1. Introduction

During the aftermath of the Russian revolution the renowned oceanographer Harald Sverdrup spent three years aboard the sailing vessel Maud while it was trapped in the ice of the East Siberian Sea. The goal of this epic 1922 to 1924 expedition was to repeat Nansen’s earlier drift on the Fram across the Arctic Ocean with the polar ice pack. The Maud failed to leave the vast shelf seas to the north of Siberia; however, the scientists aboard collected extensive salinity, temperature, and velocity data along the ship’s ice bound drift (Sverdrup 1929). These data constituted the only readily accessible hydrographic information for the area until the military vessels USS Barton and USCGS Northwind surveyed the area within 3 nm of the Siberian coast at the height of the Cold War in 1962 and 1966. In 1995 the R/V Alpha Helix was the first U.S. research vessel to enter the East Siberian Sea. During this summer cruise we deployed 39 surface drifters in ice-free shelf waters while conducting hydrographic surveys in order to explore the influence of the Kolyma River on the shelf circulation. This study constituted the first oceanographic experiment in the East Siberian Sea after the Cold War. It was conducted in a spirit of international cooperation. For this spirit to prevail, however, we must remember the men, women, and children that earlier this century vanished in the Kolyma region (Conquest 1978).

Arctic shelf processes impact ice cover, thermohaline circulation, and biochemical budgets of the Arctic Ocean and adjacent seas (Björk 1989; Carmack 1990; Gawarkiewicz and Chapman 1995; Jones and Anderson 1986). For example, the salinity of the surface waters determines the amount of salt rejected from sea ice as it grows. The dense bottom water that is formed by these brines on shallow shelves maintains the halocline of the central Arctic basins (Aagaard et al. 1981). The supply of the halocline constitutes an important element in the global climate system because it insulates surface waters and thus the ice cover from the vertical heat flux of warmer waters at depth. Melling (1993) argues that shelf processes strongly influence dense water formation as they control the distribution and dispersal of freshwater from riverine discharges prior to winter freezing. More specifically, the removal of freshwater from the shelf decreases vertical stratification, increases surface salinity, and thus enhances the amount of brines produced on the shelf.

The wide continental shelves off Siberia receive large amounts of riverine freshwater (Aagaard and Carmack 1989) and sediments (Milliman and Meade 1983). The river inflow establishes strong vertical density gradients that inhibit vertical exchange. Schlosser et al. (1994) estimate that riverine freshwater resides about 2–5 years on Siberian shelves. Due to its remote and strategically sensitive location, however, few published oceanographic observations are available from the East Siberian Sea. Codispoti and Richards (1968) and Coachman and Rankin (1968) report hydrographic and spotty current observations from a summer survey of the East Siberian shelf and Long Strait, respectively. They find summer surface salinities below 15 psu in bands more than 200 km across the shelf near rivers that drain much of the northern Asian continent of freshwater.

The main objective of this study is to explore the shelf circulation in the East Siberian Sea during the summer season when the Kolyma River discharge constitutes a potentially important buoyancy forcing. Riverine waters always represent a lateral flux of mass and buoyancy that impacts the dynamics on the adjacent continental shelf. Yankovsky and Chapman (1997) and Garvine (1999) present elegant theories on the scales of buoyant plumes and buoyancy driven coastal currents. For typical Kolyma River discharge and geometric scales (discussed below), Yankovsky and Chapman (1997) predict a surface-advected plume that extends about 30–35 km across the shelf while Garvine (1999) predicts an alongshore extent of the coastal current of about 300–500 km. Both theories exclude ambient flows, wind-driven motions, and tidal currents.

The remote location, harsh environment, and largely unexplored physical oceanography of the Siberian shelf seas render comprehensive and focused process studies prohibitively expensive. Drifter studies offer an inexpensive and attractive alternative as they provide spatial and temporal information on the flow field (Davis 1991). The deployment of a large number of drifters, however, is essential to estimate statistically meaningful fields. Individual drifter trajectories may have appealing characteristics; however, they often lack significance in spatially and temporally variable fields (Regier and Stommel 1979). We here report for the first time on the density and surface velocity fields of the East Siberian Sea.

2. Study area and data sources

The East Siberian Sea constitutes a 1000-km long, 400-km wide, and 40-m deep shelf sea in the Arctic Ocean (Fig. 1). The New Siberian and Wrangel Islands at 140°E and 180° long are its boundaries with the Laptev Sea in the west and the Chukchi Sea in the east. Since our study area includes the date line, we subsequently present all maps with longitudes measured positive westward. The Kolyma Delta is then located at 200°W (160°E) long (Fig. 1). Annually averaged discharges from the Kolyma River into the East Siberian Sea vary from just below 2000 m³ s⁻¹ to almost 5000 m³ s⁻¹ (Fig. 2); the climatological mean from 1936 to 1996 is about 3160 m³ s⁻¹ (R. B. Lammers 1998, personal communication). The annually averaged discharge for 1995 was 3080 m³ s⁻¹, close to its climatological mean. The freshwater discharge into the East Siberian Sea is highly seasonal with more than 90% of the annual discharge occurring between June and September (Fig. 2). It increases from a May average of less than 1000 m³ s⁻¹ to a June average of more than 10 000 m³ s⁻¹. Monthly averaged discharge during the winter from November through April is less than 500 m³ s⁻¹. The Lena River to the west, too, contributes to the freshwater budget of the East Siberian Sea. Its peak freshwater discharge can exceed 100 000 m³ s⁻¹ in June (Létolle et al. 1993), which is about half the peak discharge of the Amazon River.

In the summer of 1995, we deployed 39 surface drifters manufactured to our specifications by Horizon Marine Inc., Marion, Massachusetts. Each drifter consists of a cylindrical body and a separate drogue. A 1-m high, sturdy cylinder encloses batteries, electronic circuit, ARGOS radio transmitter, and antenna. A separate 2-m long “holey sock” drogue is centered 2 m below a surface buoyancy element. The drogue is tethered to the cylindrical drifter by a 3-m long line. The drifters differ from the WOCE holey sock drifter because both the drogue and the surface elements are smaller. The ratio R = Σ_D/Σ_S between drag coefficient times the frontal area of the drogue (Σ_D) and the drag coefficients times the frontal areas of the surface elements (Σ_S) is about 26 for our drifter while it is about 40 for the WOCE drifter (Niiler et al. 1995). Hence, our drifters have a slip that is about 50% larger than that of a WOCE drifter, that is, we claim that slip velocities are less than 0.015 m s⁻¹ at 10 m s⁻¹ winds. The drifters were programmed to collect data continuously during the 3-month-long summer season while in the winter, when they tracked ice motions, they collected data 8 hours daily. ARGOS transmits drifter locations almost hourly in our study area. We interpolate the irregular position data with a cubic spline to a time series with a constant time step of 6 h. Tidal currents (not shown) are generally less than 0.05 m s⁻¹; only near Wrangel Island are they larger than 0.1 m s⁻¹ (Kowalik and Proshutinsky 1994). Nevertheless, a Lanczos filter removes tidal and inertial motions from the interpolated time series of position. The filter passes less than 10% and more than 90% of the signal at periods near 30 and 60 h, respectively; the half-power point is at 40 h. We use centered differences to estimate velocity components from the interpolated, filtered, and subsampled drifter position data. Here we present data during the ice-free summer season of 1995 collected from yearday 235 (24 August) to 290 (17 October).

All drifters were deployed in ice-free waters between 5 and 50 km from the coast. Figures 3a and 3b show drifter deployment locations in August and September 1995, respectively. Also shown is the approximate location of the ice edge, which we digitized from weekly ice maps provided by the National Oceanic and Atmospheric Administration/Navy Joint National Ice Center. During the drifter deployments in the East Siberian Sea the ice edge retreated from about 70.5°N to 71.5°N lat within about 2 weeks. This ice distribution indicates an exceptionally light ice year with much melting. The freshwater input due to melting 1 m of ice over the 600 × 200 km² ice-free area of the western East Siberian Sea in 1995 is similar to the freshwater it receives from the Kolyma River (Fig. 2).

We estimate the wind field from sea level pressure forecasts of the U.S. Navy Fleet Numerical Meteorological and Oceanographic Center models to calculate geostrophic winds. Following Overland and Colony (1994), we correct for speed and direction to account for the frictional effects of the atmospheric Ekman layer. Figure 4 shows the time series of wind velocities at locations near the Kolyma River (Fig. 4a), Wrangel Island (Fig. 4b), and Bering Strait (Fig. 4c). The data are low-pass filtered with the filter described above. Winds prior to the experiment (not shown) near the Kolyma River and Long Strait were from the northeast and east and generally weak (<5 m s⁻¹). During the subsequent drifter deployment, winds reached modest strength (<10 m s⁻¹) throughout the region with magnitudes diminishing westward. During the first 3 weeks in September (yearday 242–261) winds originated from the easterly quadrant near the Kolyma River and Long Strait, but in Bering Strait they originated from the northern quadrant. In October (yearday 270) winds strengthened and became more northerly over the Chukchi Sea including Long Strait but were weaker and from an easterly direction near the Kolyma River. A weak upwelling-favorable wind component appears over a portion of our study area.

CTD and bottle casts were taken at 76 locations throughout the study area with a SeaBird 911 instrument. Pre- and postcruise sensor calibrations and comparisons with bottle salinities indicate that the accuracy for salinity, temperature, and pressure are better than 0.01 psu, 0.01°C, and 1 db, respectively. In the following discussion we will use salinity synonymously with density because at low temperatures and in shallow water the contributions of temperature and pressure to density are negligible. Salinity is furthermore an excellent tracer of both ice melt and river waters, however, in order to distinguish between these two freshwater sources, we need another tracer such as stable oxygen isotope ratios (Strain and Tan 1993).

3. Hydrography

During our surveys and drifter deployments in 1995, the coastal waters off Siberia were ice free (Fig. 3). The waters of the East Siberian Sea are both fresh and cold (Figs. 5 and 6). Surface temperatures vary between 0° and 2°C while bottom temperatures are near the freezing point. At depth we find large horizontal salinity gradients but almost no temperature gradients. This contrasts in all respects with the hydrography in the adjacent Chukchi Sea, which receives Pacific waters from the Bering Sea through Bering Strait (Coachman et al. 1975). In the Chukchi Sea surface and bottom waters reach 7°C and 2.5°C, respectively. In contrast to the East Siberian Sea, temperature gradients in the Chukchi Sea are large while salinity gradients are small. The transition occurs at Long Strait near the date line where we find both the coldest (<0°C) and most saline (>32.5 psu) waters. These are remnant waters from the previous winter (Coachman et al. 1975) and were probably formed within the polynya that frequently forms along the south coast of Wrangel Island (Cavalieri and Martin 1994).

On both shelves inshore waters are generally colder, saltier, and thus denser than offshore waters. This trend is particularly clear in Fig. 7, which shows salinity and δ¹⁸O distributions on sections K, A, B, and C (see Fig. 5 for locations). Section K extends alongshore from the Kolyma plume toward Ayon Island while sections A, B, and C extend across the shelf. We deployed drifters along sections A, B, and C. Frontal features are not resolved well by the 20-km station spacing of the survey; however, large horizontal salinity (and thus density) gradients occur throughout the study area. The ice edge is present on sections B and C about 160 km and 110 km from the coast, respectively (Fig. 7). Property gradients are most pronounced in Long Strait (section C) as compared with the more westward sections B and A. Low δ¹⁸O values indicate that the river plume extends far offshore in dilute form. We find fresh (S < 28 psu) and isotopically light (δ¹⁸O > −4‰) waters both about 100 km to the north of Ayon Island on section A (Fig. 7b) and 100 km to the west of Ayon Island on section K (Fig. 7a). Inshore salinities increase from the Kolyma River (section K) eastward toward Long Strait (section C). These reach values of about 30, 32, and 33 psu along sections A, B, and C, respectively (Fig. 7); however, along section K we find Kolyma plume waters inshore where salinity and δ¹⁸O drop below 19 psu and −9.0%. The waters thus become progressively fresher and isotopically lighter toward the Kolyma Delta along section K and offshore along section A.

The property distributions of section B warrant additional comment since they foreshadow the main result of this study: a mean westward flow due to baroclinic forcing from a retreating ice edge. The temperature patterns are decoupled from density patterns and reflect solar heating of the surface waters (Fig. 8). The freshest waters are offshore at the location of the ice edge; however, we also find a shallow wedge of warm and somewhat fresher surface water about 30 km from the coast. A 30-m thick pool of cold and salty waters overlies a bathymetric depression 40 km wide about 50 km from the coast. Dense bottom waters are advected preferentially over this depression from polynyas around Wrangel Island. Drifters deployed at the three inshore stations of section B indicate a cyclonic surface circulation over this topographic depression. Isopycnals offshore of this depression contact both the surface and the bottom. Figure 9a shows the thermal wind shear relative to no flow at the bottom; however, it indicates a baroclinic westward surface circulation of about 0.05 m s⁻¹. Figure 9b compares this estimate with alongshore velocities derived from drifters that pass section B. While the agreement is not perfect, the drifter-derived velocities show, in the mean, the same trend as the thermal wind, that is a 0.05 m s⁻¹ westward surface flow. We discuss more details of the observed velocity fields in section 5.

4. Water masses

In this study we consider three water masses. These are resident shelf waters, freshwater from river discharges, and freshwater from seasonal ice melt. Östlund and Hut (1984), Cooper et al. (1997), Grebmeier et al. (1990), Schlosser et al. (1994), and Macdonald et al. (1995) all use salinity and δ¹⁸O measurements to distinguish between river discharge and ice melt. In order to properly do so, however, one needs to prescribe numerical values for the end members that potentially contribute to the admixture. While it is straightforward to find values for the riverine input that we discuss next, it is more problematic to find an adequate δ¹⁸O value for the ice melt end member.

Östlund and Hut (1984) suggest a value of −21.0 ± 0.7‰ for riverine waters that enter the Arctic Ocean; Grebmeier et al. (1990) consider a value of −22.0 ± 0.3‰ appropriate for the Yukon River, Alaska; and Macdonald et al. (1995) use a value of −20.3‰ and −18.3‰ for the Mackenzie River, Canada, in the summer and winter, respectively. Létolle et al. (1993) cite river values for the Eurasian Arctic that decrease west to east from −12.7‰ for fjords in northern Norway to −23.8‰ for the Indigirka River. It is more difficult and ambiguous to assign a δ¹⁸O value for ice melt. Macdonald et al. (1995) distinguish between river and sea ice because the δ¹⁸O of their ice samples varies from −16.0‰ to −2.0‰. Ideally, we need a third tracer (e.g., barium; Guay and Falkner 1998) to separate sea ice melt from river ice melt. Such samples are not available for this study.

We interpret the two regression curves of Fig. 10 to represent (a) mixing of salty ambient polar shelf water with isotopically light Kolyma River plume waters and (b) mixing of salty ambient polar shelf water with isotopically heavy ice melt that contains a fraction of river ice melt. Surface samples from the western Chukchi Sea have δ¹⁸O versus S regressions very similar to the ice melt samples of the adjacent East Siberian Sea (Table 1). This observation could indicate similar sources and freezing/melting histories (Strain and Tan 1993). Nevertheless, we here distinguish three water masses. These are ice melt (δ¹⁸O = −0‰, S = 6 psu), ambient polar shelf (δ¹⁸O = −1.0‰, S = 33 psu), and, finally, river waters (δ¹⁸O = −22.4‰, S = 0 psu). Assuming a three-point admixture as well as the conservation of mass, salt, and δ¹⁸O (Macdonald et al. 1995), we find that East Siberian shelf waters at any location contain less than 16% ice melt, frequently contain more than 10% river waters, and always contain more than 54% polar shelf waters. Uncertainties in these percentages due to uncertainties in the assigned salinity and δ¹⁸O for the source waters are about 2%–3%.

The spatial distribution of the three water masses reveals two distinct source regions of freshwater. Figure 11a indicates the fraction of riverine waters resident in the surface waters of the eastern East Siberian Sea. Near the Kolyma River delta almost 20% of the water contains riverine freshwater; however, this fraction declines rapidly to less than 10% within about 100 km both to the north and east. The maximum of the ice melt fraction is found near the date line off Wrangel Island (Fig. 11b) where 20% of the surface waters consist of recent ice melt. Wrangel Island is also the location where the permanent polar ice pack always extends farthest south (Fig. 3) and, in some years, never disappears. In contrast, the nearshore waters along the coast off Siberia contain less than 5% of ice melt but more than 90% of ambient shelf waters. These are the cold, salty, and isotopically heavy waters that we find both at depth and in the western Chukchi Sea; however, we here also find them at the surface to the east of Ayon Island. They represent remnant winter waters.

5. Kinematics

c. Lagrangian scales

Lagrangian statistics 〈 · 〉_L result from ensemble-averaging properties of drifters that were deployed in an area of interest. We deployed 10 drifters in an area about 400 × 70 km² (Fig. 12) within about 2 days. Their individual mean velocities vary between 0.05 and 0.12 m s⁻¹ during the 50 days that they drifted freely in the East Siberian Sea. Their mean directions vary between 133° and 164° counterclockwise from true East. The strong directionality of the flow to the northwest contrasts with all known velocity observations from this shelf (Sverdrup 1929; Coachman and Rankin 1968; and U.S. Weather Bureau 1957).

Decomposing the Lagrangian velocities into a mean and fluctuating part for the eastward (u) and northward (υ) velocity components, that is,

utu_Lutυtυ_Lυt

we find that the overall mean velocity (〈u〉_L, 〈υ〉_L) is (−0.066, 0.042) m s⁻¹. The variances about this mean velocity are 〈u′u′〉_L = 0.0218 m² s⁻², 〈υ′υ′〉_L = 0.0210 m² s⁻², and 〈u′υ′〉_L = 0.0004 m² s⁻² (Table 2). The autocovariances 〈u′u′〉_L and 〈υ′υ′〉_L represent rms speeds of about 0.15 m s⁻¹. The fluctuating motions are thus about twice as strong as the mean motions. Additionally, they are spatially homogeneous at the largest scale and we can, if we assume stationarity of the velocity field, apply the theory of Taylor (1921) to study dispersion on this shelf. We will do so below after we discuss temporal and spatial scales of variability.

Assuming that velocity is a Gaussian variable, we find that the statistical uncertainty of the overall mean is δ = 1.96(〈u′u′〉_LT_L/T)^1/2 ≈ 0.028 m s⁻¹ at the 95% level of confidence. The ratio T/T_L estimates the statistical degrees of freedom for a record length T and a Lagrangian autocorrelation scale T_L. Following Poulain and Niiler (1989), we define T_L as the integral of the normalized versions of the Lagrangian autocorrelation functions R_uu(τ)/R_uu(τ = 0) and R_vv(τ)/R_vv(τ = 0) to the first zero crossing; τ represents the lag time. The functions R_uu(τ) and R_vv (τ) are defined as

View Expanded

Integrating the normalized autocorrelation functions to their first zero crossing at 4 and 5.5 days (Fig. 17), we find that meridional and zonal integral timescales have first maxima near 3.5 and 2.5 days, respectively. This constitutes an upper limit of T_L. The timescales relate to an integral space scale L_L through

L_Luu_L^1/2T_L

which in the East Siberian Sea is about 43 and 35 km in the meridional and zonal direction, respectively. Decorrelation time and space scales thus are about 3 days and 40 km. We use these scales to estimate degrees of freedom within our dataset.

The integral of R_uu(τ) and R_vv(τ) with respect to τ represents a single particle diffusivity if the flow is both homogeneous and stationary (Poulain and Niiler 1989). Figures 17b and 17c depict the diffusivities K_xx and K_yy as a function of lag time, respectively. Using Taylor’s (1921) theory of diffusion to interpret our data, we distinguish between initial dispersion for times τ ≪ T_L where the diffusivity varies linearly with lag time τ and a random walk regime for times τ ≫ T_L. The initial dispersion lasts for about a day while for lag times longer than 8 days the diffusivities reach values of about 3.5 × 10³ and 2.5 × 10³ m² s⁻¹ for the zonal and meridional component, respectively. These values are large, especially in the across-shelf (meridional) direction and resemble those of midlatitude open ocean studies. For example, Poulain and Niiler (1989) report diffusivities of about 3–4 (×10³ m² s⁻¹) for the California Current System, Krauss and Böning (1987) find diffusivities in the range of 2–8 (×10³ m² s⁻¹) for the North Atlantic, and Middleton and Garrett (1986) use icebergs off Labrador to estimate an isotropic diffusivity of about 1.1 × 10³ m² s⁻¹. In contrast, Davis (1985) finds 3.1 × 10³ and 0.08 × 10³ m² s⁻¹ for an alongshelf and across-shelf diffusivity within about 10 km off California. Münchow and Garvine (1993) report diffusivities of about 2 × 10³ and 0.5 × 10³ m² s⁻¹ for the Delaware coast in the along- and across-shelf directions, respectively. The dispersive characteristics of drifters deployed in the strongly stratified East Siberian shelf sea thus resemble open-ocean drifters at midlatitudes more than they resemble drifters released into the coastal ocean.

6. Discussion

We reported velocity and hydrographic observations from the East Siberian Sea to test the hypothesis that the buoyant discharge from the Kolyma River results in an eastward-bound buoyancy-driven coastal current. Using predictions of Yankovsky and Chapman (1997), we expected a surface-advected plume to extend 30–35 km across the gently sloping shelf. This estimate assumes Rossby and Burger numbers of about 0.1 and 0.7, respectively, due to a typical summer discharge of about 10 000 m³ s⁻¹. Using similar nondimensional and geometric scales, we also expected the coastal current to extend 300–500 km along the shelf (Garvine 1999). The fate of 39 surface drifters deployed into ice-free waters of the Chukchi and East Siberian sea, however, indicated that no such eastward flow was present in the summer of 1995. Instead our drifters described a flow in the opposite direction; that is, the surface circulation in 1995 was westward from the Chukchi into the East Siberian Sea. The 50-day mean flow reached 0.1 m s⁻¹ and was statistically significant at the 95% level of confidence. We found little river water on the East Siberian shelf downstream of the Kolyma River. Both velocity and density fields suggested that buoyant river waters did not reach Ayon Island 100 km to the east of the Kolyma River delta. Only in the immediate vicinity of the Kolyma Delta did we find clear evidence of predominantly riverine waters. The absence of an eastward flow requires a reevaluation of the conceptual view of the surface circulation of the East Siberian Sea.

We find large Lagrangian decorrelation scales of about 3 days and 40 km in time and space. The internal deformation radius inshore is 13 km and thus is a factor 3 smaller than the eddy length or spatial decorrelation scale. The square of the ratio between the horizontal scale of the motion L and the internal deformation radius L_D defines a Burger number S ≡ (L_D/L)² that scales baroclinic vortex tube stretching in the vorticity equation (Pedlosky 1987). In our study area the Burger number is 0.1 inshore and baroclinic vortex tube stretching is weak relative to the diffusion of relative vorticity in the potential vorticity equation. The flow field is linear, though, as the Rossby number ε = U/fL ∼ 0.005 ≪ 1 is small both inshore and offshore; we here use a velocity scale U ∼ 0.2 m s⁻¹, a Coriolis parameter f = l.4 × 10⁻⁴ s⁻¹, and the eddy length scale L ∼ 40 km. Farther offshore the flow is influenced by the retreat of the ice edge. Here the eddy scale is closer to the local deformation radius (24 km) and the Burger number is about 0.4. While vortex tube stretching is thus still smaller than the diffusion of relative vorticity, the two terms in the vorticity equation are now of the same order of magnitude. The dynamics is not quasigeostrophic, vertical excursions of isopycnals frequently scale with the total water depth.

The northwestward surface circulation is internally consistent with a thermal wind shear associated with the observed sloping isopycnals. The isopycnals outcrop over a scale of about 100 km, that is, the dynamics is not frontal. We interpret our density and velocity observations as the result of an ice edge in retreat. During our study period the ice edge moved northward more than 100 km in less than 2 weeks (0.08 m s⁻¹). In retreat, it leaves melt water behind, which constitutes a line source of buoyancy.

The prevailing westward drift of the surface waters in the East Siberian Sea in 1995 could also be caused by large-scale barotropic and/or baroclinic pressure gradients. Overland and Roach (1987) demonstrate with a barotropic model that a geostrophically controlled northward transport through Bering Strait of the observed magnitude can be driven by a sea level height difference from the Pacific to the Arctic Ocean. Coachman and Rankin (1968) suggest that the highly variable subtidal circulation through Long Strait is caused by different pressures in the Chukchi and East Siberian seas. We speculate that similar pressure gradients could also be imposed from the Laptev Sea in the west through Laptev Strait. The Bering, Chukchi, East Siberian, and Laptev shelf seas each respond to different wind and buoyancy forcing. It is thus conceivable that sloping sea levels result in enhanced flows at the “choke points” such as Bering Strait, Long Strait, and Laptev Strait. The dynamical control of transports through these choke points could influence the circulation in both the receiving and the discharging shelf sea. Numerical models that resolve these straits and the internal radius of deformation are needed to investigate the dynamics of the Siberian shelf seas.

Another possible explanation of the interannual variability of the Siberian shelf circulation, we speculate, may relate to decadal-scale climate oscillations. Proshutinsky and Johnson (1997) distinguish two oceanographic circulation regimes that relate to the atmospheric surface pressure distribution in the central Arctic. Following earlier Russian studies (Sokolev 1962), they describe “cyclonic” and “anticyclonic” regimes that represent oceanic conditions during periods of different interannual atmospheric pressure anomalies over the central Arctic. Our 1995 velocity and salinity observations appear internally consistent with expectation of the “anticyclonic” regime of enhanced high pressure systems over the central Arctic. During such years, according to the hypothesis, sea ice cover is much reduced in the East Siberian Sea. Wind-induced motions are postulated to enhance the flushing of ice from this shelf (Sokolov 1962; Proshutinsky and Johnson 1997). In contrast with this theory, however, Maslanik et al. (1996) report reduced ice cover during periods of enhanced cyclone activity off Siberia this decade in what would be the “cyclonic” regime. The cause and effect of this interannual variability is thus both unclear and speculative. Overland et al. (1999) argue that decadal-scale oscillations in the Arctic are related to similar oscillations in both the North Pacific (Mantua et al. 1997) and the North Atlantic (Dickson et al. 1997). The linkages between atmospheric forcing, ice cover, buoyancy forcing, shelf circulation, and its impact on the maintenance of the halocline, though tantalizing, are still unsettled.