We present high-resolution simulations and observational data as evidence of a fast current flowing along the shelf break of the Siberian and Alaskan shelves in the Arctic Ocean. Thus far, the Arctic Circumpolar Boundary Current (ACBC) has been seen as comprising two branches: the Fram Strait and Barents Sea Branches (FSB and BSB, respectively). Here we describe a new third branch, the Arctic Shelf Break Branch (ASBB). We show that the forcing mechanism for the ASBB is a combination of buoyancy loss and non-local wind, creating high pressure upstream in the Barents Sea. The potential vorticity influx through the St. Anna Trough dictates the cyclonic direction of flow of the ASBB, which is the most energetic large-scale circulation structure in the Arctic Ocean. It plays a substantial role in transporting Arctic halocline waters and exhibits a robust seasonal cycle with a summer minimum and winter maximum. The simulations show the continuity of the FSB all the way around the Arctic shelves and the uninterrupted ASBB between the St. Anna Trough and the western Fram Strait. The BSB flows continuously along the Siberian shelf as far as the Chukchi Plateau, where it partly diverges from the continental slope into the ocean interior. The Alaskan Shelf break Current (ASC) is the analog of the ASBB in the Canadian Arctic. The ASC is forced by the local winds and high upstream pressure in Bering Strait, caused by the drop in sea surface height between the Pacific and Arctic Oceans.
 The upper Arctic Ocean is strongly stratified: a low salinity Arctic mixed layer overlies a shallow Arctic halocline, in which density and salinity increase sharply with depth [e.g., Rudels et al., 2004]. Beneath the halocline resides a warm, saline, almost 800-m thick layer of Atlantic Water (AW). The AW layer, first explored by Nansen , contains sufficient heat to melt several meters of Arctic sea ice within a few years, if this heat is brought to the ocean surface [Rudels et al., 1996, 2004; Steele and Boyd, 1998; Turner, 2010]. Wind and atmospheric thermodynamic forcing were the main causes for the record Arctic sea ice retreat in summer 2007 [e.g., Perovich et al., 2008; Ogi et al., 2010]. However, AW heat was also important for preconditioning Arctic sea ice by making it thinner over several preceding decades and thus contributing to the extreme retreat [Polyakov et al., 2010]. An alternative view is that the presence of the strongly stratified halocline in the Arctic Ocean suppresses vertical turbulent mixing, impeding heat and salt exchange between the Atlantic layer and the surface [e.g., Aagaard et al., 1981; Lenn et al., 2009; Rainville and Winsor, 2008; Rudels et al., 1996] and, consequently, the increase of the Atlantic layer heat does not contribute significantly to the sea ice melting [Fer, 2009]. This stabilizing effect of the halocline layer varies geographically. Recent microstructure turbulence measurements have demonstrated that the upward oceanic heat flux across the halocline into the sea ice may vary from 1 to 10 W m−2 in the central Arctic Ocean and from 20 to 100 W m−2 at the continental margins of the Eurasian Basin; a difference of almost two orders of magnitude [e.g., Fer, 2009; Sirevaag and Fer, 2009; Lenn et al., 2011]. It is suggested this is due to: enhanced vertical mixing by winds or tides; upwelling of the Atlantic layer or eddy induced vertical circulation along the deep ocean margins [e.g., Polyakov et al., 2010]. Although the interactions between Atlantic layer and halocline are still debated, it is clear that these are important for the sea ice, heat balance and ocean circulation in the Arctic Ocean.
Aagaard et al. , Jones and Anderson , Rudels et al. [1996, 2004] and Steele and Boyd , followed by Kikuchi et al. , suggested a number of mechanisms to produce halocline waters. In the Eurasian Arctic (Nansen and Amundsen Basins) the ‘advective-convective’ mechanism through deep winter convection, accompanied by isopycnal advection of shelf waters between surface layer and AW, creates a cold halocline [Steele and Boyd, 1998]. Most of the Eurasian halocline forms in the northern Barents Sea and on the northern Barents Sea shelf [Rudels et al., 2004] (Figure 1). Rudels et al. [1999a] demonstrated that the thermohaline “staircase-like” layered structures, evidence of double-diffusive convection, are typical of the central Eurasian Arctic away from the shelf boundary current. They speculated that the layers were formed north of the Kara Sea along the front marking the confluence zone of Fram Strait Atlantic water and Barents Seawaters and subsequently advected into the basin interior. Lenn et al.  found double-diffusive layers in the shelf boundary current in the Laptev Sea but concluded that vertical double-diffusive heat fluxes are too small to impact the boundary current and halocline. In the Canadian Basin the halocline structure is more complex, reflecting interaction of the Atlantic and Pacific inflows and Siberian Shelf waters through thermohaline intrusions and diapycnal mixing [e.g., McLaughlin et al., 2002; Itoh et al., 2007; Woodgate et al., 2005, 2007] and input of saline waters due to the sea ice formation in the eastern Chukchi Sea [Shimada et al., 2005].
 Direct velocity measurements in the Arctic Ocean are sparse, thus much of what is known about the Arctic circulation has been inferred from hydrographic measurements and from a few current meter moorings. These data suggest a weak, eddy-rich interior circulation and a topographically guided system of boundary currents called collectively the Arctic Circumpolar Boundary Current (ACBC), flowing cyclonically along the margins of the ocean basins [e.g., Aagaard, 1989; Rudels et al., 1994, 1999b]. We propose this name (ACBC) for the system to avoid confusion generated by the use of the name “Arctic Circumpolar Current,” whose abbreviation (ACC) has other uses (e.g., Alaskan Coastal Current discussed in this paper, or Antarctic Circumpolar Current in the south).
 The cyclonic pathway of the ACBC is documented by current meter measurements along the Siberian shelf as far as the Lomonosov Ridge [e.g., Woodgate et al., 2001; Dmitrenko et al., 2008; Ivanov et al., 2009]. The North Atlantic inflow enters the Arctic Ocean by two routes: (1) through Fram Strait, when it is subsequently known as the Fram Strait Branch (FSB) of the ACBC and (2) through the Barents Sea, when it is subsequently called the Barents Sea Branch (BSB) of the ACBC (Figure 1). Both branches are assumed to flow cyclonically along the Siberian continental slope, forming the ACBC [Rudels et al., 1994, 2000]. West of St. Anna Trough (Figure 1) the ACBC carries warm AW from Fram Strait [Ivanov et al., 2009]. To the east, the current carries two water masses: AW, above the lower continental slope where the depth exceeds 1800 m, and cold Barents Seawater above the upper slope where the depth is shallower than 1800 m [Schauer et al., 2002]. The majority of the Barents Seawater arrives into the Nansen Basin of the Arctic Ocean through St. Anna Trough [Rudels et al., 2004].
 The pathway of the ACBC to the east of the Lomonosov Ridge is less well-known [Woodgate et al., 2007]. To reach the Canada Basin, the ACBC negotiates the complex topography of the Mendeleev Ridge and the northern slopes of the Chukchi Sea (Figure 1). Part of the current follows the continental slope, with a part of it diverted into the Arctic interior along the Mendeleev Ridge and the western slope of the Chukchi Plateau [Woodgate et al., 2007]. Limited current meter measurements carried out in the Canada Basin and on the Canadian Shelf [Newton and Sotirin, 1997] suggest uninterrupted eastward boundary flow in these regions and thus continuity of the ACBC throughout the Arctic Ocean [Rudels et al., 1999b].
Jones  summarized that “the circulation of halocline waters is not well delineated. A reasonable hypothesis is that the flow resembles that of the Atlantic Layer, with additions of Pacific origin water from the Chukchi shelf.” This agrees with schematics by Rudels et al.  who analyzed the hydrographic measurements and proposed an eastward flow of the halocline water along the continental shelves and spread of this water into the Arctic Ocean interior along the submarine ridges (Figure 1). More recent hydrographic observations in the Canada Basin and north of the Chukchi Sea demonstrated an interleaving structure of the Pacific and Atlantic waters [e.g., McLaughlin et al., 2004, 2009; Shimada et al., 2005; Woodgate et al., 2005, 2007]. The expected dynamics from these data suggests that the Pacific halocline water spreads northward into the Canada Basin and eastward along the Alaskan continental slope. However, current measurements from a high-resolution mooring array north of the Alaskan coast have demonstrated that the strength of the eastward flow of the Pacific halocline might be overestimated [Spall et al., 2008; Nikolopoulos et al., 2009], with the main part of the Pacific inflow entering the Arctic Ocean interior through the western Chukchi Sea or further west, during the reversal of the Siberian Coastal Current [Abrahamsen et al., 2009].
 Summarizing, we conclude that, although advances have been made in regional observations of the ACBC, the basin-wide structure of the current is still unclear. The present study examines the flow along the Arctic shelves using an eddy-permitting/eddy-resolving simulation and recently obtained closely spaced observations. It addresses the following questions. (1) To what extent can the ACBC be regarded as a continuous flow around the whole of the periphery of the Arctic Ocean? (2) What are the dynamical causes of the flow along the Arctic continental slope? (3) What is the water mass composition of the flow? And finally: (4) What are the implications of the flow for the shelf-basin exchange and for the halocline layer in the Arctic?
 The paper presents the most complete observationally supported modeling results to date to clarify the structure of the ACBC. We will demonstrate that the ACBC consists not of two but of three cores: the well-known Fram Strait and Barents Sea Branches, plus a third, newly identified feature, the Arctic Shelf Break Branch, hereafter ASBB (Figure 1). ASBB transports halocline waters from the Barents and Kara Seas and has a velocity core located next to the shelf break.
 We will show that ACBC is continuous throughout the Arctic Ocean when all its constituents are considered (ASBB, FSB and BSB). Of these, the FSB is a continuous cyclonic current throughout the Arctic Ocean. It enters, flows eastward and leaves via the Fram Strait and is continuous throughout its circuit around the Arctic. The cyclonic ASBB takes a shorter path. It enters the Arctic via St Anna Trough, leaves via the Fram Strait and is continuous between these locations. The BSB is also continuous between St. Anna Trough and Fram Strait; while flowing cyclonically around the rim of the Arctic Ocean it entrains waters cascading from the Arctic shelves and makes detours around seabed topography. We discuss the ASBB in detail and some relevant features of the FSB, but detailed studies of the BSB (a more complex current) will be the subject of future studies.
 The paper is arranged as follows. Section 2 describes the methods, including a novel theoretical formulation of Montgomery function analysis. Section 3 presents the simulated flow along the Arctic shelves, investigates how well the models represent the real flow focusing on Fram Strait, the Siberian and Alaskan shelves, where observations are available, and discusses in brief the flow along the Canadian Shelf, where they are not. Section 4 addresses mechanisms to generate the branches of the ACBC, analyses effects of the slope flow on the halocline and Atlantic waters and discusses implications for modeling and observational strategies in the Arctic. Section 5 summarizes the findings.
 To analyze the ACBC and its contribution to renewal of the halocline waters in the Eurasian Basin, we use the high-resolution global Ocean Circulation and Climate Advanced Model (OCCAM) and observational data from the long-term moorings in Fram Strait (Alfred Wegener Institute and Norwegian Polar Institute) [e.g., Schauer et al., 2004, 2008], long-term current meter data and summer hydrographic surveys on the Siberian continental slope obtained via the Nansen and Amundsen Basins Observational System (NABOS) [Polyakov et al., 2007], together with current-meter and hydrographic measurements from an array in the Beaufort Sea, north of the Alaskan coast [Nikolopoulos et al., 2009]. We use potential vorticity (PV) and the Montgomery function, presented in this section, to analyze the forcing mechanisms of the ACBC.
 OCCAM is a primitive equation, z-level global Ocean General Circulation Model (OGCM) model with a free surface [Killworth et al., 1991]. The model has been described in detail elsewhere [e.g., Marsh et al., 2009; Aksenov et al., 2010a]; here we only highlight model features relevant to the present study. The ocean model has global coverage with two separate grids joined along the Atlantic equator and through the Bering Strait via a linearized channel model, resulting in a nearly uniform global grid with ca. 8-km horizontal resolution. The model has 66 vertical levels, including 27 levels in the upper 400 m, of which 14 are in the upper 100 m. The high vertical resolution allows the representation of the key vertical gradients, including those of Arctic halocline and AW. The ca. 5-m resolution near the surface is sufficient for the KPP mixed-layer model to give a realistic evolution of the mixed-layer depth [Large et al., 1994]. The model is eddy-resolving globally, including the central Arctic Ocean, where the local first baroclinic Rossby radius is ca. 15 km [Nurser, 2009], but is eddy-permitting at best near the Arctic continental slope and on the outer shelf.
 The sea ice model comprises sea ice dynamics with elastic-viscous-plastic rheology [e.g., Hunke and Dukowicz, 1997] and Semtner thermodynamics [Semtner, 1976] configured with two layers for sea ice and one for snow. The sea ice model is coupled to the ocean on each baroclinic time step through the quadratic drag law, allowing forcing of the ice covered ocean with high frequency winds [Aksenov et al., 2010a]. The coupling conserves volume and takes account of the pressure loading of the sea ice on the ocean [Griffies et al., 2004; Campin et al., 2008].
 To compensate for the absence of continental run-off and to prevent surface salinity drift, the simulated surface salinity is relaxed toward monthly climatological values from Boyer et al. . This is done by interpolating monthly climatological sea surface salinity fields at the end of each baroclinic time step and converting the difference between model and interpolated climatological sea surface salinity into an equivalent salinity flux computed using a piston velocity and calculating sea surface height correction. The relaxation procedure preserves the ocean salt content [Webb et al., 1998]. No salinity correction is applied below the top model level. The value of the piston velocity is such that relaxation is equivalent to adding or removing fresh water on a time scale of 40 days for the top 20 m of the ocean. For the period 1989–2006, the integral relaxation term for the Arctic Ocean was 104 ± 16 mSv (1 mSv = 103 m3 s−1), in the range of the corresponding estimate for continental runoff of 99–135 mSv [Barry and Serreze, 2000]. It should be noted that the climatological salinity fields [Boyer et al., 1998] used for the surface relaxation of the model exhibit an erroneous radial pattern near the North Pole due to sparseness of the available data in the area and interpolation problems [Steele et al., 2001]. However, the error is not detrimental to the study, as the main distortion of the climatology occurs in the central Eurasian Basin and does not have an impact on the boundary current. Besides, the strong stratification at the surface prevents surface relaxation having a significant effect on the circulation at the 100–600 m depth range analyzed in the present study.
 The model was forced with the U.S. National Centers for Environmental Prediction (NCEP) 6-hourly reanalysis [Large et al., 1997], extended to 1985–2006. Details of the forcing technique are given by Marsh et al.  and Aksenov et al. [2010a]. The simulations were initialized from rest with an initial state derived from merged global and Arctic ocean climatologies (World Ocean Atlas, 2005); the initial sea ice and snow cover was taken from Romanov .
 It was impossible for the model to reach a steady state during the 22-yearlong run (1985–2006). However, because realistic initial fields were used, most features of the upper ocean circulation spun-up during the first years of integration: global mean kinetic energy, transports through the Barents Sea, Fram and Bering Straits settled after four years, and heat and salt content of the ocean stayed fairly constant after the first three years of integration. The barotropic component of the ocean circulation spun up within days. Sea ice reached a quasi-equilibrium state after the first four years. We used the period 1989–2006 for the analysis.
 Applying the technique developed to study the volumetrics of potential temperature (θ) classes by Walin , and of density classes by Speer and Tziperman , further generalized to potential temperature-salinity (θ–S) water classes by Speer  and Large and Nurser , we calculate formation rates of halocline water. We link the formation rate of water within a given θ–S class to the surface fluxes of heat and freshwater acting on ocean surface outcrop of the specified water class. The surface heat includes radiative, latent and sensible atmospheric heat fluxes and also heat of sea ice formation and heat associated with meltwater runoff due to sea ice and snow ablation. The freshwater forcing takes into account precipitation, evaporation, sea ice and snowmelt and the restoring term. By comparing halocline formation rates inside a closed domain (net export out plus the rate of increase in volume within the domain) to the surface-flux driven divergence of the volume fluxes across the θ–S surfaces we calculate the depletion of the halocline due to mixing and diffusion.
 The Ekman pumping is calculated from the curl of the combined interfacial stress acting on upper ocean surface. The interfacial stress in its turn is computed as the sum of the ice-ocean stress, weighted by the ice fraction in each model grid cell, and the wind stress, weighted by the open water fraction in the cell. The method is consistent with the forcing of the model.
 Three sets of observations (summarized in Tables 1 and 2) are analyzed in the paper: (1) current-meter data from the long-term moored array at 78°50′N in Fram Strait and temperature measurements from CTD (conductivity-temperature-depth) sensors at the moorings, (2) current meter measurements from the long-term moorings at ∼78°N, 126°E together with CTD measurements at cross-slope synoptic sections carried out in summer-autumn in the Laptev Sea along ∼126°E, and (3) current-meter and hydrographic data from a high resolution current meter array in the Beaufort Sea, north of the Alaskan coast at ∼71°N, 152°W, hereafter the Alaskan array (Figure 1).
Table 1. Inventory of CTD Measurements in Fram Strait and the Arctic Ocean Used in the Studya
Here CTD is conductivity-temperature-depth probe.
Siberian Shelf in the Nansen and Amundsen Basins
Table 2. Inventory of Current Meter Moorings in Fram Strait and the Arctic Ocean Used in the Studya
Current Meters (Depth (m))
Here MMP is McLane moored profiler, RCM is rotating current meter; CMP is coastal moored profiler, ADCP is upward-facing acoustic Doppler current profiler, and ACM is profiling acoustic current meter.
RCM, ADCP, ACM (50–2500)
RCM, ADCP (50–2500)
Siberian Shelf in the Nansen and Amundsen Basins
ACM on MMP (165–2600)
RCM (68, 210, 992)
ACM on MMP (100–2400)
ACM on MMP (100–880)
ADCP, ACM on CMP and MMP (45–600)
 The array of 14–16 moorings in the northern Fram Strait has been maintained since 1997 as a joint effort of AWI and NPI. Moorings extend from the shelf west of Spitsbergen (Svalbard) to the East Greenland shelf, covering ca. 300 km wide section with spatial resolution from ca. 10 km on the shelf slope to ca. 30 km in the deep area. The array is instrumented with current meters (Aanderaa RCM8 and RCM11, RDI ADCPs, in early years also FSI ACMs; here RCM – Rotating current meter, ADCP – Upward-facing acoustic Doppler current profiler, ACM – profiling acoustic current meter) and CTD sensors from Seabird Inc. (SBE16, SBE37), distributed at five nominal levels: in the subsurface layer (ca. 50 m), in the AW layer (ca. 250 m), at the lower boundary of the AW layer (ca. 750 m, since 2002), in the deep layer (ca. 1500 m) and above the bottom. Until 2002, the Fram Strait array consisted of 14 moorings and its western part (west of 0°) was shifted to 79°N. Since 2002 the array has been leveled along 78°50′N and augmented with two additional moorings in the deepest, central part of the strait and an additional level of instruments at 750 m. Data are recorded at 1 or 2 h intervals, de-tided and averaged on to required time steps (daily or monthly). For long-term means shown on Figure 2 monthly data were used, averaged separately for 1997–2002 and 2002–2006 to avoid bias from a different distribution of moorings and instruments in the two periods (on Figure 2 positions of individual moorings are indicated with triangles, positions of instruments with circles). Observation and data processing details are given by Schauer et al. [2004, 2008].
 Hydrographic measurements in the Nansen Basin and Laptev Sea were carried out using a shipboard SeaBird (SBE19) CTD (Table 1). Individual temperature, conductivity and pressure readings were accurate to 0.005°C, 0.0005 S/m and 0.002% of full scale range respectively (http://nabos.iarc.uaf.edu/index.php). The distance between CTD stations was about 10 nautical miles (1 nm = 1.8525 km) in the deep basin, reduced to 3–5 nm on the steep continental slope. During transects in the Laptev Sea in 2007 and 2009, detailed surveys of the upper continental slope with distance between stations ca.1 nm were carried out. These data are of specific interest as they resolve the Rossby radius in the area. Vertical resolution of all CTD measurements used in this study is 1 m. Ocean currents were measured by an ACM at the autonomous mooring site M1 in the Laptev Sea (Table 2). The ACM was mounted on a McLane moored profiler (MMP), which profiled vertically along the mooring line at a speed of 0.25 m s−1 and a sampling period of 0.5 s once a day (http://www.mclanelabs.com/mmp.html). Continuous ACM measurement record used in this study was for September 2003–2004 and covered a depth range of 100–1000 m. The ACM current velocity precision and resolution were ±3% of reading and 2 × 10−3 m s−1, with compass accuracy of ±2%. Vertical resolution of these current meter measurements is 2 m.
 Our discussion also makes use of hydrographic sections and current meter data from the long-term moorings in the Nansen Basin at ∼81°N, 31°E and at ∼81°N, 104°E (Table 2) [Ivanov et al., 2009] and published results from current meter measurements in the Lincoln Sea at ∼84°N, 63°W [Newton and Sotirin, 1997].
2.3. Potential Vorticity and Montgomery Function Analysis
 We employ the Montgomery function M to visualize and analyze the modeled Arctic Ocean circulation because M is a stream function for geostrophic flow on surfaces of constant density anomaly. Geostrophic flow can be calculated from the lateral gradient of M on a surface of constant density anomaly the same way as it can be found from the lateral gradient of pressure on a surface with constant depth. For steady flow at small Rossby numbers, the Montgomery function is also a stream function for layer-integrated PV flux, so knowledge of it helps us to diagnose the PV fluxes. Although M is not an exact stream function for flow on surfaces of constant potential density, we can define surfaces of constant density anomaly that lie close to surfaces of constant potential density. This requires some theoretical development, described below.
2.3.1. The Boussinesq Montgomery Function
 Surfaces of constant steric (volume) anomaly are defined by constant values of
where θ and S are potential temperature and salinity, with constant, ‘reference’ values θ0 and S0, and p is ‘gauge’ pressure, relative to mean atmospheric pressure pao, 101325 Pa. Surfaces of constant steric anomaly with the standard choices θ0 = 0°C and S0 = 35 psu do not always coincide closely with neutral surfaces or surfaces of constant potential density. However, where potential temperature and salinity vary little on the surface, choosing S0 and θ0 to be the median values of S and θ gives a surface of constant steric anomaly that is reasonably close to a neutral surface [McDougall and Klocker, 2010].
where p″ is pressure, η is the sea surface height anomaly, g is gravity and pa is the atmospheric pressure anomaly. Lateral gradients of M on surfaces of constant steric anomaly are then proportional to horizontal pressure gradients at constant height, so geostrophic flow ug obeys
where k is the unit upward vector, ∇z denotes the lateral gradient at constant geopotential height z, and ∇δ the lateral gradient at constant steric anomaly.
 For simplicity, the OCCAM model (as do most OGCMs) uses the Boussinesq approximation, where in situ ρ is approximated by a constant ρ0 in (3). The steric anomaly in (3) is replaced by pseudo-potential density rB, where
Here p*(z) is the reference pressure field in hydrostatic balance with ρ*(z) = ρ(θ0, S0, p*(z)), pa = pa0, η = 0 and σ0(θ0, S0) = ρ(θ0, S0, 0) is potential density. This rB is the Boussinesq approximation to r = δ−1 + σ0(θ0, S0). We shall refer to surfaces of constant rB as ‘pseudo-isopycnals’. The consistent Boussinesq Montgomery function that ensures
(here the perturbation pressure p′ = p − p*(z), and is the lateral gradient at constant rB) is [cf. Marshall et al., 2001, equation (9)],
2.3.2. Potential Vorticity Fluxes and Circulation Integrals
 Except for the exchange through Fram Strait, the Arctic upper intermediate depth (300–1000 m) flows are generally confined within the Arctic basin. Such confined flows are usefully diagnosed with the flux form of the PV equation. This takes the form [cf. Haynes and McIntyre, 1987, 1990]
Here the PV is expressed as
where absolute vorticity q = Ω + ∇ × u (with Ω the planetary vorticity), u is velocity and J is the PV flux. For consistency with the Boussinesq equation [e.g., Marshall, 2000] Haynes and McIntyre's in situ density ρ has been replaced by a constant density ρ0 in (7a) and (7b), and also in the PV flux, giving [Marshall et al., 2001, equation (11)]:
Here F is the frictional and body force per unit volume, while
is the non-advective forcing of rB. Here w is the vertical velocity, and D/Dt denotes the material derivative. The key property of J is that it does not cross surfaces of constant rB (pseudo-isopycnals)—Haynes and McIntyre's  impermeability theorem.
 Therefore, when equation (7a) is integrated between neighboring isopycnals rB and rB + dr and over some area A, the fluxes across the isopycnals disappear, leaving only the fluxes of J along the isopycnals out through the edges of A, ∂A. These simplify to give a generalized version of Kelvin's circulation integral
where = q · rB/∣∇rB∣ and = B/∣∇rB∣ are the absolute vorticity component and relative velocity across the rB surface, and n is the unit outward normal along the rB surface. In fact, equation (8a) simply states the circulation on a closed curve on an isopycnal changes only as a result of Coriolis force from fluid crossing the curve and frictional and body forces along the curve. We apply this circulation integral to areas bounded by grounding lines, outcrops, and straits for various density surfaces. The sections across straits are generally fixed, but surface outcrops and grounding lines may evolve with time.
 In the steady state the PV advection term and buoyancy forcing terms of (8a) reduce to the Coriolis force along the line element dl,
where the acceleration (Bernoulli) potential π = u2/2 + MB, it follows that the PV influx across the line element dl,
Since for most oceanographic flows the Rossby number is small, π ≈ MB, and therefore the change in Montgomery function gives the PV influx.
 In the following section we analyze the output from the model to establish the structure of the ACBC and study its variability. We use a representation of the oceanic flow on pseudo-ispoycnals to reduce the vector velocity field to the scalar Montgomery function field, and to relate this to the influx of PV. This allows us to elucidate the mechanisms which drive the current. But beforehand, to gain confidence in the model results, we compare the model results to the available measurements in the study region.
3.1. Model Validation
 We compared simulated and observed oceanic currents and θ-S fields on the key sections for this study: Fram Strait, at the section east of Svalbard at ca.30°E [Ivanov et al., 2009], at the section ca. 104°E at the Severnaya Zemlya Polygon [e.g., Walsh et al., 2007], Laptev Sea Shelf at ca.126°E (Table 2) and north of the Alaskan coast at ca.152°W [Nikolopoulos et al., 2009] (Figure 1). We also compared model results with current meter measurements in the Lincoln Sea at ∼84°N, 63°W [Newton and Sotirin, 1997].
3.1.1. Eurasian Arctic
Figure 2 depicts the cross-section velocity at the mooring array along ca. 79°N in Fram Strait in 1997–2006 [Schauer et al., 2008] along with the virtual section from the model (the section denoted as FS in Figure 1). The bi-directional flow in the strait, comprising the northward West Spitsbergen Current (WSC) and southward East Greenland Current (EGC), and main topographic recirculation structures are captured well by the model, although the structure on the eastern side of the strait is more barotropic in observations than in the model. The agreement in θ of the Atlantic inflow and outflow of the Polar water (PLW) is also good. The model simulates the warm AW core on the Spitsbergen side of the strait and the cold PLW core on the Greenland side, although in the model the warm AW core spreads more westward toward the deeper lower shelf slope than in the observations; also the lower boundary of the AW (0°C isotherm) is deeper in the model than from the observations. A difference between the model and observations is the position of the Arctic Front between shoreward boundary of AW in the WSC and cold waters on the Spitsbergen shelf which originate from Sørkapp Current. In the model this front is shifted westward. We cannot see its position at the mooring section, which does not go that far to the shelf, but geostrophic currents inferred from CTD sections and also velocity measurements from ADCP show that the front is definitely east of the easternmost mooring.
 The detailed comparison of the simulated and observed circulation and oceanic exchanges in Fram Strait, in the Barents Sea and Canadian Archipelago are given by Aksenov et al. [2010a, 2010b].
 Downstream of Fram Strait, the AW flow turns eastward and flows north of Svalbard. Comparison of the model velocity with the current meter measurements September 2004 to December 2005 at the section at ca.30°E (section is denoted as SVA in Figure 1, model fields are depicted in Figure 3) shows the close agreement between observed and simulated flow with the maximum velocity at ca. 220 m depth of 0.17 m s−1 and 0.14 m s−1 from current meter observations and model respectively [Ivanov et al., 2009]. θ-S fields are also in good agreement: both the model and the observational data show the warm (up to 4.5°C) and saline (up to 35.05) AW core next to the continental shelf break (cf. Figure 3c of this paper and Figure 2 of Ivanov et al. ). The two cores of the Atlantic Inflow, the FSB above the lower continental slope and the ASBB near the shelf break, are evident in the θ-S and velocity fields (Figures 3b and 3d) on the section at ca. 104°E (the region marked as SZP in Figure 1) in the Nansen Basin. The model gives very similar structure of the branches of the inflow with Barents Sea dense water being slightly warmer than in the observations (not shown).
 We compared observed along-slope velocities from the yearlong September 2003–2004 mooring M1 in the Laptev Sea (Table 2) with the simulated velocities at the same location (region marked SIS in Figure 1). For comparison, the observed velocities have been averaged into 5-day periods coinciding with the simulated 5-day mean velocities and then binned into the depth intervals corresponding to those of the model levels. The model and observations show similar velocity structure with a core at ca. 200–300 m (Figure 4). However, the northward component in the model is stronger than is observed above ca. 180 m and below ca. 700 m. Since the data is unavailable above ca. 100 m it is impossible to compare model results to measured currents and explain why observed currents decrease to zero at about 150–180 m depth, whereas the simulated flow still weak ∼1.5 × 10−2 m s−1, but nonzero. The time series analysis demonstrated a reasonable correlation between simulated and observed along-shelf currents with an averaged correlation coefficient for the depth range of 100–900 m R = 0.7, however the correlation for the northward (across shelf) component at 150–180 m depth was lower, 0.3–0.6. The temporal variability of the observed and simulated currents on the scale between synoptic (ca. 5-day) and seasonal (ca. several months) is comparable (thin lines in Figure 4). Since this time-scale is attributed to transient eddies [e.g., Woodgate et al., 2001] it suggests the model reasonably simulates the eddy structure of the ACBC except for the 150–180 m depth range, where the eddies interact with halocline. Dmitrenko et al.  reported a transient eddy passing in the vicinity of the mooring M1 in 2005. They observed that the structure of the flow was remarkably similar to the one depicted in Figure 4 and is evident of the similar decrease of the cross-slope velocity component at ∼150 m depth [cf. Dmitrenko et al., 2008, Figure 4]. Therefore we speculate that the model overestimates the subsurface cross-shelf flow on the seaward periphery of the FSB due to biases in simulating the transverse component of the eddy velocity. To understand the causes of this bias, more analysis is required which is beyond the scope of the present study. The observed and simulated density structure at 200–900 m depth is consistent with the geostrophic flow in the FSB.
 To assess the quality of AW simulations we calculated model positions of θ and S maxima within the AW layer (50–600 m depth range) for summers 2002–2006 and compared these with cross-slope summer sections in the Laptev Sea at ∼126°E (NABOS, http://nabos.iarc.uaf.edu/index.php). Average model depth and offshore distance of the θ and S maxima (254–283 m, 139 ± 121 km and 352–393 m, 100 ± 59 km respectively) match those observed (255 ± 48 m, 122 ± 103 km and 444 ± 43 m, 106 ± 96 km) within one standard deviation. We interpret this result to mean that the model simulates the location of the AW accurately.
 Model agreement with observations in the Laptev Sea is also illustrated in Figures 5 and 6 by comparing mean summer simulated θ and S distributions at ca. 126°E and those observed at the sections ca. 123°E (September 2007) and ca. 126°E (September 2009). As follows from these plots, the model simulates the correct position of AW θ -core, although overall the AW layer is too thick (cf. Figures 5d, 6c, and 6d). The depths of the density surfaces in the range of 25.0–27.9 kg m−3 in the model are in good agreement with observations, whereas the 28.0 kg m−3 surface is too deep. However, for the purpose of the present analysis, it more important that the model simulates well the wedge of cold, fresh water next to the shelf break and horizontal density gradients in the vicinity of the continental slope and the outer shelf. This gives us grounds to expect that the current structure predicted in the model is a real feature and not a model artifact. Moreover, the shear in the geostrophic velocities calculated from the observed density structure suggests a fast eastward current next to the shelf break, in agreement with the simulation results (cf. Figures 5a, 6a, and 6b); the discussed earlier mooring M1 is marked with gray broken line.
3.1.2. Canadian Arctic
 We compared simulated velocities with current meter observations at the high-resolution mooring array (mooring spacing ca. 5 km) in the Beaufort Sea at 152°W, north of the Alaskan coast [Nikolopoulos et al., 2009]; the region is marked as ALS in Figure 1. The averaged fro the period 1 August 2002 to 1 August 2004 along-stream (azimuth 125°) velocity component in both observation and model presents an eastward velocity core at ca. 100 m, which is identified as the Alaskan Shelf break Current (ASC) (Figure 7). This jet is fed by the Alaskan Coastal Current (ACC) flowing into the Beaufort Sea via the Barrow Canyon [e.g., Spall et al., 2008; Pickart et al., 2011]. The correspondence between the simulated and observed velocity is good, considering that the model horizontal resolution (ca. 8 km) cannot fully represent the steep topography in the area. The simulated θ-S fields are also in reasonable agreement with the observed ones (Figures 7a and 7b). However, the surface layer over the Alaskan shelf in the model is too cold due to excessive summer sea ice cover in the Chukchi Sea and consequently reduced solar insulation of the ocean.
 A boundary current (which in the model is identified as the FSB) along the Alaskan continental slope is thought to be one of the principal pathways of the AW in the Canadian Arctic [e.g., Woodgate et al., 2007]. Figure 8 depicts the ASC in the context of the large-scale flow in the Beaufort Sea. From the model results there are two eastward currents: the ASC, residing over shelf break, is the weaker; the ACBC, residing at 250–600 m depth over continental slope, is the stronger. Although velocities in the cores of the both currents are comparable (1989–2004 means are ca. 0.04–0.06 m s−1), the ACBC is wider. In contrast to the Eurasian Arctic, at this model section the tail ends of the ASBB and FSB become undistinguishable, forming a single velocity core of the ACBC, although there is hint of separation (Figure 8a).
 We also inspect the modeled ACBC north of Greenland in the Lincoln Sea (Figure 9). The simulated velocity in the core of the current is 0.06–0.07 m s−1, close to the range of 0.05–0.06 m s−1 obtained from current meter moorings by Newton and Sotirin . Observed and simulated θ-S fields are also in reasonable agreement. In the Lincoln Sea the ACBC forms a single core (Figure 9a).
 Summarizing, we conclude that the model simulation of the upper and intermediate depth circulation in all regions of interest is in line with observation and the model is fit for the purpose of the study.
3.2. Flow of the Fram Strait and Barents Seawaters Along the Siberian Shelf
 The model results show that to the east of the St. Anna Trough the strongest flow is in the ASBB (maximum speed 1989–2004 velocity 0.13 ± 0.05 m s−1 at ca. 104°E,) and is located at depths of 100–350 m near the shelf break. Figures 5a and 5b depict mean simulated velocity along the continental slope for the averaged March and August 1989–2004. The model section is chosen to coincide with the high resolution Polarstern and NABOS transects. In the simulations the ASBB enters the Eurasian Basin on the eastern side of the St. Anna Trough and then flows counter-clockwise (cyclonically) along the Siberian shelf over the 1500 m isobath and transports the cold, fresh shelf water from the Barents Sea. This water has θ and S of the Arctic halocline waters (−0.6 ± 0.2°C, 34.33 ± 0.05). Following Rudels et al.  we identify this water as the Barents Sea halocline water (BHW). In the simulations the heavily diluted Kara Seawaters enter the western Laptev Sea through Vilkitsky Strait and continue east above the outer shelf, forming the uppermost part of the ASBB (we will discuss this in section 3.4).
 The second core of the ACBC (maximum mean 1989–2004 speed of 0.09 ± 0.03 m s−1 at ca. 104°E) is at a depth of 200–400 m and above the ca. 3000 m isobath. The core transports AW and is seaward of the ASBB. This is the well-known FSB. In the simulations it has a velocity maximum offset shoreward from both the θ and S maxima within the AW (Figure 5).
 The lowest part of the ACBC (depth 900–1300 m) contains the densest, most saline fraction of the outflow from the St. Anna Trough. The source for this outflow is Barents Seawater (BSW), formed through convection in the Eastern Barents Sea [e.g., Schauer et al., 2002; Aksenov et al., 2010a]. In the model, BSW flows on the eastern side of the deepest part of the St. Anna Trough and descends into the Nansen Basin to a depth of ca. 900 m, then spreading within the depth range of 900–1300 m eastward along the continental slope and partly away from the continental slope into the basin interior. The simulated eastward flow of BSW is weak, velocities <0.02 m s−1, and of an intermittent nature. This agrees with observations. For example, in the NABOS CTD transects completed in summer 2002–2009 at ca. 126°E, the BSW signal can be found only in 2006, 2007 and 2009 (http://nabos.iarc.uaf.edu/index.php). The eastward flow of BSW is bounded by the lower continental slope and represents the conventional BSB [e.g., Schauer et al., 2002]. In the model the BSB velocity core is not always distinguishable from the lower part of the ASBB velocity core due to the weakness of the former.
 Simulations with passive tracers confirmed the sources of the three branches of the ACBC. In contrast to “off-line” tracer experiments, which utilize model velocity output to advect tracers with some added prescribed diffusion, in our simulations we employed “on-line” tracers. This allowed the tracers to evolve with full ocean dynamics during the run, thus improving the accuracy of the simulations. The tracers were initialized to unity on 1 January 2004 in the Fram Strait (AW tracer), in the top 150 m in St. Anna Trough (BHW tracer) and in the lower (below 150 m) portion of model cells in the trough (BSW tracer); points of release are marked in Figure 1. Figures 10a and 10b depict annual averaged concentrations of the tracers for 2006, along with cross-section velocity and θ, averaged for the same period.
 The correspondence between the branches of the ACBC and the source water masses is evident (Figures 10a and 10b). The second, offshore maximum of the AW tracer is due to the meandering of the FSB through the section. The meander is also present in the θ field as a second maximum in the simulations as well as CTD sections (cf. Figures 6d and 10b). The BSW tracer concentration is much smaller than the other tracers, possibly owing to the smaller velocities in the BSB. However, the tracer distribution is probably not in equilibrium due to the short integration period; this also may explain the low concentration of the BSW tracer. In the paper, we examine the FSB and ASBB in more detail to gain confidence that these are not merely model artifacts; the BSB will be discussed only briefly.
 The ASBB has a strong seasonal cycle with the maximum eastward along-shelf flow occurring in winter and spring and the minimum in summer and autumn (Figure 11a). The seasonal cycle of the FSB is much weaker, with the eastward velocity increasing in autumn and winter and decreasing in spring and summer. Here we analyze the velocity cores, defined as the maximum monthly mean eastward velocity in the branches, thus filtering out the short-term velocity fluctuations due to transient eddies. The velocity increase is accompanied by the deepening of both the ASBB and FSB (Figures 5a and 5b) and by the cross-shelf displacement of the cores by 20–30 km shoreward (Figure 11a). The θ and S maxima in the FSB shift shoreward in summer and autumn. The branch is colder in winter and spring and warmer in summer and autumn; the same pattern has been seen in observations [Ivanov et al., 2009; Dmitrenko et al., 2009]. The simulations show significant interannual variations in strength of the FSB and ASBB velocity cores (Figure 11b). The cross-shelf (latitudinal) position of the FSB also varies whereas ASBB is more stable. The corresponding salinity of the velocity cores is relatively constant in the FSB and varies in the ASBB on the annual and interannual scales (Figures 11a and 11b). The freshening and cooling (not shown) of the ASBB in autumn and winter occurs because of the entrainment of the fresh shelf waters.
3.3. Flow Along the Alaskan Shelf
 In the simulations the FSB flows above the lower continental slope, transporting AW and Arctic halocline waters cyclonically in the Beaufort Sea (Figure 8). This agrees with limited hydrographic observations in the Beaufort Sea [e.g., Shimada et al., 2005; Woodgate et al., 2007]. In the Beaufort Sea the ASBB is much weaker (velocity of ca. 0.02–0.03 m s−1) than in the Eurasian Arctic. From the passive tracer simulations (not shown) it was found that the ASBB contains BHW.
 The ASC resides above the upper shelf/shelf break as described in section 3.1 (Figures 7 and 8). This current has been discussed in detail by Spall et al.  and Nikolopoulos et al. . In the model the ASC begins at the Bering Strait inflow through the Barrow Canyon. The simulations show significant variation of the ASC in strength, with occasional reversal from eastward to westward flow, similar to observations [Nikolopoulos et al., 2009]. The model transport of PW within the ASC is 0.11 ± 0.21 Sv compared to 0.13 ± 0.08 Sv observed. The model and observations [Nikolopoulos et al., 2009] also suggest the presence of another weaker, eastward core with velocities 0.02–0.03 m s−1, located at the depths 200–250 m, ca. 20 km seawards from the shelf break (Figure 7). The core carries waters with halocline properties with θ ∼ −0.4–0°C, S ∼ 34.40–34.60, similar to BHW.
3.4. Branches of the ACBC on Pseudo-Isopycnals
 Following the method described in section 2.3.1, we employed the Montgomery function to diagnose the simulated ocean flow fields. Figures 12–14 show various fields from the simulations, averaged over March and August 1989–2004, on the pseudo-potential density (rB) surfaces approximating the potential densities σ0 = 25.07, 27.70, 27.90 and 28.00 kg m−3. The surfaces were chosen to represent (1) the PW inflow through Bering Strait (25.07), (2) the BHW inflow into the Nansen Basin (27.70), (3) the circulation of the AW entering Arctic through Fram Strait (27.90), and (4) the BSW inflow through St. Anna Trough (28.00). We refer to these below as the PW, BHW, AW and BSW surfaces. We note that the deviation between rB and σ0 is small: <0.05 kg/m3 near the Arctic shelf and ∼0.01 kg/m3 in the Arctic Ocean interior. Figures 12–14 show potential temperature θ, Boussinesq Montgomery function MB and depth of the specified pseudo-isopycnal. We show winter and summer fields for BHW and AW surfaces to demonstrate seasonal variability of the water properties and circulation on these surfaces. We chose the winter fields for the BSW surface as this water mass is formed through winter convection in the Barents Sea. The summer was chosen for the PW surface as this season is characterized by stronger inflow through Bering Strait, thus giving more distinctive flow along the Alaskan shelf.
 In summer the BHW, AW, and BSW surfaces do not outcrop but instead ground against the ocean floor (Figures 12b, 12d, 12f, 13b, 13d, and 13f, BSW surface is not shown for summer). The PW-surface outcrops to the ocean surface in the Barents Sea, Eurasian Arctic and in the Chukchi Sea, next to Bering Strait (Figures 14b, 14d, and 14f). Most of the PW surface outcropping in the Arctic Ocean occurs under sea ice, suggesting modification of this density class by buoyancy loss during sea ice formation. In winter the BHW surface outcrops to the ocean surface in the Greenland Sea and eastern Fram Strait, in the Barents Sea and around Franz-Josef Land (Figures 12a, 12c, and 12e). There is a moderate outcrop of the AW surface in the central Greenland Sea, southeastern Barents Sea and in the Bear Island Trough of the Barents Sea (Figures 13a, 13c, and 13e). The BSW surface outcrops in the seabed depression in the southern Barents Sea (Figures 14a, 14c, and 14e); this area is known for winter convection [e.g., Schauer et al., 2002]. Although there is no apparent connectivity between the outcrop area and the Arctic Ocean in the multiannual mean winter fields, the individual winters demonstrate continuous flow of BSW from the outcrop area into the Arctic Ocean.
 To complement the discussion in sections 3.2 and 3.3, here we describe the Arctic Ocean circulation inferred from the Montgomery analysis, starting from the upper layers. The PW surface presents anti-cyclonic geostrophic flow in the Canadian Arctic. The flow manifests itself with higher values of MB in the central Beaufort Sea and lower values toward the periphery of the Arctic Ocean (Figure 14d). Obviously at the shallow depths (10–20 m) of this PW surface, flow is not purely geostrophic but influenced also by Ekman drift. The analysis (supported by data from the Freshwater Switchyard Project (M. Steele, personal communication, 2011)) also suggests westward flow north of CAA and Greenland, the reverse of contemporary views. However, analysis of this feature is outside the scope of the paper; more discussion is given by Aksenov et al. [2010b].
 The flow of cold surface waters from the Kara Sea enters the Laptev Sea through Vilkitsky Strait and continues eastward along the shelf break as the uppermost part of the ASBB (Figures 5 and 14d). Beyond the Lomonosov Ridge, these waters become entrained into the anti-cyclonic circulation, spreading through the Canadian Basin.
 On the Pacific side of the Arctic Ocean, the inflow through Bering Strait generates a strong northward current, which flows in the eastern Chukchi Sea. The current turns eastward along the Alaskan coast, eventually becoming the ASC (Figures 7 and 14d). The MB fields on the PW surface and on deeper surfaces down to rB = 26.80 (not shown) are evidence of the bifurcation of the circulation north of Point Barrow, at about 156°W. The lower part of the flow continues eastward along the Beaufort Sea shelf and into the Canadian straits, with the upper part turning westward and joining the outer anti-cyclonic flow (Figure 14d). This results in PW crossing the Arctic Ocean along two pathways: one through the Beaufort Sea and one along the Siberian shelf. This is also consistent with the spread of a PW “color” tracer released in the model for the period 1996–2006 in the northern Bering Strait. The results of this tracer experiment have been supplied by Dr Martin Wadley, University of East Anglia, UK. The simulations (not shown) show westward spread of the PW toward the Siberian Shelf in the subsurface, shallower than 50–70 m, layers and flow of the PW eastward along the Alaskan Shelf toward the Canadian Archipelago in the layers deeper than 70 m. Model results are in agreement with observations on the PW distribution in the Arctic Ocean obtained from silica concentration and phosphate/nitrates ratio [e.g., Jones et al., 2003, 2008; Shimada et al., 2005; Nikolopoulos et al., 2009; Abrahamsen et al., 2009].
 The flow on the deeper surfaces (i.e., BHW, AW and BSW) is cyclonic, evidenced by the lower values of MB in the center of the Arctic Ocean, and the higher values along the Arctic shelf (Figures 12c, 12d, 13c, 13d, and 14c). The flow follows submarine ridges, resulting in cyclonic gyres in each of the basins of the Arctic Ocean; the strongest gyre is in the Nansen and Amundsen Basins, the others are in the Makarov Basin and the Beaufort Sea.
 The continuous cyclonic boundary current, the ACBC, enters the Arctic Ocean through the eastern Fram Strait, flows east along the Arctic continental shelf with occasional excursions into the trenches of the Arctic shelf, and exits the Arctic Ocean on the western side of Fram Strait. In this circumpolar circulation the branches of the ACBC are present on all three surfaces, BHW, AW and BSW (Figures 12c, 12d, 13c, 13d, and 14c). There are large cross-slope gradients of MB showing the ACBC's location and strength, and the magnitude of the gradient gradually reduces downstream along the continental slope as the current weakens (Figures 12c, 12d, 13c, 13d, and 14c). The ACBC flows eastward along the Siberian Shelf reaching the Chukchi Plateau, where the current divides in several topographic jets (Figures 12c, 12d, 13c, 13d, and 14c). After negotiating the topography of the Plateau the current merges into a single core and continues eastward along the Beaufort Sea shelf and Canadian Shelf. The upper part of the current branches into the Canadian Archipelago with the main flow leaving the Arctic Ocean through the western Fram Strait (Figures 12c and 12d). The lower part of the ACBC partly diverts offshore along the Lomonosov Ridge, Mendeleev Ridge and around the outer rim of the Chukchi Plateau (Figure 14c).
 Along-track modification of the ASBB is gradual; the increase in MB at the BHW surface above the Chukchi Plateau and next to the Canadian Shelf is weak in comparison to the drop MB undergoes downstream of the Barents Sea (Figures 12c and 12d). In contrast, on the AW surface the along-track decrease in MB is smaller (Figures 13c and 13d). At the BSW surface the MB increases downstream in winter due to the cascading of the winter water from the Barents Sea Shelf through St Anna Trough and from the Chukchi Sea in the vicinity of the Northwind Ridge. The current, evident at the BSW surface, begins in the vicinity of the northern continental slope of the Barents Sea and continues along the Arctic continental shelves to Fram Strait (Figure 14c). There is noticeable downstream deepening of the ASBB core from ca. 150 m depth on the Eurasian side of the Arctic to ca. 250 m depth on the Canadian side (cf. Figures 3b, 5a, 5b, 8a, 12e, and 12f).
Figures 12c, 12d, 13c, and 13d are evidence of the separation of the ASBB and FSB cores along the Siberian Shelf and around the Chukchi Plateau. Cold (ca. −1.0°C) BHW flows in through St. Anna Trough within ASBB, and can be traced as the core of the rim current as far east as 160°E (Figures 12a and 12b) and further in the Canadian Arctic. Much warmer (ca. 2.4°C) AW flow enters the Arctic Ocean through Fram Strait and spreads in a cyclonic manner through the entire Eurasian Basin of the Arctic Ocean (Figures 13a and 13b). The first separation of the FSB from the shelf break and displacement warm AW away from the shelf is apparent in the vicinity of St. Anna Trough (Figures 13a–13d). In the Canadian Basin the AW flows mostly along the shelf with weak offshore flow; further it flows eastward along the continental shelf of the Canadian Archipelago, leaving the Arctic Ocean as a much colder (ca. 0.4°C) water mass through the western Fram Strait.
4.1. Mechanisms Driving the Continental Slope Currents
 What drives the cyclonic ASBB in the model? Local winds along the Siberian coast of the Laptev Sea are southwesterly, suggesting an eastward along-shelf flow. In the sea-ice-covered Eurasian Basin the sea ice modifies the wind-forcing of the ocean, and so the local stress on the ocean is, in our model, too weak to drive the current (Figures 15a and 15b). Thus we conclude that over the interior Arctic Ocean, as far as the model is concerned, the Ekman suction is small, and local wind cannot drive the ASBB. Following sections investigate alternative processes driving the flow.
4.1.1. Application of the Circulation Integral to the Arctic Basin
 In the previous section we inspected the Montgomery stream function to study the Arctic circulation. The method allowed us to identify the structure of the ACBC and its branches, to analyze the continuity of the flow and water mass constituents. As we demonstrate below, the Montgomery function also allows us to identify the mechanisms driving the flow.
 Our approach is first to apply the circulation integral (8) to the interior Arctic Basin. For the multiannually averaged March and August rB = 27.7 kg m−3 surfaces (Figures 12c and 12d) we choose a contour crossing Fram Strait, then passing eastward along about 80–82°N to Franz Josef Land, across the St Anna Trough, and then returning to Greenland following grounding lines along the Siberian and Canadian continental slopes. For flows evolving on timescales longer than inertial timescales, and small Rossby numbers, the layer-integrated PV flux (or Coriolis acceleration) across a strait is simply the difference in MB across the strait (8d). Our results show a large change in MB across St. Anna Trough, driven by high values of MB over the Barents Sea, but little net change in MB across Fram Strait, supporting the conclusion of Karcher et al. , that the inflow through St Anna Trough is the main driver of cyclonic flow in the Arctic Basin. The Montgomery function gradually declines moving cyclonically around the Siberian and Canadian shelves, consistent with frictional dissipation of the (cyclonic) boundary current in these regions.
 Cyclonic flow on the AW surface, rB = 27.9 kg m−3 (Figures 13c and 13d) is again driven by inflow through St Anna Trough (though more weakly) and slowed by friction elsewhere. The BSW surface, rB = 28.0 kg m−3 (Figure 14c) also shows higher MB eastward of St Anna Trough, together with net outflow through Fram Strait. As discussed in section 3.2 and section 3.4 above, although the BSW surface does not exist in the multiannual mean in the St Anna Trough, it does exist there intermittently. ‘Blobs’ of BSW formed in the Barents Sea pass northward down through St Anna Trough and eventually cross the mean grounding line, carrying PV into the interior Arctic basin, and spinning up cyclonic flow.
 We suggest that a similar mechanism drives the eastward flow along the Alaskan shelf. The values of MB on the pseudo-isopycnals rB = 25.0–26.0 kg m−3 are high in the Bering Strait and along the Alaskan coast and lower offshore. Figure 14d, on rB = 25.07 kg m−3, suggests that strong PV inflow through the Bering Strait is the main driver of the cyclonic ASC; this mechanism is discussed in section 4.1.2 below.
 The PV-induced flow may be further enhanced by the ‘Neptune effect’ [Holloway et al., 2007; Holloway and Wang, 2009], a mechanism by which eddies may drive rectified mean flows along the continental slope. Merryfield and Scott  found that two high-resolution (ca. 0.1°) OGCMs generated more cyclonic flows along continental slopes in the Arctic than equivalent coarse resolution models. The same is evident in OCCAM: the eddy-permitting/resolving 1/12° model has a stronger boundary current than the 1/4° (27 km resolution) which does not resolve eddies in the Arctic. We compared these results with the integrations of 1° and eddy-permitting 1/4° configurations of the global Nucleus for European Modeling of the Ocean (NEMO) model (described, e.g., by Madec ). These configurations are run at the National Oceanography Centre and have 24–48 km and 6–12 km resolution in the Arctic respectively comparable to OCCAM 1/4° and 1/12°. Similar to OCCAM there is a stronger ACBC in the eddy-permitting NEMO configuration than in the coarser-resolution 1° NEMO. Thus, eddy processes appear to have a significant impact on the boundary currents and we expect the strength of the ACBC will increase in higher resolution, fully eddy-resolving models.
4.1.2. Sources of PV
 The above circulation integral analysis has suggested that flow from the Barents Sea through the St Anna Trough is primarily responsible for driving the cyclonic boundary current in the Eurasian Arctic. The next question is then what drives this flow. Unfortunately it is difficult to apply the circulation integral (8) over the Barents Sea, where winds, density outcrops and buoyancy forcing vary so greatly over the seasonal cycle (Figure 16). Our conclusions are therefore somewhat tentative. It may be easier to simply consider the processes that might create the high pressure in the dense waters of the Eastern Barents Sea that drives the flow through the St Anna Trough.
 First, inflow into the Barents Sea by the Nordkapp Current between Svalbard and Norway (Figure 12d) acts as a strong source of waters with pseudo-density rB = 27.70 (BHW). There is however no connection on the deeper BSW surface, so the water must be locally formed within the Barents Sea. Our analysis demonstrates that buoyancy loss is significant in winter over much of the Barents Sea; in particular over outcrops of both the pseudo-density surfaces rB = 27.70 kg m−3 (BHW) and rB = 27.90 kg m−3 (AW) (Figure 16a), and so generates the deeper AW and BSW waters. This is consistent with Karcher et al.'s  finding that buoyancy forcing in the Barents Sea is a main driver of the flow.
 The impact of wind-forcing is less clear. In winter (March, Figure 16b) there is Ekman suction over most of the Barents Sea, but pumping in the extreme east, near Novaya Zemlaya. This favors northward flow toward the St Anna Trough. Wind-forcing is considerably weaker in summer (Figure 16d). With sea ice declining, both buoyancy forcing and local wind driving are expected to play a greater role in future forcing of intermediate water circulation in the Arctic Ocean.
 A similar PV mechanism acts in the Canadian Basin: the sea surface drop between the Northern Pacific and the Arctic Ocean creates a pressure increase in the Bering Strait and drives the ASC. Nikolopoulos et al.  reported the maximum eastward flow within the current in spring-summer (March–September) and the maximum westward flow in winter (October–February), which coincides with high PV inflow in summer (high Bering Strait inflow and stronger stratification due to surface warming) and low in winter (low and weakly stratified Bering Strait inflow). This mechanism likely complements the local wind-forcing of the flow [Nikolopoulos et al., 2009].
4.1.3. ASBB in Other Models
 A similar triple-core structure of the ACBC in the Eurasian Arctic is present in other high-resolution z-level models, for example, in the 46-level, 1/4° resolution global NEMO model (described by Madec  and Lique et al.  with 6–12 km resolution in the Arctic) and in the 64- and 75-level versions of this model developed by the National Oceanography Centre, Southampton (see section 4.1.1). In these models velocities in the ASBB and FSB cores are very close to those evident in OCCAM. Since all these models have different architecture, including different model grids, vertical resolution, ocean physics, atmospheric forcing, restoring fields and sea ice models, we conclude that the circulation feature is not model specific but represents a real physical process.
4.2. Shelf-Basin Exchange
 To understand the implications of the ASBB for the Arctic halocline we examine water mass transformation for closed regions of the Eurasian Basin and Laptev Sea, using the Large and Nurser  method described earlier in section 2.1. To compare model results and observations we chose region boundaries close to the NABOS observational transects and with the Severnaya Zemlya Polygon. For water mass analysis and to compute oceanic transports across the boundaries of the regions, we defined AW with T ≥ 0°C, 34.70 ≤ S < 34.95 and halocline water with T < 0°C, S < 34.70.
 In the model, the ASBB is the main route for the shelf waters of the Barents and Kara Seas to enter the Arctic Ocean, bringing as much water through the St Anna Trough into the Arctic Ocean (1.01 ± 0.32 Sv at ca. 104°E) as the inflow through Bering Strait. In the Eurasian Basin the strength of the ASBB is about 40% of the FSB (2.55 ± 0.65 Sv at ca. 104°E). Part of the ASBB (0.13 ± 0.10 Sv) diverts into the Eurasian Basin; the rest (0.79 ± 0.24 Sv) flows into the Canadian Basin over the Lomonosov Ridge contributing to the lower halocline waters in the basin. In the Laptev Sea the model shows that the halocline formation rate due to atmospheric heat and freshwater fluxes (0.29 ± 0.15 Sv) is ca. 30% less than its depletion through mixing (0.39 ± 0.14 Sv) with the adjacent water masses, thus the halocline effectively loses 0.10 ± 0.07 Sv water in the area. This is because the halocline depletion (ca. 0.15 Sv) is larger than its gain through atmospheric heat and freshwater forcing (ca. 0.07 Sv) in the central Laptev Sea. In the western Laptev Sea the mixing loss and gain of the halocline through atmospheric heat and freshwater forcing are comparable (both ca. 0.25 Sv). In the vicinity of the Lomonosov Ridge (the eastern Laptev Sea) the moderate simulated mixing rates (ca. 0.04 Sv), compared to ca. 0.08 Sv of flow divergence, point out that the lateral exchange can be as important as mixing for shelf-basin interactions in the area, as observations also suggest [e.g., Lenn et al., 2009]. All rates exhibit large variations due to the strong seasonal cycle. The simulated effective loss rate of the halocline of 0.09 ± 0.06 Sv north of Severnaya Zemlya is within the 0.06–0.12 Sv range obtained from hydrographic surveys in the Severnaya Zemlya Polygon [Walsh et al., 2007]. In the model, the net halocline export from the Barents, Kara and Laptev Seas into the Eurasian Basin is ca. 0.07 Sv, the same as the halocline in the basin gains because of the atmospheric heat and freshwater forcing (ca. 0.06 Sv); therefore the halocline in the basin is renewed through both advective and convective mechanisms [Steele and Boyd, 1998]. The renewal timescale of the halocline water in the Eurasian Basin due to its loss in the Laptev Sea in the model is ca. 26 years, in the range of observational estimates, ca. 20–39 years [Walsh et al., 2007].
4.3. Implications for Observational and Modeling Strategies
 There are challenges in observing the Arctic Ocean. The perennial sea ice, remoteness and inaccessibility of the area make observations very difficult. Significant temporal variability of the ocean circulation presents difficulties in interpreting the measurements [Melling et al., 2008]. Last, the small Rossby radius in the Arctic results in circulation features of a few-kilometer spatial scale [Nurser, 2009; Woodgate et al., 2001], necessitating measurements of high spatial resolution [Schauer et al., 2008]. Some Arctic sections have been recently instrumented with closely spaced moorings [e.g., Nikolopoulos et al., 2009] or with ship-mounted ADCP. This has led to a revised view of oceanic circulation, which can be attributed to the improved resolution or a longer time series, or both [e.g., Schauer et al., 2008]. The use of eddy-permitting/eddy-resolving models clearly demonstrates improved realism in simulations of the Arctic circulation [Maslowski et al., 2008; Clement Kinney et al., 2009; Aksenov et al., 2010a, 2010b]. Modeling and sustained hydrographic observations in key locations can help to overcome the paucity of observations. Models can provide an alternative to extensive in situ observations to infer ocean circulation and be useful in offering target hypotheses for observational campaigns on the quasi-synoptic scale.
 Finally, we would like to comment on why the ASBB and associated near-slope halocline structure have not been registered in the observations for so long. The ASBB is located at the upper shelf-slope, which is a difficult place to put moorings in, and moreover, the current and the halocline “wedge” are narrow features, with width only a few Rossby radii. Thus, they were able to slip between “standard” station positions with relatively wide spacing. Additional hydrographic stations completed on the Laptev Sea shelf break in September 2009 confirmed the existence of the modeled features, prompting a search for indications of the ASBB in previous measurements.
 Model results suggest that AW and halocline waters flow along the Siberian shelf of the Laptev Sea as a triple-core current: a conventional Fram Strait Branch (FSB) and Barents Sea Branch (BSB) and the newly identified Arctic Shelf Break Branch (ASBB). In the model, the ASBB is the main route for the shelf waters of the Barents and Kara Seas into the Arctic Ocean whereas FSB transports Atlantic Water (AW) from Fram Strait. The dense water formed in the Barents Sea (BSW) constitutes the BSB. The strength of the ASBB is almost half of the FSB and this current brings as much light water into the Arctic Ocean as the inflow through Bering Strait. The present analyses consistently show the ACBC is continuous throughout the Arctic Ocean when all three cores (ASBB, FSB and BSB) are considered. The model thermohaline structure agrees reasonably with recent observations across the continental slope, and modeled and measured velocities compared well.
 Buoyancy loss over the Barents Sea is the main mechanism responsible for driving the cyclonic ASBB through the inflow of the PV into the Eurasian Arctic Ocean via the St Anna Trough, possibly further enhanced by winds and eddy-topography interaction. An analogous PV mechanism acts in the Canadian Arctic where the sea surface drop between the Northern Pacific and the Arctic Ocean creates a pressure difference across the Bering Strait and drives the ASC, pointing at the two primary sites in the Arctic Ocean (Barents Sea and Bering Strait) where PV influx occurs and drives the upper ocean circulation.
 Using the model results and observations we have demonstrated that the ASBB is the main route for the shelf waters and that lateral exchanges with the ASBB can be as important as vertical mixing for halocline evolution. The model shows halocline loss in the Laptev Sea through mixing (0.39 ± 0.14 Sv) and export (0.13 ± 0.10 Sv) into the Eurasian Basin. The principal areas of halocline depletion are the western and central Laptev Sea. The simulations demonstrate that the relative contributions of the shelf-basin exchanges and air-sea-ice interactions in renewal of the Eurasian Arctic halocline are comparable, therefore suggesting both the advective and convective renewal of the Eurasian Basin halocline with the renewal timescale of ca. 26 years. Further observations on the continental slope and the shelf break, including measurements of the currents, are needed to support this model conclusion and to help us understand more fully the dynamics of the Atlantic and halocline waters in the Arctic.
 We would like to acknowledge the use of the observations collected in the Beaufort Sea near Alaskan Shelf and kindly supplied by Robert Pickart and Wilken von Appen (Woods Hole Oceanographic Institution, USA) along with the plots of these observational results. We also would like to thank Robert Pickart and Wilken von Appen for providing invaluable comment on the manuscript. We are very grateful to Beverly de Cuevas (National Oceanography Centre Southampton, UK) for the extensive help with the model data analysis and to Martin Wadley (University of East Anglia, UK) for contributing Pacific Water tracer simulations. We would like to thank the anonymous reviewers for the thorough comments which helped to improve the manuscript. This research has been primarily supported by the Arctic Synoptic Basin-wide Oceanography Consortium, Natural Environment Research Council (NERC), UK, and by the Oceans 2025 Research Programme of the above research council. Alberto Naveira-Garabato was supported by NERC through an Advanced Research Fellowship (NE/C517633/1). Vladimir Ivanov acknowledges support from NABOS and IARC for this research and Igor Polyakov acknowledges support from NASA, NOAA, NSF, and JAMSTEC. Long-term observations in Fram Strait have been supported by the EU projects ASOF-N (Arctic-Subarctic Ocean Flux Array for European Climate: North), DAMOCLES IP (Developing Arctic Modeling and Observing Capabilities for Long-term Environment Studies), and currently by ACOBAR (Acoustic Technology for Observing the Interior of the Arctic Ocean, grant agreement 212887). We also acknowledge the use of UK National high performance computing resource.