Kugmallit Valley spans the midsection of a transect across the Mackenzie Shelf of the Beaufort Sea. Its greatest relief is 20 m, and its width is 20 km. Using a yearlong record of ice drift, ocean current, temperature, and salinity acquired near the center of the valley, we describe a pattern of flow that is correlated with wind stress and ice motion and discuss its similarity to flow within larger submarine canyons that cut through the shelf break. As in such canyons, there is enhanced cross-shelf transport within Kugmallit Valley during upwelling-favorable surface stress. The data also document the down-valley flow of dense water from a flaw lead.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 There are several shallow valleys that cross the Mackenzie Shelf of the southern Beaufort Sea (Figure 1). These features are relict channels of the Mackenzie River, dating from times of lower sea level during the ice ages. Since these valleys are shallow and do not extend as far as the shelf edge, there is uncertainty as to whether they enhance cross-shelf exchange of water in response to upwelling- and downwelling-favorable winds as do deeper canyons that cut though the shelf edge [Hickey, 1997; Allen, 2004]. Should they do so, these valleys may be effective pathways for supplying deeper, nutrient-rich water from the outer shelf to the photic zone near shore.
 Kugmallit Valley, which has the greatest relief, is the focus of this paper. The valley begins at the 25-m isobath and runs northward to past the 70-m isobath, where it fades out about 25 km short of the shelf break (Figure 2). The shelf break is at about 85-m depth. Where Kugmallit Valley is most pronounced, the valley bottom is about 20 m below its eastern rim. The valley is also asymmetric: The slope of the eastern wall (20 m/km) is much steeper than that on the western side (1 m/km).
 The geographic width of Kugmallit Valley is not necessarily an indication of its width in relation to ocean dynamics. An important measure of width relevant to fluid dynamics is the ratio of physical width to the Rossby radius of deformation, NH/f, where N is the buoyancy frequency, f is the Coriolis parameter, and H is the depth of the canyon [e.g., Klinck, 1996; Hickey, 1997]. For Kugmallit Valley near site 1, f = 1.37 × 10−4 s−1; N, estimated from conductivity-temperature-depth (CTD) casts, ranges over 2.3–3.2 × 10−2 s−1, depending on the season and on what part of the water column is considered; H is either 20 m (the relief of the valley) or 60 m (the depth of the valley near site 1). With these values for f, N, and H the Rossby radius is in the range 3.5–14 km. Since Kugmallit Valley is 10–20 km wide, its width is 1–6 Rossby radii.
 Another parameter is the ratio of canyon depth at the shelf break to the depth of the adjacent shelf break. Here we modify the definition slightly, using the dimensions of Kugmallit Valley at site 1: 60 m deep with the eastern rim rising to 40-m depth. The ratio is only 1.5. In comparison, Astoria Canyon [e.g., Hickey, 1997] is 4–8 times narrower than the Rossby radius of deformation and 3 times deeper than the shelf break. Therefore Kugmallit Valley is shallow and moderately wide. Also, because the depth is only ∼60 m, surface and bottom boundary layers could occupy a significant proportion of the water column.
 Using a yearlong record of ice drift, ocean current, temperature, and salinity acquired near the center of the Kugmallit Valley, we describe a pattern of flow that is correlated with wind stress and ice motion and discuss its similarity to flow within larger submarine canyons that cut through the shelf break. The data also document the down-valley flow of dense water from the flaw lead in late winter.
2. Data and Methods
 Instruments along Kugmallit Valley were operated on short taut line subsea moorings from September 2002 to September 2003. Site 1 (70°20′N, 133°44.4′W), near the middle of Kugmallit Valley, was in 57 m of water; site 9 (70°5.0′N, 133°30.0′W) was on the 35-m isobath in the upper valley; and site 2 (70°59.3′N, 133°45.0′W), in 116 m of water, lay beyond the mouth of the valley on the upper continental slope. The instruments providing the data used here were upward looking acoustic Doppler current profilers (ADCP) (RD Instruments narrow band at 300 kHz) positioned 5 m above the seabed, temperature-salinity recorders (Sea Bird SBE37) 2 m above the seabed, and pressure recorders (Paroscientific sensor within ice-profiling sonar, ASL Environmental Sciences model IPS4) 5 m above the seabed.
 The ADCP at site 1 sampled at 30-min intervals within 10 depth bins spanning the water column between the sonar and the zone of interference from surface echoes; the bins were centered at depths of 11.1, 15.0, 18.9, 22.8, 26.7, 30.6, 34.5, 38.4, 42.3, and 46.2 m. The bin depths are indicated on the cross section of Kugmallit Valley in Figure 2a; note that the deepest bin is 10 m above the seabed and so just above the most dramatic valley topography. The data were low-pass filtered (half amplitude at 33-h period) to remove variance dominated by tides and inertial waves. The SBE37 sampled seawater temperature and conductivity at 15-min intervals, and the IPS4 sampled pressure every 6 min.
 Data are also shown from moorings CA1, deployed in 63 m of water ∼70 km up shelf (WSW) of Kugmallit Valley, and CA2, deployed in 66 m of water ∼35 km down shelf (ENE) of Kugmallit Valley. The velocity, temperature, and salinity data shown for these two moorings are from Aanderraa RCM7 instruments located approximately 14 m above the bottom: 49.1 m deep at CA1 and 51.4 m deep at CA2. Data were recorded hourly.
 To estimate the stress from wind on the ocean surface, we use a flat plate formulation and assume ice-free conditions: τwind = ρaCDair-water ∣U∣U, where ρa (1.2 kg/m3) is the density of air, CDair-water (0.0015) is the 10-m drag coefficient [Fairall et al., 1996], and U is the wind velocity (m/s). We used the wind velocity at 10-m elevation, provided every 6 h by National Centers for Environmental Prediction (NCEP), for this calculation and averaged values from 10 grid points surrounding Kugmallit Valley (see Figure 1). We favor NCEP winds over observations made at coastal stations (e.g., Tuktoyaktuk) or on offshore islands (e.g., Pelly Island; see Figure 1) because of concern that such data were substantially altered by the presence of land [Williams et al., 2006].
 The velocity of pack ice relative to the ocean is the critical determinant of surface stress on the ocean in this area since the ice concentration is high for much of the year [Canadian Ice Service, 2002]. The stress exerted by moving ice on the ocean depends on ocean velocity and stratification, on ice roughness and concentration, on skin and form drag, and on internal hydraulic effects [McPhee and Smith, 1976; Pite et al., 1995; McPhee, 2002]. In addition, the application of drag formulations appropriate for logarithmic boundary layers over rough flat plates, as for wind stress above, is not appropriate beneath the ice, where surface roughness elements frequently extend not only through the log layer but also through the entire planetary boundary layer [McPhee, 2002]. It is common to wrap up the complexity of under-ice flow into a geostrophic drag formulation, τ = ρw CDICE-water ∣UICE-Ug∣ (UICE-Ug) eiθ [Wadhams, 2000], where τ is the stress exerted by the ice on the ocean at the base of the planetary boundary layer, ρw (1020 kg/m3) is seawater density, and CDICE-water (0.0055) is a geostrophic drag coefficient [Hibler, 1980]. In the complex number notation used, UICE-Ug is the difference between the ice velocity and the water velocity below the surface mixed layer, Ug, assumed to be in geostrophic balance, and θ (25°) is a turning angle resulting from Ekman dynamics [Wadhams, 2000].
 Ice velocity was measured by the ADCP at site 1, and these data are representative of drift over much of the shelf since the correlation length scale for ice velocity is about 300 km [Thorndike, 1986]. However, we were not confident that reliable regional estimates of ice–ocean surface stress could be calculated from the data available. Instead, ∣UICE∣UICE was used as a general indication of the magnitude and direction of the ice-ocean stress; that is, the ice velocity (without reference to current) was used as an indicator of surface stress over the Mackenzie Shelf near Kugmallit Valley when ice was present.
 Reference directions in subsequent discussion of currents around the Kugmallit Valley are as follows: The along-valley direction runs along the valley axis (∼350°T, close to true north), with the down-valley direction toward deeper water; the cross-valley direction is in the along-shelf direction and runs parallel to the shelf break (∼52°T or ∼232°T), with the down-shelf direction in the direction of Kelvin wave propagation (toward Amundsen Gulf). Note that stress from wind or ice in the up-shelf direction may cause upwelling via surface and bottom boundary layer processes.
 For site 1 the along-isobath (∼130°T or ∼310°T) and cross-isobath (∼40°T or ∼220°T) directions are also used. Site 1 is located on the western side of the valley, and so these directions are not coincident with the along-shelf and cross-shelf directions. However, a barotropic, geostrophically balanced along-shelf flow is constrained to flow along isobaths, so the along-isobath direction at site 1 is in this sense equivalent to the along-shelf direction. Departure from the along-isobath direction at site 1 is likely due to topographic effects of the valley or boundary layer processes or interaction between the two. Flow at site 1 that predominantly crosses isobaths rather than flowing along them can be considered as upwelling or downwelling.
3.1. Wind and Ice Stress
 Stress at the sea surface is due to the wind when the ocean is ice-free and a combination of wind and ice motion when ice is present. Since wind dominates in forcing ice motion [Wadhams, 2000], it is useful to consider the nature of wind stress even when ice cover is complete.
Figure 3 is a rose diagram for wind stress from NCEP for the yearlong period of observation in Kugmallit Valley. The largest values of stress were applied on a roughly WNW-ESE axis, which is between the along-shelf and cross-shelf directions. Stress toward the WNW was more probable during 2002–2003, so that the 12-month mean wind stress was a (weak) 8 × 10−3 Pa along 269°T, and mean wind stress in the along-shelf direction was 7.5 × 10−3 Pa up shelf, the upwelling-favorable direction. However, there is appreciable year-to-year variation in mean wind stress, as calculated by NCEP; in some years the mean along-shelf wind stress has been downwelling favorable [O'Brien et al., 2006].
 A time series of the along-shelf wind stress is plotted in Figure 4a, where the general predominance of upwelling-favorable stress is clear. The longest periods of downwelling-favorable stress occurred at the beginning of October 2002, in July/August 2003, and in September 2003, when Kugmallit Valley was generally free of ice. Upwelling-favorable wind stress predominated during the ice-covered period. The surrogate for ice stress, ∣UICE∣UICE, is overlaid on wind stress in Figure 4a. Note that the ice tends to move in response to upwelling wind stress but not always during downwelling events, when stress transmitted through the ice from the coast stops its drift [Williams et al., 2006]. Since downwelling-favorable wind stress occurring during periods of ice cover is frequently not transmitted to the ocean, the presence of pack ice tends, on average, to enhance upwelling-favorable surface stress over the Mackenzie Shelf.
3.2. Flow Over the Valley at Site 1
 Stick plots of ocean current for six levels of measurement by the ADCP are shown in Figure 5, and the corresponding velocity roses are shown in Figure 6. In general, the flow becomes faster toward the seafloor: The mean speed is 5.6 cm/s at 11.4 m depth (the top level) and 8.1 cm/s at 46.2 m depth (the bottom level). There are two preferred flow directions: a fast flow, generally up valley (toward the south-southwest, between the up-valley and cross-isobath directions) and a slower flow, generally cross valley and down shelf but with rather variable direction. The up-valley flow rotates southward toward the seafloor and becomes faster until middepth (∼11 cm/s) whereas the cross-valley flow has similar speeds and remains diffuse at all levels (Figure 6). Flows toward the northeast are both slow and rare (Figure 6).
 Because of their geostrophic balance, low-frequency flows over the continental shelf and slope tend to flow along isobaths. The low-frequency flow at site 1, however, displays an asymmetrical cross-isobath flow pattern which is very typical of flows found in cross-shelf canyons [Allen, 2004; Hickey, 1997; Klinck, 1996]. In a cross-shelf canyon the potential vorticity dynamics are asymmetric for up-shelf versus down-shelf flow impinging on the canyon [Williams et al., 2001]. This asymmetry produces different flows within the canyon. An up-shelf flow impinging on the canyon causes an up-canyon flow below the canyon rim which then upwells onto the shelf on the up-shelf edge of the canyon [e.g., Hyun, 2004]. This aspect of canyon flow is similar to the cross-isobath flow toward the southwest found at site 1. A down-shelf flow impinging on the canyon does not cause down-canyon flow but a flow that crosses the canyon [She and Klinck, 2000]. This aspect of canyon flow is somewhat similar to the variable cross-valley flow found at site 1.
 These up-valley and cross-valley flows at site 1 are correlated with the low-frequency along-shelf flows found at ∼50 m deep at nearby moorings CA1 and CA2 (see Figure 1 for location and Table 1 for correlations), indicating that they are part of the larger-scale along-shelf flow. The level of correlation (∼0.7) implies that about 50% of the variance at site 1 is explained by the variance at CA1 and CA2. The correlation of site 1 with site 2 is much lower (∼0.4), which suggests that there are shelf break flows at site 2 which are not present on the shelf at site 1. A coherence analysis (not shown) between site 1 and CA1, CA2, and site 2 reflects the correlations. Coherence between site 1 and CA1 and CA2 is about 0.5–0.6 for time periods of longer than 5 d and then rolls off to about 0.3 at a period of 2 d. The coherence between sites 1 and 2 at these frequencies is much lower at 0.1–0.15.
Table 1. Correlation Coefficients Between the First EOF of the Low-Passed Flow at Site 1 and the First EOFs of the Low-Pass-Filtered Flows at CA1, CA2, and Site 2a
Low-passed flow was 46.2 m deep at site 1, 49.1 m deep at CA1, 51.4 m deep at CA2, and 102.5 m deep at site 2. EOF, empirical orthogonal function.
 Along-shelf geostrophically balanced flows can be generated by along-shelf surface stress. With this in mind, Figure 7 shows progressive vector plots of the site 1 currents, with the deepest level of measurement colored to indicate the along-shelf wind stress. There is good correlation between upwelling-favorable wind stress and up-valley flow and downwelling-favorable wind stress and the cross-valley flow. The correlation indicates that flow in Kugmallit Valley responds to along-shelf flow generated by surface stress in a similar fashion to that found in a cross-shelf canyon, despite the fact that Kugmallit has low relief and does not intersect the continental slope.
 To begin to quantify the relationship between the wind and ice motion and low-frequency flow at site 1, correlation coefficients were calculated between the along-shelf wind stress and along-shelf ice velocity and the first empirical orthogonal function (EOF) of the low-pass-filtered flow of the bottom (46.2 m) velocity bin at site 1. The correlation of the along-shelf wind stress with the flow at site 1 is ∼0.6, but this varies over the year because of wind and ice conditions. Prior to ice formation in autumn of 2002, there is a strong upwelling event followed by a strong downwelling event (see Figure 7), and the correlation is ∼0.75. During ice cover the correlation with the along-shelf wind stress is still 0.6, but the correlation with the along-shelf ice velocity is ∼0.7, possibly reflecting that internal ice stress can prevent the ice from moving in response to the wind. After the spring thaw the correlation with the along-shelf wind stress is again ∼0.6.
 The variation of current with depth at site 1 was very different during the ice-covered and ice-free time periods. During ice cover, there was greater vertical shear in the horizontal velocities which is evident in divergence of the progressive vectors in Figure 7 after the onset of ice cover. The mean speed of the top and bottom velocity bins and the mean difference in direction between the two for ice-covered and ice-free periods is given in Table 2. There is a large reduction in speed and strong anticyclonic turning of the velocity vector during ice cover which is highlighted by the examples of upwelling circulation in Figure 8.
Table 2. Mean Speed of the Flow in the Top and Bottom Velocity Bins From the Acoustic Doppler Current Profiler and the Mean Difference in Direction Between the Two Bins During Ice-Free and Ice-Covered Conditions
11.1 m mean speed, cm/s
46.2 m mean speed, cm/s
Mean difference in direction (top minus bottom)
 An example of the difference in flow during upwelling-favorable surface stress with ice cover and during upwelling without ice cover is shown in Figure 8. In the example with no ice (26–28 October 2002, Figure 8a), upwelling flow is directed between south and south-southwest at all levels, and there is little reduction in speed toward the surface until the topmost level. In the example with ice present (27 February to 1 March 2003, Figure 8b), there is a large reduction in speed, and the velocity turns strongly toward the ice velocity when moving from the bottom velocity record toward the surface. We propose that the difference between the two examples is due to the difference in thickness of the surface boundary layer. In late September, there is typically strong stratification between 10 and 30 m depth at site 1, whereas in late March the water column is homogeneous over the upper 30 m and the stability of underlying water is relatively weak. The likely effect of the autumnal density structure is an Ekman transport confined to the top 10 m, so that shear over the range of depths observed by the ADCP is weak except for the top two velocity bins. If the surface boundary layer during 26–28 October 2002 was ∼10 m thick, the wind stress present would cause a surface boundary layer velocity of ∼14 cm/s roughly toward the north, on the basis of Ekman transport. This speed is comparable to the speed of the up-valley flow but in the opposite direction, so the reduction in speed of the topmost velocity bin (11.1 m depth) is roughly consistent with this basic surface mixed layer estimate. In late winter, there is no such depth constraint, and shear within the Ekman layer extends into the depth range of observation and appears to reach ∼40 m deep. The data suggest a superposition of up-valley flow and the surface Ekman layer.
 During upwelling events the deepest velocity measurements at site 1 show a swift onshore flow directed between the cross-isobath and up-canyon directions. The comments above imply an extrapolation of this swift upwelling flow to the bottom, noting that there is an increase in the flow speed below middepth. Two processes that could reduce the upwelling flow near the bottom are a frictional bottom boundary layer and an offshore (along-canyon) pressure gradient. The deepest velocity measurements are 10 m above the bottom (roughly half the depth of the valley) and do not show influence of the bottom boundary layer. In a cross-shelf canyon the maximum depth of upwelling is roughly the depth at which the along-canyon pressure gradient that is driving flow onshore, up the canyon, is reduced to zero by the along-canyon density gradient [Allen, 2004]. This along-canyon density gradient is due to isopycnals within the canyon tilting upward onshore in response to up-shelf flow. It is clear from the site 9 minus site 1 pressure difference time series (Figure 4c), the time series of along-shelf stress (Figure 4a), and the stratification over the shelf that Kugmallit Valley is not deep enough for this effect to retard the flow near the bottom.
 Kugmallit Valley is dynamically wide as a whole, but it has a steep and narrow down-shelf edge. Up-shelf flow impinging on this side of the valley will stretch vertically as it crosses the steep slope and will generate cyclonic vorticity. In observations of narrow canyons this is how the cyclonic rim depth eddy is thought to be generated [Allen et al., 2001]. If the stretching reached to the bottom of the valley, salinity would decrease at the bottom at site 1 during up-shelf flow events. However, salinity either increases or stays the same during up-shelf flow (see section 3.3), suggesting that upwelling flow at the bottom at site 1 is from farther down the valley and possibly is similar to upwelling flows within narrow cross-shelf canyons, where the upwelling flow is primarily driven by the cross-shelf pressure gradient. The example of upwelling flow under ice-free conditions (Figure 8a) shows upwelling flow at site 1 that has similar speed and direction at all measured depths beneath the surface boundary layer, suggesting that deep upwelling flow within the valley is congruent with rim depth flow at site 1.
 While the data show enhanced cross-shelf transport up Kugmallit Valley during upwelling-favorable wind stress and the pattern of flow is similar to that within cross-shelf canyons, they do not show the final onshore displacement of water parcels. This onshore displacement is a necessary part of assessing the importance of Kugmallit Valley to cross-shelf transport. Note that numerical modeling of upwelling in cross-shelf canyons can show upwelling flow that is onshore within the canyon but then turns offshore over the shelf up shelf of the canyon [e.g., Hyun, 2004]. It appears that the canyon generates a kind of lee wave up shelf of the canyon as the up-shelf advection of the disturbance caused by the canyon is balanced by the tendency for long-wavelength shelf waves to propagate down shelf. It is possible that a similar kind of flow is associated with Kugmallit Valley and, if this is the case, could reduce the final cross-shelf transport of water that has interacted with the valley.
3.3. Near-Bottom Temperature and Salinity
 Temperature and salinity data were collected at 54 m deep (2 m off the bottom) at site 1 (Figure 4). Temperature, salinity, and velocity data were also collected at site 2, CA1, and CA2 (Figure 4) and are used here to place the site 1 data in a broader context. Site 2 is located at 116-m depth on the upper continental slope beyond the mouth of Kugmallit Valley. The T-S measurements made there were 2 m above the bottom. CA1 and CA2 were located on either side of the valley but closer to the shelf break than site 1. The distance to the shelf break at CA1 is ∼10 km, at CA2 it is ∼17 km, and at site 1 it is ∼65 km. The deepest T-S measurements at CA1 and CA2 were 14 m, rather than 2 m, from the bottom and so are less likely to be affected by the bottom boundary layer than at sites 1 and 2.
 Between mid-September and mid-December 2002, site 1 was alternately engulfed at the seabed by warm water of low salinity and cooler water of higher salinity (Figure 9). Upwelling events during this period tended to bring water of higher salinity up Kugmallit Valley and corresponded to upwelling-favorable surface stress. These upwelling events are highlighted by grey bands in Figure 4. In several cases, salinity increased rapidly at the beginning of an upwelling event and then increased more slowly, which suggests an initial adjustment of the stratified water column in the valley to upwelling wind stress that is similar to that found in cross-shelf canyons [Allen, 2004]. Downwelling events tended to lower the salinity at site 1 and correspond to downwelling-favorable surface stress. The largest downwelling event during this period occurred in the second week of October and caused lowering of salinity at sites 1 and 2, but there was little response 14 m from the bottom at CA1 and CA2.
 Overall, between mid-September and mid-December 2002, there is a salinity increase at all four moorings due to predominantly upwelling-favorable surface stress. However, at CA1 and CA2 the increase in salinity has less variation on weekly timescales than at sites 1 and 2, which may be because the instruments there were farther from the bottom and likely to be above the bottom boundary layer.
 From mid to late December, water with a different T-S relationship (of relatively high salinity and warmer) moved over site 1 during down-shelf flow (see Figure 9). Given the general T-S relationship with depth in the Beaufort Sea, it is possible that this anomalous water could be created by mixing between shallow shelf water and deeper slope water. This could occur via horizontal mixing of deep water upwelled onto the shelf at Mackenzie Canyon, a site of known large-amplitude upwelling [Carmack and Kulikov, 1998; Williams et al., 2006]. Mackenzie Canyon is ∼150 km up shelf of site 1, and down-shelf advection during the first half of December could have brought the anomalous water to site 1 (see Figure 7).
 During January and February, there was little variation in temperature and salinity near the seabed at site 1 despite several events of up-valley flow (Figure 4) and evidence for strong upwelling on the upper slope between mid-January and early February at site 2 (Figure 4). We take this as evidence that there is little cross-shelf variation in bottom salinity and temperature at this time over the midshelf.
 Early in March, more saline water at the freezing temperature arrived at site 1, associated initially with down-valley flow in the lower half of the water column and cross-valley flow above this. The likely source is a reservoir of cold salty water created by brine rejection from growing ice within the flaw lead closer to shore [e.g., Melling, 1993]. The mechanism for its sudden arrival at site 1 is discussed in section 3.4.
 Bottom water at site 1 is at the freezing temperature for only about a month. During April through July the salinity decreases slowly to its value prior to March and the temperature increases. There is little evidence via fluctuating bottom temperature and salinity for events of upwelling or downwelling at site 1 despite strong signals in the flow 10 m above the seabed. This is similar to January and February and is again expected to be due to continued low cross-shelf variation in bottom temperature and salinity near site 1.
 With the onset of downwelling-favorable winds in July the seabed salinity starts decreasing in Kugmallit Valley from late winter values, and there is a corresponding increase in temperature. This pattern also occurs at site 2, CA1, and CA2, so that this appears to be a broad-scale pattern.
 Up-valley flow at site 1, associated with upwelling-favorable surface stress, does not always result in significantly higher salinity than at similar depths at CA1 and CA2 during upwelling, so that it remains somewhat unclear whether there is larger-amplitude upwelling in Kugmallit Valley than on the adjacent shelf. However, also note that it is difficult to assign a depth to a particular temperature and salinity over the shelf because of the greater variability of the water masses there compared to deeper on the slope [Williams et al., 2006].
3.4. Down-Valley Flow of Flaw Lead Water
 The ADCP shows a 10–15 cm/s flow down Kugmallit Valley (toward the north and northeast) from 6 to 21 March 2003 (see Figure 5). This event is the only prolonged period of down-valley flow in the velocity time series. It is bottom intensified (largest below 30 m deep) and also coincides with the arrival of dense, salty, freezing point water at site 1 (see the vertical blue band in Figure 4).
 During the winter the flaw lead between landfast ice and the mobile pack ice opens and closes with the ice motion [e.g., Melling and Riedel, 1996; Melling, 1998, Figures 12 and 13]. When open, brackish sea ice is formed in the flaw lead, and concentrated brine is released into the underlying seawater. Convective turbulence driven by the flux of buoyancy from the freezing interface causes the mixed layer to thicken by entrainment of slightly warmer and more saline water within the pycnocline. Thus the mixed layer beneath the flaw lead tends to become thicker and denser as winter progresses. By late winter, the location of the flaw lead is typically between the 20- and 30-m isobaths [Barry et al., 1979; Melling, 1998]; it rarely extends as far offshore as site 1. The arrival of near-freezing salty water at the bottom at site 1, coincident with down-valley flow, is therefore suggestive of a down-valley transport of the dense water from the flaw lead above the head of Kugmallit Valley.
 A pseudotrajectory of ice drift, calculated using ADCP data from site 1, is displayed in Figure 10. Between mid and late February the pack ice was essentially motionless. A rapid movement to the west-northwest (up shelf and offshore) began early on 26 February, opening the flaw lead to a width of 70 km in some places by 4 March. Figure 11 shows a surface temperature image of the Mackenzie Shelf on 4 March when the flaw lead was at its widest. It shows the flaw lead as 70 km wide, in agreement with the ice drift data, and encompassing the entire length of the Mackenzie Shelf, stretching from the middle of Amundsen Gulf to over 100 km west of Mackenzie Trough. At this time the flaw lead is largely filled with fractured grey or grey/white ice, and the region of open water that would contain frazil ice is confined close to the edge of the landfast ice. From Figure 11 the open water region is approximately 3 km wide over the Mackenzie Shelf; earlier images showed the region of open water up to about 10 km wide. Strong upwelling-favorable winds were also present during this upwelling-favorable ice motion. Toward the end of the 4 March the direction of ice drift began backing onshore, toward the south-southeast (Figure 10), ending the upwelling impetus to the water column. The cold outflow through Kugmallit Valley began shortly thereafter, possibly released by the cessation of upwelling circulation. Following closure of the lead on 12 March, there was virtually no movement of the pack for over a month.
 New ice at the outer edge of the flaw lead would have survived for 16 d before being crushed into stamukhi by the onshore ice motion; the lifetime at the inner edge was zero. Since the thickness of 16-d-old ice in late winter is about 0.5 m in this area [Melling and Riedel, 1996, Figure 4], the average growth of ice within the flaw lead during this event was about 0.25 m. This growth would have released about 6 kg/m2 of salt to the water column. If the waters of the flaw lead were a static receiving volume for this salt, bounded by the 20- and 50-m isobaths, the salinity would have increased by 0.17 practical salinity units (psu) during this event. However, the onset signal at site 1 in early March is much larger than this at ∼1 psu. The maximum possible salinity increase can be estimated by considering the maximum rate of ice formation that occurs in the narrow region of open water alongside the edge of the landfast ice near the 20-m isobath. An ice-profiling sonar deployed at site 1 showed that approximately 30 cm of ice formed in 3 d during the flaw lead event (not shown). Because of the nonlinear growth of ice this leads to an estimate of approximately 10 cm of ice growing in 12 h in the open water of the flaw lead. This would have released about 40 kg/m2 of salt to the water column during the 7 d of the lead opening. If the open water region was a static receiving volume for this salt, the salinity could have increased, on average, by at most 2 before the lead began to close on 5 March. This increase is of approximately the correct size, based on CTD casts, to produce an onset salinity signal at the bottom at site 1 of ∼1 psu. However, since the open water region in the flaw lead is quite narrow, the volume of water produced with this estimate is much too small to produce the 10–15 cm/s flow observed that lasted for 2 weeks. A more reasonable explanation of the down-valley flow of dense water is that the waters near the edge of the landfast ice under the flaw lead had been progressively filled with cold brine for some months because of repeated opening and closing of the flaw lead. This has previously been observed over the Mackenzie Shelf [Melling, 1993; Melling and Moore, 1995]. The lack of freezing point water at the bottom at site 1 prior to March and the occasional appearance of a freezing temperature signal at the bottom at site 9 (35-m isobath) as early as January support this contention (not shown). Also, it is too idealistic to presume that the open water of the flaw lead forms a static receiving volume for salt rejected by ice formation. For example, offshore flow in the surface boundary layer, driven by up-shelf and offshore-directed surface stress, will have caused some offshore export of the brine formed in the open water of the flaw lead.
Melling and Lewis  described the flow of dense polynya water across the Mackenzie Shelf using a stream tube model and found, because of the extremely gentle cross-shelf bottom slope (∼0.001) and effects of friction on a thin (10 m) layer of dense water, that the time period for the water to reach the shelf break was of the order of 50 d with flow speeds of about 3 cm/s. With such slow transit speeds the cross-shelf transport of dense water is likely to be dominated by other shelf processes, such as upwelling/downwelling, because of surface stress from the wind and ice. We expect that the dense polynya water is not observed at site 1 prior to 6 March because of the predominance of up-shelf surface stress during the winter, a result of both the wind direction and the blocking of down-shelf ice motion [Williams et al., 2006].
 The swift down-valley flow of dense polynya water occurs on the cessation of 7 d of swift upwelling-favorable ice motion. We speculate that because of high ice/water drag coefficients for rough ice the ice motion caused a period of intense upwelling circulation. This circulation caused the dense flaw lead water, which may have begun to spread across the shelf, to be advected onshore in the bottom boundary layer and to accumulate near its origin under the flaw lead, thus becoming thicker and gaining potential energy. This thick reservoir of dense water would have a tendency to move offshore under the influence of gravity, friction, and baroclinic instability. When the upwelling-favorable ice motion abruptly stopped on 4 March, subsequent adjustment of the dense water then produced the thick, fast down-valley flow observed, though it is not possible to know the dynamics of that adjustment given the data available.
 A similar scenario of cross-shelf dense water transport is modeled by Chapman and Gawarkiewicz , in which they study the effect of a canyon on cross-shelf transport of dense polynya water. In this paper, instability of the front that forms at the boundary of their dense polynya water causes eddying motions to develop. Eventually, an eddy crosses the head of the canyon, and dense polynya water then flows down the canyon in a gravity current, thereby enhancing cross-shelf transport. In a similar study, Kämpf , both numerically and with a rotating water tank, showed how dense water on a shelf will preferentially descend a canyon on its down-shelf side and in so doing drive an upwelling exchange flow on the other side of the canyon. These two idealized studies do not include many factors like the influence of a background ocean circulation (with up-tilting isopycnals) or strong forcing by surface stress. Nonetheless, they provide dynamical evidence that canyons can focus cross-shelf flow of dense polynya water and so leave open the possibility that this is occurring in Kugmallit Valley.
 It is difficult to establish from the observations at hand whether the cross-shelf flow of brine-enriched water was confined to Kugmallit Valley or whether it was a general occurrence on the Mackenzie Shelf at this distance from the flaw lead. Although not in 2003, the wintertime hydrographic surveys of the Mackenzie Shelf reported by Melling  and Melling and Moore  show no tendency for seaward protrusion down Kugmallit Valley of isohalines associated with cold shelf water. The five to six hydrographic sections across the shelf during each survey are fairly congruent and show an apparent but weak tendency for cold brine to move down shelf and accumulate well to the east of Kugmallit Valley. Moorings CA1 and CA2 provide some contemporary along-shelf resolution but are closer to the shelf break than site 1, and their temperature and salinity and velocity data are at least 14 m from the bottom. Mooring CA2, 35 km down shelf of Kugmallit Valley, does not have a dense polynya water event near the beginning of March. Salty, near-freezing-point water arrives at CA2 toward the end of March and is associated with a gradual increase in salinity and no anomalous cross-shelf flow. However, data at CA1, 65 km up shelf of Kugmallit Valley, do show a similar event to the one in Kugmallit Valley and at the same time. This shows that cross-shelf flow of dense polynya water was not confined to the Kugmallit Valley when the flaw lead closed on 4 March. CA1 is located close to the shelf break in a very broad and shallow depression in the shelf topography. It is possible that this topography also focused the cross-shelf flow of dense polynya water. It is also possible that more dense water could have been present at the eastern end of the shelf because of up-shelf advection during the flaw lead event.
 The flow of cold saline water at site 1 was not observed to reach the bottom at site 2. Anonymously salty, and therefore dense, water persisted at site 1 until the end of April, long after the down-valley flow had ceased. This suggests that some of the brine produced at the flaw lead remained on the shelf in early March rather than immediately draining to the Canada Basin. There is some evidence of cold salty water associated with brine rejection at site 2 at the end of April (Figure 4), but note that if cold salty water did cross the shelf break at site 2, it could also have done so higher in the water column than above the near-bottom temperature/salinity sensor. The cold saline water at site 1 is of comparable density to the warm saline water at the bottom at site 2, and mixing as the water transits across the shelf could lower the density of the site 1 water, so that it detaches from the bottom before site 2.
 On the basis of the time series of near-bottom temperature at site 1 that began in 1985, near-freezing-point water is observed at site 1 at some point during the winter in about two thirds of the years observed (not shown). It is useful to consider what conditions may be necessary to cause these events as they are plausibly connected with down-valley flow. Obviously, ice motion in the winter must be conducive to the formation of large amounts of new ice, requiring that the prevailing up-shelf ice motion is accompanied by larger than normal off-shelf ice motion to open the flaw lead. In addition, wind stress from storms in the autumn may precondition the shelf by removing the buoyant Mackenzie Plume [Melling, 1993]. These autumnal storms tend to possess downwelling-favorable wind stress which could have the effect of pushing the plume onshore, resulting in a down-shelf coastal current [Fong and Geyer, 2001] that advects the plume waters toward Amundsen Gulf. If the Mackenzie Plume is removed in the autumn, the reduction in surface stratification will allow deeper convection from brine rejection during winter and greater maximum densities.
 In general, at site 1 in Kugmallit Valley, upwelling-favorable surface stress is associated with up-valley flow, and downwelling-favorable surface stress is associated with down-shelf flow. This pattern is similar to that found for cross-shelf canyons, although cross-shelf canyons typically have a much larger relative depth and also intersect the topography of the slope. Kugmallit Valley fades out before it reaches the shelf break, so that upwelling found within Kugmallit Valley must transport water across the shelf, rather than across the shelf break.
 Seasonal variation in stratification over the Mackenzie Shelf leads to seasonal variation of the response at site 1 to upwelling-favorable surface stress. During summer, when the depth of the mixed layer is likely to be shallower than the highest-velocity measurement, up-valley flow is relatively uniform over the vertical range of the ADCP at site 1. During winter the mixed layer depth increases because of brine rejection from ice formation and causes the surface Ekman layer to extend ∼40 m into the water column. Upwelling flow is then a superposition of up-valley flow and surface Ekman layer flow, which leads to a strong rotation of the velocity vectors toward the up-shelf direction. The velocity data set does not extend into the bottom boundary layer, and so we do not have information as to how bottom friction affects the flow in the valley.
 There is seasonal variation in the bottom stratification. During autumn of 2002, upwelling within Kugmallit Valley advected cold salty water past the bottom of site 1. However, for much of the winter, up-canyon flow does not result in large changes in the near-bottom temperature and salinity at site 1, apparently because near-bottom temperature and salinity have little cross-shelf variation during this time.
 Dense flaw lead water flows down Kugmallit Valley for 2 weeks in early March 2003. This event coincided with the closure of the flaw lead immediately after 1 week of swift, upwelling-favorable wind stress and ice motion. We suggest that during upwelling the dense water that had been forming throughout the winter moved back onshore, toward the flaw lead, thus becoming thicker and gaining potential energy. When upwelling-favorable stress ceased and the flaw lead began to close, the potential energy of the dense water was released, eventually causing the swift down-valley flow observed 2 d later.
 Field measurements were supported by the Federal Panel on Energy Research and Development, under the Northern Hydrocarbons Objective. Moorings were serviced with the able assistance of the officers and crew of the CCGS Sir Wilfrid Laurier, in collaboration with the Canadian Coast Guard and the Japan Marine Science and Technology Center. The satellite image was kindly provided by G. M. Schmidt and T. J. Weingartner. Aqua satellite data were obtained from the Ocean Color Data Processing Archive, NASA/Goddard Space Flight Center Greenbelt, MD, USA. Support for this research was provided by the Natural Sciences and Engineering Research Council of Canada via the Canadian Arctic Shelf Exchange Study.