Vertical and spatial variability of tidal currents in the northern part of the Bering Sea near St. Lawrence Island were examined using observations made at 17 new and historical current meter mooring locations. The semidiurnal currents dominate over the diurnal currents. Semidiurnal currents exhibit a seasonal modulation with currents being somewhat stronger during the summer and fall when the density structure displays two-layer stratification. Diurnal currents exhibit a weak semiannual modulation. Examination of the vertical structure of tidal currents shows that maximum semidiurnal currents are observed in the upper or middle layers down to a depth of about 20 m, and below this depth the energy of the semidiurnal currents decreases rapidly. However, the energy of diurnal band currents remains almost constant with depth, decreasing only in close proximity to the bottom. Rotary spectral analysis applied to the records shows further differences: the M2 clockwise (CW) component of rotation dominates in the upper and middle water column but the counterclockwise component (CCW) dominates near the bottom. Interpretation of these observed results points to the different bottom boundary layers (BBL) for the CW and CCW motions in addition to the role of density stratification in organizing the BBL. We compare results of a vertically averaged two-dimensional model to the observationally derived tidal ellipse parameters and show that the barotropic formulation provides a first-order description of the tidal regime within the domain.
 Until now only a handful tidal current measurements have been reported from the northern Bering Sea [Pearson et al., 1981; Mofjeld, 1984]. In this paper we utilize these historical records as well as newly available current meter data from the vicinity of St. Lawrence Island in attempt to construct the spatial and temporal variability associated with tidal currents around this island. St. Lawrence Island is situated on the Bering Sea shelf (Figure 1) about 250 km south of Bering Strait and 500 km northeast of the shelf break.
 An encompassing oceanographic background for the Bering Sea shelf, including discussion of tides, hydrography, circulation, ice and meteorology, is given by Hood and Calder . Fall storms tend to mix and cool the water column until by December the northern shelf waters are vertically homogeneous and at the freezing point. Ice cover progresses from north to south [Niebauer et al., 1999] and by the end of December the entire northern shelf is ice covered. During the winter, a polynya is generated immediately to the south of St. Lawrence Island by periodic strong northerly wind events [Schumacher et al., 1983; Johnson and Kowalik, 1986]. Opening and closing of the polynya occurs until late April, when increasing solar radiation and warmer air temperatures melt the sea ice. Spring and summer insolation, the fresh ice melt and the spring river freshets progressively stratify the water column so that by the end of summer the northern shelf is essentially a two layer fluid: a homogeneous lower layer of cold, salty water remnant from the previous winter (mixed by tidal currents) and a homogeneous upper layer (mixed by winds) comprised of relatively fresh and warm waters. Surface cooling and fall storms then break down this stratification and the cycle begins anew. The unique location of the northern Bering Sea as a transition zone between the midlatitudes and the Arctic results in three different water types (Alaska Coastal Water, Bering Shelf Water, and Anadyr Water [Coachman et al., 1975]) that pass through Bering Strait to their eventual fate in the layers of the Arctic Ocean. The basic physical environment annual cycle of the northern Bering Sea region is depicted in Table 1.
Table 1. Basic Environmental Components of the St. Lawrence Island Region and Their Approximate Annual Distributionsa
Shaded boxes denote nominal months of influence by relevant processes.
High river discharge
Highly stratified water column
Homogeneous water column
 Tides are an important and consistent driving force in the Bering Sea. Over the southeast Bering Sea shelf, where shoaling and/or resonance effects in Bristol Bay dominate, up to 90% of the kinetic energy is of tidal origin. According to Coachman , in the St. Lawrence Island region tides account for only 10–40% of the total kinetic energy. Tide waves enter the Bering Sea as progressive waves from the North Pacific Ocean, mainly through the central and western passages of the Aleutian-Komandorski Islands [Sünderman, 1977; Pearson et al., 1981; Kantha, 1995]. The Arctic Ocean is a minor secondary source of tides, and therefore tides in the northern Bering Sea are considered to be the result of co-oscillation with the Pacific and, to a lesser degree, with the Arctic Ocean. Pearson et al.  constructed empirical charts for four major tidal constituents (M2, N2, K1, and O1) over the Bering Sea shelf. Mofjeld [1984, 1986] and Mofjeld et al.  analyzed data for three diurnal components (O1, P1, K1) and three semidiurnal components (M2, S2, N2) over the northeastern Bering shelf and made comparisons against numerical models and tidal theory. Observations show that over most of the shelf regions, the tides are primarily semidiurnal. However, near some of the Aleutian Islands, in Norton Sound and at some locations along the shelf break, diurnal tides can be dominant. The tides on the northeastern Bering Sea shelf are in a transitional regime, between low tides north of Bering Strait and high tides along the Bering shelfbreak.
 The major semidiurnal constituent M2 displays approximately 20 cm amplitude near St. Lawrence Island. The M2 tide is under the influence of amphidromic systems located near Bering Strait, southeast of St. Lawrence Island and in the Gulf of Anadyr [Kowalik, 1999]. The dynamics of diurnal tides can be described using the major constituent K1. K1 wave has a maximum amplitude of approximately 40 cm in the region of the shelf break and this amplitude decreases from the shelf break to about 7–8 cm in the St. Lawrence Island region, thus indicating the presence of trapped shelf waves [Baines et al., 2005]. Mofjeld  suggested that such behavior of the K1 tidal wave on the Bering Sea shelf resembles a Sverdrup wave generated by an incident wave from the deep Bering basin. The Sverdrup wave model predicts well the amplitudes and phases of the tides on the Bering Sea shelf. These waves, in a frictionless ocean below the critical latitude (for K1 tide, Φ = 30°), propagate as free waves; above this latitude they change to subinertial evanescent waves. Data described by Mofjeld  from the northwestern shelf show that the tidal harmonic constants change seasonally. The diurnal amplitude increased and semidiurnal amplitude decreased during winter. The increase and decrease amounted to a few percent of the mean values. The maximum change in amplitude is close to 3 cm and the maximum change in phase is up to 15° for the M2 tide in Bering Strait. The relatively large changes in amplitude (37% of the mean amplitude) at this site possibly expresses damping of the tides by ice in the Arctic Ocean [Kowalik and Proshutinsky, 1994].
 Five current meter stations located near St. Lawrence Island were described by Pearson et al.  and Mofjeld . For comparison to the observations newly available to this study, the moorings deployed in the 1970s and 1980s (NC19B, NC25A, NC26B, LD2, and LD3) are reproduced in Tables 2 and 3. These historical measurements give broader spatial coverage than the newer observations but none of the historical records were made over the course of an entire calendar year or over the entire water depth.
Table 2. Mooring Names, Locations, Water Column Depth, Instrument Depth, Instrument Type, and Record Lengtha
Water Depth, m
Instrument Depth, m
Record Length, hour
RCM, Aanderaa RCM-4 and RCM-7 instruments; ADCP, RD Instruments 300 KHz Acoustic Doppler Current Profiler.
Table 3a. Constituent M2 Tidal Parameters Derived From Overlapping 29-Day Analyses of the Current Meter Records and From the Barotropic Modela
Semimajor Axis, cm/s
Semiminor Axis, cm/s
Greenwich Phase, deg
Std. Dev., standard deviation. Given are semimajor and semiminor ellipse axes (cm/s), rotation direction (CW, clockwise; CCW, counterclockwise), the phase angle (degrees) referred to Greenwich and the inclination (degrees) of the semimajor axis referenced in a counterclockwise fashion from due east. Mean and standard deviations are computed with respect to the ensemble of 29-day analyses, beginning 15 days from the start of the record. Model parameters that fall within the 95% confidence limits of the observed parameters are shaded with boldface type. “VA” in the depth column for F6 and F8 refer to analysis of vertically averaged currents from these two ADCP moorings.
 Here we report on tidal analyses from 12 additional yearlong stations in the vicinity of St. Lawrence Island. The objectives of this paper are (1) to examine the temporal and spatial variations of the observed tidal currents, (2) to compare the observed currents with the calculated barotropic currents of a numerical model, (3) to investigate rotational spectra and the vertical distribution of tidal currents within the water column and (4) to apply bottom boundary layer (BBL) theory for an explanation of the observed structure of the tidal current characteristics within the water column. This paper is organized in the following way. In section 2 the tidal currents are examined using harmonic analyses and tidal current ellipses are compared to the results from the vertically averaged model. Seasonal and vertical variability in the tidal currents is described in section 3, including rotary spectra analysis to discern different behavior of the semidiurnal and diurnal components. Interpretation of the observations with the help of simple equations for the BBL in a rotational, density stratified flow is given in section 4. Section 5 provides discussion and concluding remarks. In Appendix A, a basic set of equations for the BBL is briefly described.
2. Spatial Distribution of Tidal Currents
 Twelve moorings were deployed to collect velocity data at hourly intervals over the period of one year from September 1998 to September 1999. Figure 1 shows a detailed map of the St. Lawrence Island region with these mooring locations and the five historical mooring locations taken from Pearson et al.  and Mofjeld . Table 2 summarizes the mooring names, instrument types, spatial coordinates, current meter depths, water column depths, and record lengths for the data sets employed.
 Recently deployed moorings utilized RD Instruments 300 KHz Acoustic Doppler Current Profilers (ADCPs) and Aanderaa RCM4 and RCM7 current meters. Linear interpolation between pre- and postcalibrations were applied to RCM data sets when possible. All records were corrected for clock drift when possible then interpolated with cubic splines to the top of each hour. RCM data was likely subject to some degree of jellyfish biofouling. The records that appeared to be affected primarily were P1 in February, P2 during summer months, FTS8 in winter, the upper instrument at H1 in late summer, and intermittently at H3 and H4. Comparison of speed distributions to the nearby ADCP instruments and other RCM instruments leads us to believe that the periods of intermittent rotor stalls have only a minor effect on the outcome of our harmonic fits, in part because of the averaging employed. ADCP derived velocities were corrected for speed of sound at the transducer head based on a salinity value measured by a nearby T/C recorder and temperature measured by the ADCP. In addition, the following quality control criteria were applied to the ADCP data: data failing any of the following tests were rejected. (1) Data determined to be in bins above the sea surface, (2) data within the beam sidelobe reflection shadow, (3) correlation of less than 64% on any one beam, (4) percent-good four-beam solutions <10%, and (5) percent-good four-beam solutions + percent-good three-beam solutions <20%. Hour-long ADCP ensembles collected 105 pings per ensemble so upon application of the above screening parameters, the standard deviation of each ADCP measurement is between 0.6 and 1.3 cm/s.
 Analysis software T-TIDE by Pawlowicz et al.  was used for all tidal harmonic fit computations. New results given in Tables 3a and 3b have been computed in a consistent manner to these historical analyses for ease of comparison. The essence of this analysis approach is consecutive 29-day harmonic analyses spaced every 15 days from the beginning of the record. Three rows are dedicated to each ADCP mooring: one near-surface depth, one near-bottom depth and the vertical average. Tidal ellipse parameters were not recomputed for the historical records. Rather, we have taken the previously published findings for these moorings [Pearson et al., 1981; Mofjeld, 1984], adjusting the inclination (and phase when necessary) to conform to the current standard of reporting ellipse inclination with respect to due east. Results from moorings LD2 and LD3 were published with one month of data so the standard deviation entry in Table 3 is blank for these moorings. Moorings LD2 and LD3 represent midsummer conditions; NC19B, NC25A, and NC26A represent tidal currents from the November through June time frame. See Table 1 and the discussion below for ramifications of these temporally restricted sample periods.
 The record-length analyses show that the M2 constituent contains the most energy, with current speeds between 3.1 and 16.0 cm/s. K1 is the largest diurnal constituent, with current speeds between 1.8 and 7.5 cm/s. For records with an entire years worth of data, the M2 (K1) current speed maxima and minima are 16.0 (4.1) and 6.2 (2.0) cm/s respectively. At ADCP mooring F8, vertically averaged M2 and K1 are 14.3 and 3.9 cm/s, respectively; the third through tenth largest resolved constituents are all in the range between 1.2 and 3.7 cm/s. The first 10 constituents at this mooring, in order of magnitude, are M2, K1, N2, MM, MSM, MF, 01, S2, SSA, and P1. All ten constituents have a signal-to-noise ratio greater than 2. The sum of these first ten amplitudes is about 38 cm/s. The mean current speed of a yearlong tidal prediction that includes all resolved constituents is about 12 cm/s. Larger standard deviations of the diurnal constituent K1 may reflect the difficulties in resolving K1 from P1 in the monthlong analysis even with the use of inference or may be an indication of the lower signal-to-noise ratio for K1. Propagation of the time of maximum M2 and K1 currents generally progresses from east to west for the region to the south of the island with the notable exception of mooring H5 off Southeast Cape. The three moorings located off Chibukak, Northeast and Southeast Capes have the three largest Greenwich phase values, indicating that these promontories may act to disrupt the wave propagation.
2.2. Vertically Integrated Model
 We compare tidal current ellipses from the measurements to the output of a tidal model constructed around St. Lawrence Island. The model employs a spatial resolution of approximately 1.85 km and the boundary conditions for the local model were taken from a large-scale model of the entire Bering Sea. The formulation, mechanics and performance of the model are described in detail by Kowalik .
 The ellipses for the M2 tide are shown in Figure 2 and ellipses for the K1 tide are shown in Figure 3. Ellipse parameters from the model grid points closest to the mooring locations are also given in Tables 3a and 3b. We applied two-sided t-tests to determine 95% confidence bounds for each of the observed ellipse parameters, assuming that the number of degrees of freedom is equivalent to the number of months sampled. Confidence limits computed in this fashion provide similar results to those given by the T-TIDE package. Modeled ellipse parameters that fall within the 95% confidence limits are shaded in boldface type within Tables 3a and 3b. We are unable to construct confidence limits for historical moorings LD2 and LD3 because only one month of data was reported. Semiminor axes for which the confidence limits cross zero indicates uncertainty in the direction of rotation and is reflected in appropriate shading of the rotation direction.
 For the M2 constituent, the mean magnitude of difference between the observed and modeled ellipse parameters given in Table 3 are the following: semimajor axis, 3.2 cm/s; semiminor axis, 1.4 cm/s; inclination, 17°; and phase, 18°. For the K1 constituent, the differences are the following: semimajor axis, 1.2 cm/s; semiminor axis, 0.5 cm/s; inclination, 18°; and phase, 19°. Many of the inclination and phase angles for both constituents are resolved to within 10° of the observations even when they do not fall within the 95% confidence limits. The model resolves the semimajor current speeds nearly equally well for the two constituents (M2 mean error is 33%; K1 mean error is 43%), expressed as a fraction of the RMS error to the mean amplitude of the observed semimajor axis. On the basis of the confidence limits, the model fully or nearly resolves both M2 and K1 currents at some sites (e.g., P1-F5 and H4), but has trouble at other nearby locations (e.g., F6 and NC25A). The M2 transition from counterclockwise (CCW) rotation just south of the island to clockwise (CW) rotation farther south places F6 and NC25A in a transitional regime with nearly degenerate ellipses. At F6, the model underestimates the M2 vertically averaged currents by 6 cm/s, although only 15 km away the model predicts a semimajor axis that is within 1 cm/s of the F6 current. With its high horizontal resolution the model has the ability to resolve a wide dynamic range over a short spatial scale, underscoring the importance of having an accurate bathymetric grid. The larger standard deviation of phase at some sites (e.g., F6 for M2 at 37 m) is an indication that the tidal signal can be nonstationary. To fully resolve the tidal currents at such locations may require in part a model that includes physics beyond the vertically averaged case (e.g., inclusion of seasonally varying density structure (see sections 3 and 4 for further development of this aspect)).
 The three historical measurements to which we can make a statistical comparison fare poorly compared to more recent measurements: none of the semimajor axes fall within the confidence limits, and only a few of the other parameters are resolved. Historical M2 semimajor currents, with the exception of LD2, are smaller than the computed ellipses. Mooring NC25A has only half the semimajor axis M2 amplitude of moorings F8 and F4 (spaced < 40 km apart) even though it was located midway between these two other sites and in a region of gently sloping bottom relief. Similar differences at NC25A are seen in the diurnal constituent. Perhaps the other historical results are suspect as well, but we have no other mooring results to provide another perspective at or near these locations. Rather than rejecting any ellipse parameters as necessarily faulty, we leave it to the reader to make their own assessment of the utility of the model results and/or observed parameters.
 Divergence between the modeled and observed ellipse parameters may be due in part to (1) inaccurate bathymetric grid employed by the model, (2) density structure is absent from the model, (3) location of the recording instrument within the bottom boundary layer (often so placed to observe dense water and to avoid winter ice keels), and (4) the mismatch of comparison between a vertically averaged parameter and a measurement at one particular depth. We will show in sections 3 and 4 that the latter three sources of uncertainty should be particularly strong for the M2 constituent. The first source of uncertainty, deficiencies in our current realization of the true bottom relief, may be the most limiting factor to model accuracy for both constituents.
 The resolved ellipse orientation is generally parallel to the bathymetric contours and ellipses are typically more rectilinear for locations closer to the shore and in the straits. The model delineates regions of stronger currents in the St. Lawrence Island domain, in particular off of Southeast Cape and to a lesser extent off of Chibukak and Northeast Capes. The stronger tidal currents of these locations are probably due to shallower bathymetry which conforms to the cape structure. It is interesting that Anadyr Strait, located between the island and Siberia, strongly controls the tidal motion both in the semidiurnal and diurnal bands by changing the flow from elliptical to rectilinear, whereas off of Northeast Cape in Shpanberg Strait K1 appears much more elongated than M2. The K1 wave has approximately twice the wavelength of the M2 wave so it is not surprising that the K1 ellipses are less able to exhibit rotational character within the straits. The wider nature of Shpanberg Strait (∼200 km) apparently allows for more rotation of the M2 wave than does Anadyr Strait (∼75 km). However, the model does not properly capture the low-eccentricity nature of the M2 tide as observed at the easternmost two mooring locations. We ascribe these differences to the probable causes noted above and point out that approximately 25 km to the east of LD2 the model does predict more circular ellipses.
 In summary, the spatial variability of the tidal ellipses over the course of our domain depict a dynamic region that is strongly influenced by the coastlines and bottom topography. Both M2 and K1 ellipses exhibit a range of speed that varies by factor of five and can be either highly rectilinear or have no eccentricity whatsoever. Despite the limitations inherent within both the vertically integrated model and the measurements, many computed ellipse parameters turned out to be in agreement with those obtained from the harmonic analyses, thus implying that a barotropic description provides a first-order estimate of the tidal regime in this region.
3. Structure of Temporal and Vertical Variability
3.1. Mean Vertical Distribution
 To examine the primary characteristics in the tidal current distribution along the vertical direction we describe yearlong analyses at mooring F8. Structure of the tidal currents through the water column based on the annual analysis for all depths at mooring F8 is shown in Figure 4. Confidence limits for the yearlong analyses are taken from the T-TIDE output. M2 tidal currents are relatively constant (15 ± 2 cm/s) from the surface down to approximately 15 m depth, depict a slight maximum around 20 m, and then decay to the instrument depth of 43 m. This character is seen in both the semimajor and semiminor axes. The direction of rotation for the M2 ellipse is negative (CW) and only in close proximity to the instrument the direction changes to positive rotation (CCW). This reflects that in the entire layer of decreasing current magnitudes, progressing toward the bottom, the sense of rotation tends more and more to the CCW direction. The M2 current ellipse only slightly changes eccentricity in the surface layer, as both major and minor axes are relatively constant. Below 25 m, the eccentricity decreases until at about 10 m above the bottom the ellipse becomes rectilinear, indicative of the change in sense of rotation. The M2 inclination and phase both are constant in the upper 15 m of the water column then undergo a linear decrease of about 15° to the instrument depth.
 In contrast to the M2 currents, the K1 tidal current semimajor and semiminor axes are invariant over the course of the water column. The phase and inclination of K1 progress from the surface to the instrument depth over a range of about 20° in an opposite sense to the change of the M2 parameters. The diurnal BBL must be located below the deepest measurement depth, i.e., 6 m above the bottom. Mean variations through the water column are in the range of ±10° and ±0.5 cm/s. The direction of rotation for the K1 ellipse does not change from CW throughout the measured water column.
3.2. Seasonal Variability
 For the two moorings with profiling ADCPs (F6 and F8), we also computed overlapping monthly tidal harmonic analyses for the yearlong series at each depth bin. This allows us to assess the seasonal character in conjunction with the vertical structure for a comprehensive description of the tidal currents. Figure 5 describes the yearly cycle of the M2 tidal currents from F8 and shows the development of a nonuniform vertical distribution of currents during the summer. The character of the M2 tidal currents changes over the course of the year as the water column progresses between the cold, salty, ice-covered winter conditions between January and March and the two-layer, sharply stratified conditions of July through September. For M2, the time periods of October to December and May to June are transition periods between these two extremes. The main periodicity is a yearly cycle with weaker winter currents and stronger summer currents. The monthlong analyses reveal that the M2 currents varied from as little as 7 cm/s (near the bottom in the winter) to as much as 21 cm/s (near the surface in the summer). Minimum values were observed near the bottom during all months, the smallest values occurring between January and March. Tidal currents were fairly constant between the surface and about 25 m during this time. During this period of ice cover, we do not see a near-surface boundary layer of tidal current attenuation similar to that observed near the bottom. This is one indication that the pack ice is not significantly influencing the tidal current profile.
 During the stratified period (approximately June through September) at mooring F8 a subsurface maximum of the M2 currents occurred between 20 and 25 m, whereas at F6 (just 20 km away) the M2 currents are maximal at the surface and decay throughout the water column toward the bottom. From March to October, ellipse inclination and phase both are constant in the upper 20m and then decay from 20m to the bottom. This describes the tendency of ellipses, with decreasing distance to the bottom, to be aligned progressively more in the CCW sense. From November through April the phase and inclination have a smaller range of values through the water column.
 The diurnal tidal currents are smaller than the M2 currents and exhibit markedly different seasonal character. K1 does not have an annual modulation. Constituent K1 (T = 23.934522 h) is not resolved from constituent P1 (T = 24.065873 h) in a 29-day analysis, so instead P1 was inferred from K1 using the relationship determined by the yearlong analysis [Foreman and Henry, 1989; Pawlowicz et al., 2002]. Linear superposition of K1 and P1 creates a beat period of 1/2 year [Pugh, 1987]. In the St Lawrence Island region, the manifestation of this semiannual period (Figure 6) results in approximately 2.5 cm/s difference between February/August (2.5 cm/s) and November/May (5 cm/s). Although 5 cm/s is not a particularly energetic flow, it is of the same order of the mean background flow (3.6 cm/s) at NC25A [Schumacher et al., 1983] and is approximately 40% of the annual mean tidal flow and 14% of the peak tidal flow. The major seasonal cycle for the diurnal currents is not related to the annual cycle of density that dominates behavior of M2, but the semiannual cycle caused by the linear superposition of the K1 and P1 constituents. Beating of close constituents also occurs for semidiurnal tides. We will show later that M2 and N2 exhibit a monthly beat that dominates the total tidal energy profiles.
3.3. Rotary Spectra
 Tidal current parameters were also evaluated based on rotary spectra analysis [Gonella, 1972; Mooers, 1973; Emery and Thomson, 2001]. Figure 7 depicts the rotary spectra for the 10 m and 43 m depths at mooring F8. This analysis shows that the energy of semidiurnal tides dominates both the upper and lower water column. The contribution of diurnal tides to the total energy is smaller, although the second strongest maximum in the energy spectra belongs to the K1 tide. Tidal maxima show slight seasonal variability, but it is not the strongest feature of the tidal regime depicted by the rotary spectra.
 The feature that sets apart the tidal currents in the upper and lower water column are the CW and CCW components of rotation. At 10 m depth, the energy of the CW component at the M2 period is 2 to 3 times greater than the energy of the CCW component, while at 43 m depth the CCW component slightly dominates. The same behavior, i.e., prevalence of CW motion at the surface and growth of the CCW motion toward the bottom can be discerned for both seasons as well as the other semidiurnal tides, cf., S2 and N2. The ratio of CW to CCW energy (approximately 3.5) in the diurnal band does not appreciably change with depth or season.
 The broad inertial oscillation maximum of CW currents (Figure 7c) exhibits strong seasonal variability. Amplification of near-inertial clockwise currents in the upper layer of the water column during the summer period is seen in this figure. Wind-generated motion supplies energy only through the surface layer of the ocean in the wide frequency band around the inertial frequency. The strong summer stratification appears to limit the enhancement of near-inertial motion to the upper layer. Figure 7c also suggests that the summer enhancement is not limited to inertial motion, as the clockwise currents at the N2 frequency have been magnified with respect to the N2 currents shown in Figure 7a. However, the noise floor of the semidiurnal band in the near-surface winter spectrum (Figure 7a) appears depressed with respect to the other three spectra, so such characteristics are difficult to assess. In winter, there is no near-surface maximum in the power spectra around the inertial period of oscillation. One may be tempted to speculate that a possible cause of the diminished wintertime inertial-band energy is reduced air-sea coupling due to sea ice cover, but the vast majority of ice in the northern Bering Sea is in free drift and the wind stress is larger in the winter than in the summer. Maas and van Haren  demonstrated a dramatic increase in energy of the inertial oscillations, in the central part of the North Sea, from the unstratified winter conditions to the stratified summer conditions, despite the lack of ice cover during the winter and the wind-forcing during the two seasons is not significantly different.
 We have described in sections 3.1–3.3 the temporal and vertical structure of tidal currents at this one location on the northern Bering Sea shelf. In the next section we use numerical models to explain observed differences between the M2 and K1 currents.
4. BBL Structure for the Subinertial and Superinertial Tides
 In this final section, the vertical velocity structure and rotational characteristics described previously will be examined numerically. We seek an explanation of the different behavior of the M2 and K1 currents through the water column by deconstructing the different BBL regimes for the subinertial and superinertial tides. We do this by outlining the pertinent BBL dynamics and then looking into effects of density stratification and constant and variable eddy viscosity coefficients upon the CW and CCW rotating components of currents.
 Equations for boundary layer dynamics, along with the analytical solutions are given in Appendix A. The BBL for the rotating (tidal) flow, although similar to the Ekman BBL concept, differs strongly in magnitude. The magnitude of the thickness δ of the BBL can be evaluated from the exponent in the solution (A13), assuming decay of and is equal to e−1.
 Here Nz denotes the eddy viscosity coefficient, ω is the tidal frequency, and f is the Coriolis parameter. Negative signs represent clockwise rotation and positive signs represent counterclockwise rotation. At latitude 60°N the disparity between the two scales at the M2 frequency is quite strong, since δ−/δ+ ≃ 4, but at the K1 frequency this disparity diminishes to δ−/δ+ ≃ 2. In Figure 8 the vertical distribution of the magnitude of velocity is shown for the CW and CCW motions, as obtained from the analytical solution (A14). Motion is considered in shallow water of 50 m depth and is imparted at the bottom by a circular current whose components for the clockwise motion are u0 sin(ωt) and v0 cos(ωt). The current magnitude is 10 cm/s and is equal to . We consider circular motion so that the input into CW and CCW motion is the same and therefore the differences in the boundary layer thickness are not induced by unequal forcing.
 The first case considered is with constant eddy viscosity set to 50 cm2/s and the latitude set to 60°N. The M2 and K1 velocity profiles are generated with circular motion imparted at the seafloor (Figure 8). Profiles of the CW motion and CCW motion are both forced with the same magnitude. Strong disparity in the behavior of the CW and CCW motion exists. Because the M2 wave is located in the superinertial frequency domain and its period = 12.42 hr) is close to the inertial period (Ti = 13.82 hr), while the K1 wave is located in the subinertial domain and its period = 23.93 hr) is much larger than the inertial period, the thickness of the BBL for M2 flow is greater than for K1 flow.
 A few conclusions follow from Figure 8 and from the solution given in Appendix A. Comparison of the M2 profiles shows that the dominant component of velocity at the free surface is rotating clockwise, while close to the bottom the CCW rotation is stronger. The relation between M2 and K1 components can be deduced from Figure 8 as well. Large differences are between clockwise rotating currents: M2 tends to dominate in the surface layer, but K1 is stronger in the bottom layer.
 As suggested by Prandle  the properties of the BBL are further elucidated by the use of a variable eddy viscosity (see equations (A17) to (A20)). The results of this scheme, based upon turbulent energy closure, are not radically different from previous studies employing piecewise linear eddy viscosity [e.g., Fjeldstad, 1936; Maas and van Haren, 1987; Souza and Simpson, 1996; van Haren, 2000; Howarth, 1998]. To investigate the differences between constant and variable eddy viscosity coefficient, the vertical exchange of momentum is first considered in a uniform (unstratified) water column. Results of numerical computations for this case are shown in Figure 9a, for the M2 wave and 9b for the K1 wave. All parameters are defined for the geometry considered in Figure 8. For comparison, previous results with constant eddy viscosity are given as well. The influence of the depth-dependent eddy viscosity is expressed in a narrowing of the BBL, both for the semidiurnal and diurnal tides and for the clockwise and counterclockwise rotations. As before, the motion imparted at the bottom influences the higher levels of the water column mainly through the clockwise M2 component. With a variable eddy viscosity, the magnitude of the CW component of M2 current is still growing even at 50 m from the bottom. The CW motion in K1 component reaches constant magnitude below the free surface. The vertical distribution of the eddy viscosity coefficient for the CW and CCW rotation of the M2 wave is shown in Figure 10. The velocity distribution given in Figure 9 influences the vertical distribution of the eddy viscosity. For CW motion the maximum value of the eddy viscosity is greater and is located higher in the water column than the maximum for CCW rotation.
 A typical density profile south of St. Lawrence Island at the late summer shows very strong fresh water and solar heating influence (Figure 11a). The water column structure depicts two-layers of well-mixed density which are divided by a narrow pycnocline. The density stratification resulted in the new profile of the eddy viscosity (Figure 11b). Owing to the density stratification, turbulence generated in the BBL has been strongly reduced and at about 20 m above the bottom it is completely damped. Therefore the eddy viscosity changes only in the near-bottom region and above 20 m from the bottom the eddy viscosity is reduced to a constant, near zero value. Such distribution of the turbulence generates a velocity profile that has a small maximum at the bottom of the pycnocline and constant magnitude above (Figure 11c). The presence of a tidal current maximum for the M2 constituent in the pycnocline region has been noticed in earlier observations [Fjeldstad, 1936] and the explanation invokes the dynamics of the rotational boundary layer in a stratified fluid [Maas and van Haren, 1987; Howarth, 1998; van Haren, 2000].
 The influence of the density stratification on the K1 wave is negligible. The different behavior of the M2 and K1 waves is related to the different thickness of the BBL for these two waves. The boundary layer thickness of the K1 wave is smaller than the distance to the pycnocline, therefore the pycnocline region is encountered by the constant velocity profile without any gradients. On the other hand, the boundary layer for the M2 wave is greater than the distance to the pycnocline. The M2 wave profile exhibits a marked maximum in the pycnocline region. Since the CW constituent of M2 has the largest thickness of the BBL it also depicts the strongest influence by the density stratification. Above the pycnocline the eddy viscosity is strongly damped, therefore in this region the vertical structure exhibits nearly constant velocity. This is in contrast to the velocity profile computed for the uniform density in the water column (cf. thick dashed line in Figure 9a). The key factor for the generation of different velocity profiles appears to be the location of the pycnocline in relation to the BBL. Both observations and theory point out that the changes in the vertical profiles will be dominated by the CW component of M2 [Souza and Simpson, 1996; van Haren, 2000].
 The above investigations were made for circular motion where the relative input of the CW and CCW motion was the same. To consider the possibility of generating an elliptical CCW motion in close proximity to the bottom by elliptical CW motion imparted at the surface we take the surface currents from the F8 mooring and use them as the forcing function for the boundary layer currents. In Figure 12 we compare the results of this final numerical experiment with the current ellipses observed at F8. We consider motion generated by the M2 surface current whose components are u0 cos(ωt − gu) and v0 cos(ωt − gv). Amplitudes and phases of the current are: u0 = 14.0 cm/s, gu = 239.33°, v0 = 5.3 cm/s, gv = 88.90°. The density distribution for this computation is the same as in Figure 11. We point out that the actual density distribution at F8 for this period is unknown.
 The M2 tidal current ellipses have been plotted in Figure 12 at four levels (distances are given from the sea surface). In the surface layer (10 m), near the pycnocline (24 m), below the pycnocline (36 m) and close to the bottom (44 m). While in the upper layer (10–15 m from the surface) the current ellipses conserve shape, magnitude, and sense of rotation. In the lower layer the ellipses change magnitude, shape, and sense of rotation. The clockwise motion dominates in the entire water column with an exception of the narrow bottom layer where the M2 current rotates counterclockwise. The decrease of the current toward the bottom is associated with the changes in the ellipse shape. Close to the bottom, due to the different decay rate of the CC and CCW components, the flow becomes more rectilinear. Especially interesting is the behavior of the tidal current ellipses in the pycnocline region where a maximum of the current occurs both in observation and model, although the model predicts smaller values than the observed ones.
5. Discussion and Summary
 The main purpose of this study was to examine the tidal current regime in the vicinity of St. Lawrence Island. We aimed at explanation of the tidal velocities and their spatial and temporal variations through the water column, across the domain and in both the summer and winter seasons. The basic observations can be summarized as the following: (1) Semidiurnal currents dominate over diurnal currents. (2) Many tidal current parameter estimates from harmonic analyses are in agreement with parameters computed from a vertically integrated model, confirming the primarily barotropic nature of the currents. (3) The horizontal distribution of tidal currents is one of bathymetrically and frictionally steered current ellipses across the domain. Current speeds have a range of more than fivefold in both the M2 and K1 constituents. (4) Structure of the tidal currents through the water column is different for the semidiurnal and diurnal constituents. Semidiurnal currents change along the vertical direction while diurnal currents are nearly constant from just above the bottom to the free surface. (5) Rotary spectra analysis shows large differences in the diurnal and semidiurnal tidal bands. The semidiurnal tides show prevalence of CW motion near the surface and growth of the CCW motion toward the bottom, whereas the ratio of CW and CCW energy in the diurnal tidal band does not change with depth.
 We show that these peculiarities in the vertical behavior are related to the BBL in the tidal flow regime of approximately 50 m depth. Here we use an approach suggested by Visser et al. . Instead of the analytical solution for the two-layer model that they used to examine tidal flow in very shallow (20 m) water, we use a numerical solution to the set of equations of motion together with turbulent closure models. This approach allows us to examine a tidally generated BBL in interaction with variable eddy viscosity. A similar approach to the investigations of the BBL was suggested by Marchuk and Kagan  and Mofjeld et al. . Even the simplest model that describes tidally generated turbulence by constant eddy viscosity points out that the differences observed in the M2 and K1 velocity profiles are related to (1) different frequency of oscillations and their relation to the inertial frequency, and (2) different sense of rotation (see equation (1)). The BBL for the clockwise rotating M2 currents extends from the bottom to the free surface, i.e., over the entire depth of the 50 m water column, while the BBL for the K1 currents is limited to 15 m for CCW rotation and to 20 m for the CW rotation. Considerations with a variable eddy viscosity without density stratifications describe similar distributions of currents with a much thinner BBL, limited to 10 m, except for the CW rotating M2 current whose BBL extends from the bottom to the surface.
 Numerical experiments with two-layer density stratified flow shows that the only tidal mode interacting with the pycnocline is the one whose BBL is thicker than the distance from the bottom to the pycnocline. There is only one mode in our experiment set with such a large BBL, i.e., the semidiurnal mode with CW rotation. Through interaction with the pycnocline, this wave organizes a local maximum in the velocity distribution. M2 is therefore the most likely constituent to induce vertical mixing by generating a region of enhanced velocity shear in the thermocline. Observations near St. Lawrence Island indicate that the change in the vertical profile of the semidiurnal tidal current is connected to the summer stratification (Figure 4). It follows from the rotary spectra (Figure 7) that at the same period a strong inertial motion develops. As processes of maintaining the summer stratification are controlled though a balance of tidal and inertial motion shear [van Haren, 2000] a picture of interaction between tides and inertial motion becomes more complicated. Vertical viscosity which modifies the tidal currents through dependence on stratification is related to the intensity of inertial motion as well.
 Numerical computations based on observational data from mooring F8 help confirm the influence of the BBL on the tidal current distribution around St. Lawrence Island. The M2 currents were computed in a stratified water column forced by the observed surface current. The resultant tidal currents show transformation from CW motion high in the water column to CCW motion in close proximity to the bottom. The observed disparity in the BBL regime of the diurnal and semidiurnal tides may be important for the understanding of exchange and transport processes driven by tides in close proximity to the bottom. Although semidiurnal currents dominate the dynamical processes at the ocean surface and at middepths, close to the bottom an exchange of mass and momentum may be driven by the diurnal currents.
 The striking differences between diurnal currents which are nearly constant throughout the entire water column and decrease of the semidiurnal currents in the thick BBL are not limited to the Bering Sea or to other shallow seas of fresh water influence [Maas and van Haren, 1987; Visser et al., 1994; Souza and Simpson, 1996], they were also observed in the southern Weddell Sea near the critical latitude by Foldvik et al. . At one of the mooring stations in the Weddell Sea the BBL thickness of the CW and CCW components at M2 frequency was estimated to differ by a factor of 15.
 Unlike regions that experience shorefast ice where the surface tidal currents are very small due to an immobile boundary as observed by Sverdrup , the observations near St. Lawrence Island are made below drifting sea ice. The amplitude of both the 10 m M2 and K1 currents (Figure 7) do not appreciably change from summer to winter, and the vertical distribution of the current magnitude (Figure 5) during winter shows very small near surface attenuation. Thus the tidal waves are not significantly damped by sea ice cover. The pack ice of the northern Bering Sea is in free drift and it behaves as a flexible membrane which weakly damps vertical motion but resists horizontal convergence. Therefore it tends to generate strong nonlinear motion. Observations of ice drift and deformation in the Bering Sea indicate nonlinear ice-water coupling in the tidal band of oscillations since the M4 tidal component of the ice velocity is relatively much stronger than the M4 in the ocean currents [Pease and Turet, 1989].
 Following Maas and van Haren  and Howarth  we have excluded free internal tides as the source of changes in the tidal current distribution. The mooring stations around St. Lawrence Island were placed over smooth topography located about 500 km from the shelf break, where free internal tides are usually generated. Owing to the intermittent character of the internal tides and the long distance of propagation from the likely source region to the mooring sites these waves will not be phase locked to the surface tides, and therefore they will tend to be filtered out through the tidal analysis procedure.
 Energy delivered to the system by tidal currents accounts for anywhere from less than 20% to greater than 80% of the total (Figure 13). The annual mean is close to 50%, larger than the expected mean of 10–40% given by Coachman  for the region immediately to the south of St. Lawrence Island. One reason for the previous underestimation of the fraction of kinetic energy due to tides is that the current meters were typically placed in close proximity to the bottom, in part to reduce risk from ice keels. Thus the measurements have been located within the M2 BBL and have underestimated semidiurnal currents. The tidal kinetic energy comes in packets with an approximately 1 month beat, the period of which is set up by superposition of semidiurnal constituents M2 and N2. These packets have the greatest energy in the summer when a density modified semidiurnal tide is generated. The beating is also enhanced during periods of constructive interference between diurnal constituents K1 and P1, a process that undergoes a semiannual modulation.
 There exists potential for the effects of this waxing and waning of tidal energy in conjunction with the 18.6-year tidal cycle to influence biological processes from monthly up to interannual and interdecadal scales. For instance, a strong and sustained phytoplankton bloom might optimally develop under periods of calm winds, strong insolation, and large tidal currents, with tidal currents above some threshold value providing nutrient resupply to the euphotic zone through mixing. In any given year these three processes may or may not simultaneously occur. The connection between monthly to decadal periods of tidal current modulations and the biota of the continental shelf ecosystem remains largely unknown. However, the tidal forcing ought to be important for such connections since tides give the largest input to the energy budget of the Bering Sea shelf.
Appendix A:: Model for the Tidal BBL
 The first step in constructing the equations for the tidal bottom boundary layer (BBL) is to assume that velocity is a combination of the zeroth-order geostrophic solution and the first order solution, which represents changes caused by friction [Munk et al., 1970]. For the zeroth-order, the assumption is that the horizontal velocity does not vary along the vertical direction,
The first-order velocity is defined by
The total velocity u = u0 + u1 and v = v0 + v1 away from the bottom will be expressed by u0 and v0 only, while at the bottom, due to friction, u = 0, and v = 0. Therefore the boundary conditions for the first-order velocity are
Thus the above condition states that the motion for the first-order solution is imparted at the bottom. Thorade  [cf. Defant, 1960; Mofjeld, 1980; Prandle, 1982] demonstrated that the vertical scale of the BBL depends on the tidal ellipse rotation, and for the clockwise (CW) and counterclockwise (CCW) rotation the two different scales occur. Following Sverdrup  the clockwise q and counterclockwise r components are defined as
With the above definition the Cartesian velocity components in terms of the CW and CCW components can be written as
Equations of motion (A3) and (A4) can be redefined for q and r components by multiplying the second equation by i and taking a sum and a difference,
Here q1 = u1 + iv1 and r1 = u1 − iv1. The boundary conditions (A5) need to be reformulated for the rotating components, and the boundary condition at the free surface ought to be changed since the distance from the bottom to the free surface is finite. At the free surface we assume a zero stress condition,
At the bottom the total velocity vanishes and the first-order velocity is defined by the zero-order velocity
Here q0 = u0 + iv0 and r0 = u0 − iv0. Solution to equation (A8) and (A9) for the constant eddy viscosity coefficient Nz and for all dependent variables changing in time as e−iωt can be written as [Mofjeld, 1980]
The total velocity is defined as
Description of the vertical exchange of momentum in the uniform and density-stratified water columns is achieved better through the depth-dependent eddy viscosity. The first step is to solve equations for temperature (T) and salinity (S)
to define the density (ρ) distribution along the vertical direction.
 In the eddy viscosity parameterization we follow the approach by Galperin et al. . The eddy viscosity (Nz) and eddy diffusivity (Dz) are computed according to
where l is the mixing length, q the square root of twice the kinetic energy, and SM and SH are functions of the vertical stratifications. The definition of these functions is given by Blumberg and Mellor  and Galperin et al. . The equation for the turbulent energy
and the equation for the length l
complete the turbulence closure. Here B1 = 16.6, E1 = 1.8 and E2 = 1.33. defines a wall proximity function, see formulas (13) and (14) in the work of Blumberg and Mellor .
 First and foremost, the authors wish to thank the investigators whose data were graciously made available for the analyses in this study and who offered many helpful suggestions to early drafts of this manuscript. The bulk of the data used in this manuscript are from a study designed to investigate dense water formation within the St. Lawrence Island polynya and the subsequent migration away from the polynya area. Knut Aagaard, Jens Meincke, Detlef Quadfasel, and Tom Weingartner all provided RCM data. Special appreciation is directed to Tom Weingartner, who provided the ADCP data and offered much guidance with the analyses. For the field program, thanks are extended to mooring technicians Jim Johnson, David Leech, Andreas Welsch, and Ulrich Druebbisch as well as the captain and crew of the R/V Alpha Helix, all of whom contributed to the successful deployment and recovery efforts. We thank the residents of the village of Gambell, who assisted us on a number of instances during both the deployment and the recovery field campaigns. Finally, we would like to thank both anonymous reviewers for their comments and suggestions to help make this a stronger manuscript. Seth Danielson was supported through NSF grant OCE 9730697.