin of the Arctic by the I/B Oden in 1991 (TA, Anderson et al. ; δ18O, Bauch et al. ). In sections 4, 5, and 6 we use data from the Oden cruise together with data from Mirai cruises to represent water properties in the Eurasian and Canadian basins, respectively; to find appropriate end-member values; and to compare estimates from TA and from δ18O.
 Before using historical data we filtered data as follows. First, data flagged in original databases as “questionable,” “bad,” or “very anomalous” were excluded. We have further filtered TA data using three more steps.
 1. The TA data from stations shallower than 50 m depth have been deleted. These stations are largely scattered in a salinity-TA plot due to seasonal TA variability in the river water, which is high in winter and low in spring [Olsson and Anderson, 1997]. Since we use a TA value for an OF end-member as an annual average, stations locally affected by seasonal variability of river TA are excluded.
 2. TA data in the Barents Sea (20 <°E ≤ 55 and °N ≤ 78) have not been used. TA in this area shows biological alteration by drawdown in the salinity-TA plot, probably due to calcification by coccolithophore. Coccolithophore blooms have been observed in satellite imagery in the Barents Sea [Smyth et al., 2004]. The large drawdown of TA, found in the region north of Kola Peninsula, is consistent with the area of coccolithophore blooms in satellite images. However, drawdown of TA is not observed north of 78°N, where the water enters into the deep Arctic Ocean. Therefore we have simply excluded Barents Sea TA data south of 78°N for the analysis.
 3. To find the outliers, the mean TA values and standard deviations of deep water (>1000 m) are calculated for each observation station. On the basis of the data obtained by the Oden and Mirai cruises the average TA value is determined to be 2309 μmol kg−1 (standard deviation equals 7.3) for the Arctic deep water. Therefore we have set a possible range of deep water TA value in the Arctic Ocean as 2287–2331 μmol kg−1 (average plus or minus standard deviation times three). Stations having a mean deep water TA out of this range, or a standard deviation larger than 20 μmol kg−1, were eliminated from our database. If this elimination was required for more than 90% of stations for a particular cruise, all data from the cruise were omitted. Data from eight cruises in the historical database were excluded by this last step. The number of data points in the final data set is 9229 and 7861 for δ18O and TA, respectively. We still have substantial TA data, although more than half of data in the database were excluded by these tests. The use of TA together with δ18O doubles the number of data points of freshwater tracers in the Arctic Ocean. Figures 1 and 2show distribution of data in space (Figure 1), time (Figure 2a), and water depth (Figure 2b). As apparent in Figure 2a, observations are frequent after the 1970s. The first observation in the deep Arctic Basin was in 1948, and data before that are from regions around Svalbard Island and in the Kara Sea. Sampling is most frequent in the surface water, and ∼5000 data points are available for the surface to 30 m depths (Figure 2b).
4. End-Member Properties
Table 1 gives end-member values used in this study. As has been mentioned, a three-component mixing scheme of ATW, SIM, and OF is considered in this study. The freshwater sources to the Arctic Ocean are sea ice meltwater, river runoff, precipitation, and the salinity deficit of Pacific water. Among them, we cannot distinguish Pacific water from the mixture of Arctic meteoric water and ATW. This is because salinity-δ18O and salinity-TA properties of Pacific water lie on the mixing line of ATW with Arctic meteoric water [Ekwurzel et al., 2001; Anderson et al., 1994]. For example, combinations of salinity 32.7 and δ18O −1.1‰, or salinity 31.5 and TA 2173 μmol kg−1, characteristics of Pacific water from Ekwurzel et al.  and Anderson et al. , respectively, produce fSIM of −0.01 and fMW of 0.07 or 0.11 from equations (1)–(3) (replace δ with TA in equations for TA) with values in Table 1. Thus the Pacific water can be represented as a mixture of ATW and meteoric water. This is why we include the salinity deficit of Pacific water into OF with river runoff and precipitation. The δ18O value of OF has been set to −18 ± 2‰, weighted mean value of Arctic meteoric water [Ekwurzel et al., 2001]. Here we should mention the variability of Pacific water properties. As measured by Roach et al. , salinity of Pacific water changes seasonally and interannually because of fluctuations in mixing with meteoric water and ice formation/melting. However, these processes influence not only salinity but also δ18O and TA, as they do in the Arctic Ocean. Therefore low (high) salinity of Pacific inflow because of high (low) content of meteoric water or SIM in the Bering Sea will be carried into the Arctic Ocean as signals of high (low) content of OF or SIM.
Table 1. End-Member Values Used in This Study
TA μmol kg−1
Average ± standard deviation of values at temperature maximum layer of each station in the Eurasian Basin (Oden cruise).
 To determine the end-member composition of ATW, values at the temperature maximum of each station of the Oden cruise were picked out, and the mean values were obtained. These values are used to represent ATW properties with a range of standard deviation. The value of SSIM of 4 ± 1 follows Ekwurzel et al. , and TASIM of 263 μmol kg−1 is from Anderson et al.  with ±25% uncertainty, assuming it has a variability similar to SSIM.
 A fixed value of sea ice δ18O is used for the whole Arctic Ocean, rather than calculating a value for each station using the surface water value plus the fractionation factor between sea ice and water as used by Ekwurzel et al. , because sea ice can move from its formation area to its melting area and we do not know from which surface water the sea ice formed. It is assumed therefore that sea ice forms from the surface water with salinity ranging from 28 to 32, a range that covers most of the surface salinity in the Arctic Ocean [Pokrovskii and Timokhov, 2002]. The δ18O range of surface water before freezing can be obtained as −3.3 to −1.3‰ from the mixing line of ATW with OF. By considering a fractionation factor of 2.6 [Ekwurzel et al., 2001, and references therein], δSIM is estimated to be 0.3 ± 1.0‰.
Anderson et al.  used the TA value of 1412 μmol kg−1 as the end-member value of river runoff in the Eurasian Basin. However, as they mentioned, the estimated fraction of meteoric water from TA is 50–80% lower than fractions obtained by Schlosser et al.  using δ18O in the same region. This might come from the fact that they obtained the value of 1412 μmol kg−1 from the intercept of a fitted line in salinity-TA plot for data collected north of the Laptev Sea. These samples would be affected by sea ice meltwater or formation, which changes both salinity and TA. Therefore we need to correct for the effect of sea ice before fitting a regression line to select the end-member value of TA for OF by
where S0 and TA0 are salinity and TA corrected for the influence of sea ice melting or formation. For the Oden cruise and Mirai cruises, fSIM can be calculated for each sample by using salinity and δ18O data and equations (1)–(3) to obtain S0 and TA0. The resulting relationship between S0 and TA0 is shown in Figure 3b. For the Oden cruise, the regression line of the observed salinity-TA plot is TA = 26.64S + 1379 (R = 0.87, where R is the correlation coefficient) (line A in Figure 3a). The intercept of 1379 μmol kg−1 is close to that of Anderson et al. . However, after correction for sea ice melting or formation the regression line becomes TA0 = 45.24S0 + 731 (R = 0.99) (line B in Figure 3b). The value of 731 (±30 at 95% confidence level) μmol kg−1 should be the representative TAOF in the Eurasian Basin. Interpretation of the value of TAOF is discussed in section 5.
 For Mirai cruises, relationships between TA and salinity observed and corrected for the influence of sea ice formation/melting are presented in Figures 3c and 3d, respectively. The deflection at salinity around 33 (above line B) and downward deviation around 25 (below line B) found in the observed salinity-TA plot (Figure 3c) disappeared in the TA0-S0 diagram (Figure 3d). This implies that the upward deflection in the salinity-TA relationship at salinity of 33.5 is caused by brine rejection and the downward deviation at salinity of 25 is due to mixing with sea ice meltwater (see dashed arrows in Figure 3a). After correcting for these sea ice effects, most samples from Mirai cruises are distributed close to line B. This indicates that the TA value of OF in this region is close to that in the Eurasian Basin. However, some samples deviate toward a higher TA0 value in Figure 3d. Deviations are anomalously high for low-salinity water near the Mackenzie River mouth. These deviations signify the influence of the Mackenzie River entering the ocean from North America, since it has higher TA value (1900 μmol kg−1 [Telang et al., 1991]) than other freshwater sources such as Russian rivers (770), precipitation (0), and the salinity deficit of Pacific water (930). Large deviations of more than 50 μmol kg−1 above line B are found in the upper 30 m at stations marked with dots in Figure 1b. Except for these samples the S0-TA0 relationship does not largely differ from the regression line of the Oden cruise. This means that the obvious influence of the Mackenzie River is very restricted both in region and in depth. Since outlying samples in S0-TA0 produce a fraction of OF higher than 0.2, we have concluded that our calculations using TA in the Canadian sector can be reliable only for waters of fOF < 0.2. There are four samples from three stations having fOF > 0.2 in the TA data in the Canadian sector other than Mirai data. They are not included in our analysis, and locations of these samples are shown in Figure 1b by triangles. Except for samples having fOF < 0.2, samples from Mirai cruises give a regression line of TA0 = 39.31S0 + 930 (R = 0.99), and 930 (±13 at 95% confidence level) μmol kg−1 should be the mean TAOF for this region. Since the relationship between TA0 and S0 is linearly correlated with high correlation coefficients in both the Eurasian and Canadian sectors and intercept values are not largely different in the two regions, we conclude that a single value of TA (831 ± 100 μmol kg−1; the median) would represent the end-member value for OF for the entire Arctic Ocean.
5. Origin of OF in the Arctic Ocean
 We have found that the TA value of OF is 731 ± 30 μmol kg−1 and 930 ± 13 μmol kg−1 for the regions of the Oden and Mirai cruises, respectively. What is the source of OF in these regions? There are several freshwater sources contributing to OF in the Arctic Ocean with different TA values. Olsson and Anderson  reported the weighted mean concentration of dissolved inorganic carbon (CT) of 770 μmol kg−1 for Russian rivers. Since HCO3− is the dominant carbonate species in the river water, the TA of Russian rivers should be very close to 770 μmol kg−1. This is close to the value of OF in the Eurasian Basin, suggesting that OF is dominated by runoff from Russian rivers. Precipitation over evaporation (P-E) is another possible source of OF in this region, and its TA value is ∼0. Annual freshwater flux by P-E is ∼1500 km3 yr−1 [Serreze and Barry, 2000], accounting for ∼30% of meteoric water input to the Arctic Ocean. However, contribution of P-E to OF in this region is calculated to be ∼5% from these TA values (731 = 770(1 − fprecipitation). This may indicate that a large part of snow that falls onto the ice-covered Arctic Ocean is exported from the ocean with sea ice before melting into seawater.
 For the Canadian sector the TAOF is found to be 930 ± 13 μmol kg−1. The Pacific water flowing into this region (salinity 31.5, TA 2173 μmol kg−1) produces the intercept of 930 μmol kg−1 at S = 0 when mixed with ATW. The contributions of low-TA precipitation (TA = 0 μmol kg−1) and high-TA Mackenzie runoff (TA = 1900 μmol kg−1) can add to OF, and their combined effects on TAOF can offset one another. However, even assuming a high contribution of precipitation (30% of meteoric water), the TA of the mixture of Mackenzie runoff with precipitation is 1330 μmol kg−1, still higher than 930 ± 13 μmol kg−1. This suggests that Mackenzie runoff is not the dominant source of OF, even in the Canadian sector. Assuming that OF in this region consists of OF in the Eurasian Basin (mainly Russian river runoff), the salinity deficit of Pacific water and Mackenzie River water, the fraction of Mackenzie River runoff in the OF will be less than 17% (930 = 1900fMackenzie + 930fPacific + 731(1 − fMackenzie − fPacific)). It will be <23% when precipitation is taken into account (930 = 1330/0.7fMackenzie + 930fPacific + 731(1 − 1/0.7fMackenzie − fPacific)) and the contribution of precipitation is <10% (<23 × 3/7). Therefore Russian runoff and the salinity deficit of Pacific water should be the main origin of OF in this region and hence in the Arctic Ocean.
6. Comparison of Estimates From Two Tracers
 With end-member values described in section 4 and summarized in Table 1, fractions of OF and SIM are calculated from the salinity-TA and salinity-δ18O relationships. Vertical distributions of fractions of these freshwater sources in the Eurasian Basin (Oden 91) and Canadian sector (Mirai) are shown in Figure 4 for comparison. The same features of freshwater distribution are well represented in results from both tracers; for example, most of surface water contains rejected brine from sea ice (negative fSIM) in the Eurasian Basin and meltwater (positive fSIM) in the Canadian sector, and water at the depth of 50–250 m in the Canadian sector contains higher fractions of brine and OF than water in the Eurasian Basin. Moreover, estimated values of fractions from the two tracers agree with each other. Direct comparison of each result shows that the difference between fractions estimated from the two tracers is smaller than 0.03 for 94% and 96% of data for SIM and OF, respectively. Linear regressions of fractions estimated from salinity-TA versus those from salinity-δ18O from Oden 91 and Mirai cruises (Figure 5) give equations of ffrom TA = 1.017ffrom δ18O + 0.002 (R = 0.87) and ffrom TA = 1.084ffrom δ18O − 0.003 (R = 0.97) for fSIM and fOF, respectively. From these analyses it may be concluded that TA can be a useful tracer for freshwater and brine in the Arctic Ocean and estimated fractions from salinity-TA can be combined with those from salinity-δ18O. The calculated fractions range within ±0.03 and ±0.04 by changing δOF and TAOF within their uncertainties, respectively, while they range within ±0.01 by similarly varying the other end-member values. TA data in the historical data set were measured by the volumetric analytical and potentiometric methods, and precision of these methods is ∼12 μmol kg−1 or less [Colony and Timokhov, 2001]. The precision of δ18O measurements is usually smaller than ±0.1‰. These can change estimated fractions within ±0.015.
7. Freshwater Distribution in the Arctic Ocean
Figures 6 and 7 show distributions of fractions of OF and SIM in seawater at 10 and 150 m water depths of the Arctic Ocean, respectively. Estimates within the top 30 m of the water column, where observations are the most abundant (Figure 2b), are used and interpolated to obtain freshwater fractions at 10 m depth (Figure 6). For Figure 7, estimates in 100–200 m depths are used. A wider depth interval for Figure 7 is set because of less data availability for the intermediate layer compared to the surface layer (Figure 2b). According to an examination using Mirai and Oden data, interpolation of estimates at 100 and 200 m depths to obtain a value at 150 m depth does not differ more than 0.03 (N = 41, where N is the number of stations) from the value directly estimated from data observed at 150 m. Since the surface layer shallower than 50 m has a strong seasonal variability [McLaughlin et al., 2004], the distributions at 10 m depth in summer (June–October) and in winter (November–May) are shown separately in Figures 6b and 6c, respectively. Note that the Chukchi Sea and Eurasian marginal seas were observed mostly in summer.
 The fraction of OF at 10 m depth in the Chukchi Sea is 5–10%. This fraction can be explained by inflowing Pacific water alone since it carries the OF fraction of 7–11%. This implies a small influence of Arctic meteoric water in the Chukchi Sea. On the other hand, surface waters in Siberian and American coastal areas contain more than 20% (note that fOF > 0.2 in the Canadian sector is only from δ18O), and a high fraction of OF (>10%) is distributed over the Canadian Basin (Canada and Makarov basins) and extends toward the western part of Fram Strait. These areas are obviously influenced by Arctic meteoric water. A front in fOF distribution lies along the Eurasian side of the Lomonosov Ridge, though the high-fOF water seems to leave the Russian shelves mainly west of the Mendeleyev Ridge.
 The surface waters in the Chukchi Sea and in the region from Fram Strait to the Kara Sea contain a relatively high fraction of SIM, >3% in summer. These regions receive inflowing water from the Pacific and Atlantic oceans, which melt sea ice on their way into the Arctic Ocean. On the other hand, negative fractions of SIM are found in the central part of the Arctic Ocean and to the east of Greenland, where sea ice covers the surface throughout the year. Negative fSIM indicates the existence of brine rejected from growing sea ice. Remarkably, negative fSIM occurs at 10 m depth on the continental shelves of the Beaufort and East Siberian seas in winter (Figure 6c) and in the Laptev Sea in summer (Figure 6b). Since sea ice formation on shallow shelves can efficiently accumulate rejected brine because of limited convection depth, the high brine content in shelf water is not surprising, especially in winter. In the Laptev Sea, water at 10 m depth has very low salinity (S < 25 at 10 m) due to the large influence of Lena River water despite its high brine content. The depth of 10 m is at or below the sharp pycnocline observed at the shallow Laptev Sea shelf close to the Lena River mouth [Létolle et al., 1993]. The pycnocline could retain the winter water with a substantial amount of brine even in summer by preventing mixing with warmer and fresher surface water.
 At 150 m depth (Figure 7), fractions of OF and SIM are close to zero in the eastern Arctic Ocean (Figure 7), showing that Atlantic water is distributed with no large modification by freshwater input or sea ice formation/melting. High OF fractions are found in some regions with negative SIM fractions. These regions are the Canadian Basin, Baffin Bay, and east of Novaya Zemlya in the Kara Sea. A combination of high OF content and brine implies the existence of water formed during winter. The atmospheric cooling and brine rejection during winter converts fresh surface water into denser water with high OF and high brine content. Stations with these signals in the Kara Sea and Baffin Bay correspond to regions where deep convection (down to 250 m or more) occurs in winter [Pavlov and Pfirman, 1995; Lemon and Fissel, 1982]. In the Canadian Basin, both fOF and brine are distributed widely at the 150 m depth. This depth is between the surface and Atlantic origin water and corresponds to the depth of a nutrient maximum in the Canadian Basin. This nutrient maximum, called the upper halocline water (UHW), has a salinity of ∼33 [Jones and Anderson, 1986] and is believed to be maintained by advection of water from the adjacent shallow continental shelves. The shelf-derived water formed during sea ice formation enters into the interior ocean and carries freshwater and brine from the surface to the layer between the surface and Atlantic layers.
 In Figures 6 and 7, outflow of the OF from the Arctic Ocean to the Atlantic Ocean occurs with the East Greenland Current in the western part of Fram Strait [Anderson and Dryssen, 1981; Östlund and Hut, 1984]. This outflowing OF must consist of Russian runoff and the salinity deficit of Pacific water according to a multitracer analysis (Ba, nutrients, and δ18O) by Taylor et al. . They reported that North American river water is undetectable from the surface to 200 m depth in Fram Strait. They suggested that North American runoff must be stored in the Beaufort Gyre and/or must drain through the Canadian Archipelago. The former was inferred from the high Ba in the Canadian Basin, which can originate in the Mackenzie River because the river contains a higher Ba concentration than the Eurasian rivers [Guay and Falkner, 1997, 1998; Macdonald et al., 1999]. The high content of OF can be seen in the Canadian Basin in Figures 6 and 7. The TAOF of 930 ± 13 μmol kg−1 for Mirai cruises, however, designates Russian river runoff and salinity deficit of Pacific water rather than Mackenzie River water as the main source of OF in the Canada Basin. The influence of the Mackenzie River is found only in the upper 30 m depth at several stations (Figure 1b) at least during Mirai cruises from 1999 to 2002. Most of the Mackenzie River water likely flows out from the Arctic Ocean relatively fast, probably through the Canadian Archipelago as suggested in the modeled flow by Karcher and Oberhuber . Thus the high Ba concentration in the Canada Basin must be explained by other sources than the Mackenzie. Although Russian rivers are low in Ba compared with the Mackenzie River, their Ba concentrations are nevertheless higher than the Ba concentration in seawater. As mentioned by Guay and Falkner , not only the Mackenzie River water but also Russian river waters can be the fluvial source of the observed Ba-salinity relation in the surface water. For example, if the mixing ratio of runoff and ATW (42–45 nmol Ba L−1) is 2:8, and if runoff consists of 20% of Mackenzie River water (520 nmol Ba L−1) and 80% of Russian river water (100–200 nmol Ba L−1), then the resulting mixed water has a salinity of 28 and a Ba of 70–89 nmol L−1 (Ba concentrations are from Guay and Falkner ). This estimated value is comparable with observed surface Ba concentration in the Canada Basin (75 nmol Ba L−1). Mixing of ATW only with Russian river runoff at 2:8 can account for Ba of 54–76 nmol L−1. Furthermore, Pacific water can supply additional Ba into the Canadian Basin since its Ba content (50–60 nmol Ba L−1) is higher than ATW. These simple calculations clearly demonstrate that high Ba in the Canada Basin does not necessarily imply the dominance of a Mackenzie River influence. Moreover, Ba is not a conservative element. Ba can be removed from surface water by biological activity, increased by diffusion from sediment, and supplied in estuaries by desorption from clays [Guay and Falkner, 1997, and references therein]. Of course, Mackenzie River water must largely influence surface water in some regions. As mentioned in section 4, some surface samples deviated in the S0-TA0 diagram toward the higher TA0 value (see Figure 1b for locations). These waters are clearly influenced by Mackenzie River water and could have high Ba concentrations as observed by Macdonald et al.  (>120 nmol Ba L−1, with high fraction of meteoric water >0.25). Nevertheless, the main freshwater source for the Canadian sector of the Arctic Ocean in this study is found to be Russian river runoff and salinity deficit of Pacific water.
 From the distribution of OF in Figure 6 with the above considerations, the surface flow pattern of river water can be depicted. Russian river water enters into the Siberian shelf seas, flows out of the shelf mainly west of the Mendeleyev Ridge, and enters into the interior ocean. A portion of this water enters into the Canada Basin, probably by being incorporated into the Beaufort Gyre. Most of Mackenzie River water likely flows out from the Arctic Ocean through the Canadian Archipelago. Figure 8 summarizes distributions and possible flow paths of freshwater described above.
Figure 9 shows the horizontal distribution of freshwater content, integrated from the surface to 300 m depth (or to the depth of bottom measurement if it is shallower than 300 m). The depth of 300 m was selected because most freshwater is restricted in the upper 100 or 250 m of water column in the Eurasian and Canadian basins, respectively, and fractions of OF and SIM are close to zero at 300 m depth (Figure 4). Integrations are done only for stations having measurements >5 in the top 300 m (or N > 1.7/100 m for shallower stations). Figure 9a shows freshwater accumulation in the Canadian basin. Total freshwater inventory is >10 m in the basin, with the maximum of 20 m. In the Eurasian Basin, total freshwater inventory is <10 m and decreases toward the mainstream of the ATW along the Barents Sea slope. In the Canadian Basin the freshwater inventory for OF is >20 m (Figure 9b), and sea ice formation removes 10 m or more freshwater from the water column (Figure 9c), resulting in a net inventory of >10 m of freshwater. Figure 9 also shows that the water inflowing from the Atlantic Ocean observed in the eastern part of Fram Strait will exit along the western part of the strait after experiencing 5 m of net sea ice formation and receiving 10 m of OF within the Arctic Ocean.
8. Transport of Freshwater and Brine Into UHW in the Canadian Basin
 Large inventories and high contents at 150 m depth of both OF and negative SIM (brine) in the Canadian Basin (Figures 7 and 9) are produced by deeper penetration in the Canada Basin than the Eurasian Basin to the maximum depth of 250 m (Figures 4 and 7). This depth corresponds to the bottom of the upper halocline layer in the Canadian Basin. The winter Pacific water is assumed to be the main source of UHW because of its high nutrient concentrations. Inflowing winter Pacific water through the shallow Bering Strait (30–50 m) increases in salinity due to the inclusion of rejected brine from growing sea ice, and it loads additional nutrients by interaction with bottom sediments on the shallow shelf on its way to the deep Arctic Ocean [Jones and Anderson, 1986; Cooper et al., 1997]. The initial salinity value before sea ice formation, S0, calculated from equation (4) with fSIM obtained from salinity-δ18O or salinity-TA, reveals that the UHW of salinity of about 33 is formed from water having salinity of about 31–32, indicating that brine from sea ice increased the salinity of the seawater by 2–1. Measurements of salinity through the Bering Strait by Roach et al.  show that the mean salinity of Pacific water inflowing through the Bering Strait is 32.0 in the eastern and 32.6 in the western channels. According to model studies [Overland and Roach, 1987] as well as an observation by Roach et al. , more than 60% of inflowing water comes from the Anadyr Strait, mainly through the western channel of the strait. Taking this into account, the S0 of 31–32 of UHW is lower than the expected value for winter Pacific water. In both channels, salinity increases during winter and reaches to a maximum of 33 or more in March. Such winter Pacific water can directly enter and contribute to the UHW, as has been thought. However, since calculation of S0 excludes the salinity change due to ice formation/melting, S0 of this saline water (S > 33) should be around 32–32.6. In other words, more or less formation of ice changes salinity of water but does not change S0 of water. Thus S0 of 31–32 implies that other additional processes are necessary to make the UHW salinity 33.
Melling and Moore  have observed winter shelf water in the Beaufort Sea shelf. They describe that preconditioning of shelf water salinity by weather events before freezing is a key to making winter shelf water dense enough to contribute to the UHW. In some years, dense shelf water successfully forms on the Beaufort Sea shelf, but its nutrient concentration is not sufficient to retain a nutrient maximum in the UHW. The water having salinity of about 33 observed on the Beaufort Sea shelf has an S0 of <31.5. Mixing of this water with nutrient-rich winter Pacific water having an S0 of >32 can form the UHW having S0 of 31–32.
Ekwurzel et al.  have also pointed out that the East Siberian Sea close to the Chukchi Abyssal Plain is a location where renewal of UHW occurs. A high value of fOF and negative value of fSIM extending northward along the 180° line in Figures 7 and 9 indicate the outflow of shelf water from the west of the Chukchi Sea. This feature comes from TA data observed at the shelf edge of Chukchi/East Siberian seas in as early as 1948 and 1979, although the data might have a relatively large error. However, the layer of negative fSIM from 50 to 150 m depths in these stations, with salinity in the range of 32–34, corresponds to the layer of low temperature and high silicate concentrations (not shown). These facts support the existence of water formed by cooling and brine rejection in this region. This water can enter into the UHW and deeper layer when it leaves the shelf. The calculated S0 for this water ranges from 28 to 32 and is 30–31 for the water with salinity of 33. Although this region receives inflowing Pacific water, its low S0 range suggests that the observed shelf water is not formed only from winter Pacific water. Winter Pacific water needs to mix with fresher water before rejected brine increases its salinity to form observed shelf water in the Chukchi/East Siberian seas. Such freshwater would originate from Russian runoff, which flows into the Chukchi/East Siberian seas as the Siberian Coastal Current [Weingartner et al., 1999] and/or fresher summer Pacific inflow. Although Weingartner et al.  presume that the absence of a fresh Siberian Coastal Current allows the formation of UHW source water in this region, our analysis suggests that Russian runoff or freshwater from summer Pacific inflow is carried into the UHW.
 At least three types of shelf-derived water can be suggested to form the UHW at salinity of 33 from the historical tracer data: (1) winter Pacific water with rejected brine from sea ice (S0 > 32 and salinity increase by brine (ΔS = 33 − S0) is <1), (2) shelf water formed on the Beaufort Sea shelf (S0 < 31.5, ΔS > 1.5.), and (3) shelf water formed on the Chukchi/East Siberian seas by mixing of winter Pacific water with fresher shelf water containing summer Pacific water and/or Russian river water and brine (S0 ∼ 30–31, ΔS ∼ 2–3).
 The high concentration of nutrients of the UHW points to the winter Pacific water as a major source, but the low S0 of UHW indicates a significant contribution of other shelf waters, which should be formed from fresher water but contain more brine. Furthermore, the TA0-S0 plot for Mirai cruises (Figure 3d) indicates that OF in the Canadian sector consists of mainly Russian river water and the salinity deficit of Pacific water but not much of the Mackenzie River water except surface water at several stations. This suggests that the contribution of the second type of shelf water (Beaufort Sea shelf water) to the UHW in the Canadian Basin is small and highlights the importance of the third type of shelf water (Chukchi/East Siberian seas shelf water). These shelf-derived waters transport both OF and brine from the surface to the deeper layer in the Canadian Basin and make the basin their reservoir.
9. Changes in Freshwater Distribution Between Cyclonic and Anticyclonic Regimes
 The interannual variability in distributions of freshwater sources from the long-term hydrochemical observations merits close examination. Even though there are data for more than 70 years in our database, they are still not enough to present a sequence of freshwater content for comparison with time series of atmospheric forcing or indices. We therefore have divided our data into two groups to present variability in freshwater distribution.Proshutinsky and Johnson  discuss the existence of two circulation regimes of wind-driven circulation in the modeled Arctic Ocean: cyclonic and anticyclonic regimes, which are the response of the ocean to atmospheric fluctuations. Our results have been divided into two groups of cyclonic years (1953–1957, 1964–1971, 1980–1983, 1989–1996, and 2001–2002) and anticyclonic years (1946–1952, 1958–1963, 1972–1979, 1984–1988, and 1997–2000) following Proshutinsky and Johnson . Most of the data in the cyclonic regime are observed during a single period of 1989–1996 (Figure 2a). Even with this split, data are still limited in spatial coverage (Figures 10 and 11), especially for cyclonic years. Since our data cover different regions in different seasons and in different regimes, it is difficult to assess changes in freshwater content quantitatively and to discuss differences in surface distribution between the two regimes. Therefore we note that discussions on horizontal distribution of freshwater in section 7 and the scheme in Figure 8 are based on a composite of two regimes. We mention only the change in freshwater inventory in the region observed in both regimes.
 According to model simulations the anticyclonic Beaufort Gyre in the Canadian Basin has a wider distribution in anticyclonic years than in cyclonic years [e.g., Proshutinsky and Johnson, 1997; Häkkinen and Proshutinsky, 2004]. Freshwater and brine accumulated in the basin are expected to be redistributed more widely in anticyclonic years. Despite the limitation of data in cyclonic years in the central Canadian Basin, observed contents of total freshwater, OF, and brine seem to be consistent with expected changes. In Figure 10b the region of high OF content (20 m or more) is restricted to the eastern part of the Canadian Basin and does not cross the 180° line. In Figure 11b, stations with similarly high OF content are evident throughout the Canada Basin, even beyond the 180° line. The same difference is found in brine distribution (Figures 10c and 11c). The area off Point Barrow, Alaska, to the north along the 180° line in the western Canadian Ocean is the region where inventories of OF and brine largely increased from cyclonic to anticyclonic regimes. This is the region where the water enters the interior Arctic Ocean from the Chukchi and East Siberian shelves. Therefore increase of OF and brine can be caused not only by redistribution of Canadian Basin water but also by active input of shelf water formed during sea ice formation. Melling  observed warming of UHW in the Canada Basin and suggested a decrease in ventilation over the shelves on the Pacific side caused by a weakened Siberian High in the cyclonic 1990s. This might support the idea that brine-injected shelf water transport into the western Canadian Basin is greater in anticyclonic years. As Melling and Moore  and Cavalieri and Martin  suggested, initial salinity of seawater before freezing commences is critical to form dense shelf water in order to be able to transport brine and OF vertically into the UHW. Therefore, even if production of ice is lower in anticyclonic years in the Chukchi/East Siberian seas [Zhang et al., 2003], higher initial salinity would increase the volume of dense water entering into the UHW and then subsequently increase the inventories of OF and brine in the western Canadian Basin. As the circulation regime changes from cyclonic to anticyclonic, the mainstream of freshwater from the Siberian shelves shifts westward [e.g., Steel and Boyd, 1998; Maslowski et al., 2001], and surface salinity would increase in the Chukchi/East Siberian seas. Therefore the higher inventories of OF and brine in the western Canadian Basin in Figure 11 as compared with Figure 10 may suggest the active transport of shelf-derived water into the basin in anticyclonic years.
 In any case, freshwater distribution in the western Canadian Basin north of Chukchi and East Siberian seas is revealed to be remarkably sensitive to regime shifts. As our data are still not sufficient in spatial and temporal coverage, a detailed comparison with model simulations would likely reveal information lurking in our database and help us to better understand the variability of freshwater and brine content and behavior in the Arctic Ocean.
 A comparison of salinity-TA with salinity-δ18O methods has revealed that TA can be used as a tracer of freshwater and brine as well as δ18O. TA values for OF are found to be 731 μmol kg−1 in the Eurasian Basin and 930 μmol kg−1 in the Canadian Basin after correcting for the effect of sea ice formation/melting by using δ18O. TAOF in the Eurasian Basin is much smaller than the 1412 μmol kg−1 proposed by Anderson et al.  but is close to the TA value of Russian river water (770 μmol kg−1 [Olsson and Anderson, 1997]). Estimated TAOF in the Canadian Basin is lower than one might expect from the high TA value of Mackenzie River water (1900 μmol kg−1 [Telang et al., 1991]).
 With appropriate end-member values, including TAOF of 831 ± 100 μmol kg−1, fractions of SIM (brine as negative value) and OF estimated from salinity-TA relationship are in close agreement with those from the salinity-δ18O relationship. Therefore results from both tracers are combined to show the freshwater and brine distributions in the Arctic Ocean based on historical data of TA since 1929 and δ18O data since 1966. The composite results show the synoptic feature of distribution of freshwater and brine in the entire Arctic Ocean. Figure 8 schematically summarizes the freshwater and brine distribution in the surface Arctic Ocean and regions where the UHW is formed. Sea ice melts at the surface along the relatively warm inflow from the Pacific and Atlantic oceans. Brine rejected during sea ice formation is distributed in the surface layer of the central Arctic Ocean, and the western part of Fram Strait. River water flows out from the Russian coastal seas, enters into the deep basin west of the Mendeleyev Ridge, and eventually flows out through Fram Strait.
 Influence of Mackenzie River water was limited in region and in depth, at least during the Mirai cruises from 1999 to 2002. The fraction of Mackenzie River water in the OF in the Canadian Basin is estimated to be ∼23% or less, indicating the importance of Russian river water and the salinity deficit of Pacific water as freshwater sources in the Arctic Ocean, even in the American side of the ocean. Therefore most of Mackenzie River runoff likely flows out of the ocean rapidly through the Canadian Archipelago. Deep penetration of freshwater is found in the Canadian Basin. Shelf-derived water from the Chukchi and East Siberian seas is suggested to carry and accumulate OF and brine into the UHW in the layer between surface and Atlantic layers in the Canadian Basin.
 The western Canadian Basin is found to be an important region, not only as an area of deeper transport of freshwater and brine but also as a highly variable region in response to atmospheric fluctuations. The anticyclonic circulation regime alters the flow of Russian river runoff westward [e.g., Maslowski et al., 2001] and increases the salinity of the surface layer in this region. This is a favorable condition for active transport of brine-injected shelf water into the Canadian Basin. This suggests an increase in the supply of shelf-derived water from the Chukchi and East Siberian seas under anticyclonic circumstances. Although the East Siberian Sea is expected to be a very sensitive region to atmospheric variability or changes, it is one of the least observed areas in the Arctic Ocean. Changes in this sea may affect the entire ocean through the transpolar drift system and through the formation of UHW, which is presently preventing the contact of warm Atlantic water with sea ice.
 We appreciate A. Murata, K. Shimada, and T. Takizawa of JAMSTEC for providing data from the past Arctic expeditions by the R/V Mirai for this study. We are greatly indebted to L. A. Timokhov and research scientists and the supporting staff at AARI for supporting M. Yamamoto-Kawai at AARI. The completion of this work reflects the efforts of many persons who collected valuable data in very difficult circumstances in the Arctic Ocean and who compiled those data. Measurement of δ18O from Mirai cruises by G. Bower is gratefully acknowledged. Some figures in this paper were illustrated using Ocean Data View software [Schlizer, 2004]. Frontier Observational Research Program for Global Change, JAMSTEC, provided major financial support during the course of this study. Two anonymous reviewers significantly contributed to improve the manuscript and therefore are greatly appreciated.