Low frequency variability on the continental slope of the southern Weddell Sea

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH: OCEANS, VOL. 118, , doi: /jgrc.20309, 2013 Low frequency variability on the continental slope of the southern Weddell Sea Mari F. Jensen, 1 Ilker Fer, 1 and Elin Darelius 1 Received 22 March 2013; revised 28 June 2013; accepted 9 July 2013; published 5 September [1] One-year long records of temperature, salinity, and currents show seasonally varying, energetic oscillations with a dominant period of approximately 35 h on the upper continental slope of the southern Weddell Sea. The data set is sampled by five moorings deployed on the slope of the Crary Fan, east of the main outflow site of the Filchner overflow plume. The characteristics of the observed oscillations are compared to idealized coastal trapped waves inferred from a numerical code. The variability at 35 h period is identified as mode 1 waves with wavelengths less than 200 km and group velocity opposing the phase speed, indicating energy propagation toward east. Filchner Depression and the nearby ridges on the slope are suggested as the generation site where the dynamics associated with the overflow plume can force the variability. Historical time series at the overflow site are revisited to identify the source of previously reported variability at 3 and 6 day time scales. Mode 2 waves at wavelengths of about 100 and 1000 km were found to bear resemblance to the 3 day and 6 day variability, respectively. The seasonal variation in the energy in the 35 h band shows small but significant correlation with the low frequency easterly winds. The presence of coastal trapped waves along the continental slope of the Weddell Sea can increase the heat exchange across the shelf break and affect the dense water production rates. Citation: Jensen, M. F., I. Fer, and E. Darelius (2013), Low frequency variability on the continental slope of the southern Weddell Sea, J. Geophys. Res. Oceans, 118, , doi: /jgrc Introduction 1 Geophysical Institute, University of Bergen, Bergen, Norway. Corresponding author: I. Fer, Geophysical Institute, University of Bergen, Allegaten 70, NO-5007 Bergen, Norway. (Ilker.Fer@gfi.uib.no) American Geophysical Union. All Rights Reserved /13/ /jgrc [2] Antarctic Bottom Water (AABW) is the cold and dense abyssal water found beneath the Arctic-derived deep water, and it is an important part of the global thermohaline circulation. Approximately 10 Sv (1 Sv ¼ 10 6 m 3 s 1 )of AABW is exported from the Southern Ocean [Orsi et al., 1999]. The Weddell Sea is the source of the coldest and the most oxygen-rich bottom waters in the Southern Ocean and one of the major sites for deep water production in the Antarctic [Orsi et al., 1999]. The AABW production in the Weddell Sea occurs over the broad continental shelf of the southern Weddell Sea through a complex interaction of warm Circumpolar Deep Water, dense shelf waters, and cold Ice Shelf Water (ISW) [Foldvik et al., 1985]. The interaction of warmer off-shelf water masses with the colder waters on the continental rise and shelf influences the AABW production rates. [3] The continental slope of the southern Weddell Sea is home to subinertial mesoscale variability at time scales varying from diurnal to several weeks [Middleton et al., 1982; Darelius et al. 2009]. Darelius et al. [2009] observed low frequency oscillations with periods of 35 h, 3 and 6 days at the site of the Filchner overflow plume, but that study was inconclusive regarding the generation mechanism of the oscillations. West of the Filchner Depression, Middleton et al. [1982] observed mesoscale variability with periods exceeding 3 days which could be attributed to coastal trapped waves (CTW). In addition, diurnal CTWs have been observed on the continental slope of the southern Weddell Sea [Middleton et al., 1987]. The subinertial variability can enhance acrossshelf exchange and vertical mixing by, for example, increasing the across-isobath flow by upwelling in a trough crosscutting the shelf [Allen and de Madron, 2009], that in turn can affect the deep water production rates. Furthermore, the penetration of warm inflow water onto the shelf and beneath the ice shelves might have consequences for the future state of the ice shelves in the area [Hellmer et al., 2012]. [4] Coastal trapped waves are trapped by a boundary, such as a coastal wall or a sloping bottom, where the depth variation in topography and the conservation of potential vorticity provide a mechanism to support the waves. The waves decay rapidly away from the boundary by which they are trapped. For an extensive discussion of CTWs, see, e.g., Mysak [1980] and Rhines [1970]. Several studies on continental slopes at other sites have shown observations of subinertial oscillations that compare well with properties of CTWs, for example, the Gulf of Guinea [Vangriesheim et al., 2005] and the California coast [Chapman, 1987]. CTWs can lead to across-shelf exchange through near-resonant response or through wave propagation from a 4256

2 Figure 1. Map of the Weddell Sea showing the bathymetry (color) from the 2 arcmin global relief model ETOPO2, land (black), and ice shelves (white). AP is the Antarctic Peninsula, RIS is the Ronne Ice Shelf, FIS is the Filchner Ice Shelf, and KN is Kapp Norvegia. The star shows the position of Halley and the large arrow demonstrates the Weddell Gyre. The delineated study site is enlarged to show the positions of the moorings (M1 M5, red squares). Black arrows are rough indications of the pathways of the outflowing ISW plume suggested by Foldvik et al. [2004]. The position of the moorings F1 F4 discussed in Darelius et al. [2009] is shown by blue squares. The inset in the bottom right corner demonstrates the coordinate rotation. site with different water properties, and are therefore important for the circulation on the shelf [Huthnance, 1995]. The CTWs can enhance mixing of water masses that can be especially important on the continental shelf break where different water masses typically meet. In addition, several studies have shown that CTWs may develop undercurrents (countercurrents opposing the mean flow) [see, e.g., LeBlond and Mysak, 1977; Middleton and Cirano, 1999; Suginohara, 1982]. Strong undercurrents have been observed below the Antarctic Coastal Current (ACoC) [Nu~nez- Riboni and Fahrbach, 2009; Chavanne et al., 2010], which may affect the exchanges between coastal and deep waters near the Antarctic continental margins. A plausible forcing mechanism for the Antarctic slope undercurrents can be CTWs [Chavanne et al., 2010]. [5] In this study, the low frequency (i.e., subinertial) variability on the continental slope of the southern Weddell Sea is further investigated using observations from the continental slope. The observations consist of year-long time series of ocean temperature, salinity, and current structure, obtained from five moorings deployed on the Crary Fan, east of the main outflow site of the ISW plume (see Figure 1 for a map with place names). We hypothesize that CTWs are the cause of the subinertial variability observed in the study area. To test this, observations are compared with a numerical code for computing the 2-D structure of CTWs. The code is adapted to the study area using local stratification and a topography representative of the area. As the continental slope of the southern Weddell Sea is situated approximately 300 km away from the coast, this study focuses on CTWs trapped by a sloping bottom. 2. Data and Methods [6] Data were collected using five moorings deployed on the continental slope of the southern Weddell Sea between February 2009 and February 2010 (Figure 1). Two arrays were positioned to the east of the main ISW outflow, in a direction across the continental slope, one (M1-M2) approximately in the middle of the Filchner Sill and one (M3-M4-M5) roughly 80 km farther east. Pairs of moorings were positioned approximately at the same isobath on the continental rise; M1 and M4 at about 1000 m and M2 and M5 at about 1900 m. Mooring M3 was placed at the continental shelf break. The moorings were equipped with Sea-Bird Electronics temperature (SBE 39) and conductivity and temperature recorders (SBE 37 Microcat), current meters (Aanderaa Recording Current Meter (RCM) 7/8, Nortek Aquadopp), and acoustic Doppler current profilers (ADCP, RD-Instrument (RDI) 300 khz Sentinel and 75 khz Longranger, Nortek 190 khz Continental). The details of the moorings are given in Table 1. The sampling rate was set to 5 min for the Microcat and SBE39, 1 h for the RCM-7/8, Longranger, and Continental, and 20 min for the Aquadopp and Sentinel. The RCMs averaged 50 evenly distributed samples per hour, whereas the Aquadopp averaged velocity measurements from the first 180 s. The RDI ADCPs sampled an ensemble of 30 pings collected in burst mode for the first 60 s (Sentinel) or 150 s (Longranger). For the Continental, 300 s of profile averaging was done at the start of each interval. Two other RDI ADCPs did not return any data (not listed in Table 1) and the velocity measurements at 4257

3 Table 1. Mooring Details a Mooring Time (UTC) (in/out) Position (lon/lat) Bottom Depth (m) Height (m.a.b.) Parameter Instrument M1 10 Feb W , 67 T, C Microcat 17: S 24 T, C, V RCM-7 46 T, P, V Aquadopp 10 Feb T SBE 39 22: T, C, P Microcat 136 T, P SBE 39 M2 11 Feb W T, P, V RCM-7 18: S 68 T SBE T, C, P Microcat 10 Feb T, P Sentinel 22:00 78:4:150 V, W Sentinel M3 13 Feb W T, C Microcat 16: S 25 T, V RCM-7 77, 154, 360 T, C, P Microcat 9 Feb , 128, 257 T SBE 39 18:00 123:4:199 V, W Sentinel 205 T, P Sentinel 308 T, P Continental 310:5:505 V, W Continental M4 13 Feb W , 78, 314 T, C, P Microcat 13: S 25 T, V RCM T SBE 39 9 Feb T, C Microcat 15:00 183, 261 T, P SBE T, P Longranger 442:16:986 V, W Longranger M5 12 Feb W , 415 T, C Microcat 22: S 26 T, C, V RCM-7 52, 104, 311, 363 T SBE Feb , 259 T, P SBE 39 22: T, C, P Microcat 415 T, P Longranger 55:16:391 V, W Longranger a T is temperature, C is conductivity, P is pressure, V is horizontal velocity, and W is vertical velocity meters above bottom (m.a.b.) from the Continental were obstructed by flotation elements. [7] For this study, the time series from the moored instruments are averaged into 1 h intervals, and low-pass filtered with a fourth-order Butterworth filter using a 25 h cutoff period. By design, the half-power point of the Butterworth filter is at the cutoff frequency, that is, the amplitude is attenuated by a factor of two at 25 h period; at the frequency corresponding to 35 h period, the attenuation is Figure 2. Time series of nonfiltered (thin) and lowpassed (thick) velocity in across-slope direction (black) and temperature (gray) measured at mooring M3 at 700 m depth. A 35 h period is shown for reference. 6%. This cutoff period is chosen to include the 35 h oscillation found in Darelius et al. [2009], whereas excluding the inertial and dominant tidal frequencies and the highfrequency variability. The local inertial period is 12.5 h. The coordinate system is rotated with ¼ 110 clockwise (CW) from north, aligned with the mean orientation of the isobaths inferred using the 2 minute Gridded Global Relief Data (ETOPO2). The rotated coordinate system is right handed, oriented with x 0 pointing upslope and y 0 pointing east, along-slope (see the inset in Figure 1). The uncertainty associated with the choice of the isobath orientation is about 20. [8] Spectral analysis of the nonfiltered hourly averaged data is used to detect the dominant periods of the temperature and velocity oscillations. All spectra are calculated using eight, half-overlapping segments, using a Hamming window. For the velocity, the rotary component spectra [Gonella, 1972] are used to obtain the variance in the CW and counterclockwise (CCW) rotating components. 3. Observations 3.1. Low Frequency Variability [9] Low frequency oscillations of velocity and temperature are observed at all moorings and at all depths. An 4258

4 Figure 3. Low-passed velocity in (a) across-slope direction (U 0, positive upslope), (b) along-slope direction (V 0, positive eastward), and (c) low-passed temperature at moorings M1, M3, and M4. example of the oscillations is shown in Figure 2 for the across-slope velocity component and the temperature measured at mooring M3, using both the low-passed and the nonfiltered data. The dominant 35 h variability is clearly visible in the nonfiltered data, which is successfully delineated in the low-passed record. The oscillations are most energetic at the continental shelf break and the upper slope (M1, M3, and M4), and decay down the continental slope. At moorings M2 and M5, offshore from the shelf break, the oscillations are substantially attenuated; therefore, the focus will be on the moorings on the upper slope. [10] Low-passed time series with energetic oscillations from selected instruments and moorings are shown in Figure 3. Both velocity components show oscillations, but the across-slope component is typically more energetic (twice as much at M1) and regular (Figure 3a). The largest amplitudes in the entire data set are observed in the across-slope velocity component at mooring M1, however, varying with time, from approximately 5 cm s 1 inmayto40cms 1 in February. At the other moorings, the variability and the difference between the two velocity components are smaller. The amplitudes of the velocity oscillations reduce to approximately 15 cm s 1 and 10 cm s 1 at moorings M3 and M4, respectively. From September to January, the middepth velocities at mooring M3 have comparable amplitudes to those at M1. [11] The low frequency oscillations are also observed in temperature (Figure 3c). The distribution of temperature in a section along the mooring array, collected in February 2009, shows the cold upper layer and the warmer deeper waters typical at the site (Figure 4). The oscillations move the front between the cold on-shelf water and the warmer off-shelf water past the moorings. Upslope and downslope velocities are typically associated with an increase and decrease in temperature, respectively (Figure 2). [12] A thorough coherence and correlation analysis using the hourly winds measured at Halley (Figure 1, approximately 150 km east of our measurement site) and our current measurements conducted using both the low-passed and the nonfiltered currents from several moorings and depths, and using various lags, showed no correlation (close to zero at the 95% significance level) for the 35 h period. The low frequency variability is therefore not driven by short-duration local wind events Dominant Frequency Bands [13] The across-slope velocity spectra at moorings M1, M3, and M4 for the entire duration of the time series (Figure 5) show two dominant subinertial frequency bands; the daily band B24 ( h) and a band B35 ( h) centered at 35 h. Whereas B35 compares with the shortest period observed in Darelius et al. [2009], B24 contains the diurnal tidal frequency, which is also subinertial at this latitude. Both bands are statistically significant, above the 97.5% confidence level, and the spectra are representative of those obtained from other levels at these moorings. There are no other significant peaks at lower frequencies. In this study, we concentrate on B35 because B24 has already been discussed and explained by Foldvik et al. [1990] and Middleton et al. [1987]. The spectra at the deeper moorings, M2 and M5, have less energy density in general, showing relatively energetic oscillations with periods exceeding 100 h (not shown). The low frequency variability at the deeper moorings is not studied further here Vertical Structure [14] The vertical structure of the velocity oscillations is presented using the horizontal eddy kinetic energy (EKE) and the current ellipses. The EKE is obtained by integrating the velocity spectra in the frequency band bounded by! 1 and! 2, 4259

5 Figure 4. Potential temperature section measured in February 2009 along the mooring line M3 M5. Moorings M3, M4, and M5 are in the vicinity of stations 101, 99, and 95, respectively. Isotherms are drawn at 0.2 C intervals. The relatively warm water with temperatures greater than 0 C is marked in gray. Z!2 EKE ¼ ðs uu þ S vv Þd!; ð1þ! 1 where S uu and S vv are the power spectral density of the across-slope, u, and along-slope, v, component of the velocity, respectively. The EKE is calculated for the two energetic frequency bands B24 and B35, using half-overlapping 4 week windows. The vertical coverage at M1 is limited, but the resolved structure of the EKE in B35 (EKE 35h ) shows bottom enhancement (Figure 6a); the velocity oscillations at 943 m depth are 5 cm s 1 larger than those at 921 m (not shown). At mooring M3, a slight bottom enhancement is also observed at 700 m depth, 25 m.a.b., relative to the homogeneous bottom 200 m. The energy then increases toward the surface, reaching a maximum at middepth. At mooring M4, the energy is relatively evenly distributed with depth, and attenuates slightly at depths shallower than 300 m. [15] The principal axes of a velocity vector time series are identified with the major axis, which contains most of the variance, and the minor axis, which contains the remaining variance. The orientation of the principal axes of the currents is found by rotating the velocity vector in 1 Figure 5. Weighted across-slope velocity variance spectra at moorings M1, M3, and M4. Vertical dashed lines mark the frequency bands B35 and B24. increments and identifying the direction with the maximum EKE 35h. The sense of rotation of the current ellipses along the principal axes is inferred from the dominating rotary component in the given frequency band. The angle between the principal axis and the orientation of the isobaths,, is calculated using the orientation of the local topography at each mooring. [16] The major principal axis is directed approximately normal to the orientation of the isobath at all depths and at all moorings (Figure 6b). Table 2 tabulates the angle between the major principal axis and the orientation of the local isobath for the annual average as well as for a winter and a summer event identified in section 3.2. There is a small (less than 10 ) decrease in the orientation of the axes with increasing height at moorings M3 and M4. CCW rotation dominates at all depths at the shallow moorings (M1 and M3), whereas CW rotation dominates at the deeper moorings (M2, M4, and M5) Phase Propagation [17] Phase difference, D, squared coherency, 2, and spectra are calculated using measurements at a given height between moorings separated in the along-slope direction. This analysis gives information about the along-slope phase propagation of a coherent frequency band. The wavelength and phase speed of the coherent signal in B35 are found using the phase difference and the known distance between the moorings. Because the currents are not coherent with the local wind for the 35 h period (see section 3.1), we assume free wave propagation in the following analysis. [18] The horizontal phase propagation along the 1000 m isobath is calculated between moorings M1 and M4, 70 km apart. Squared coherency and phase spectra show coherent motion in bands B24 and B35 (Figure 7). All phase values with squared coherency above 0.4 (the 97.5% confidence level) are extracted to calculate wave number and frequency pairs. The results are shown by markers in Figure 8 with a positive wave number denoting a westward propagating signal (the dispersion curves from the numerical code will be discussed later). Wave number and frequency 4260

6 Figure 6. (a) Mean eddy kinetic energy, EKE, (cm 2 s 1 ) at moorings M3, M1, and M4, in B35. The diameter of the circles increases with increasing EKE. Black, horizontal lines mark the bottom. (b) Principal axes of current velocity variance at moorings M3, M1, and M4 in B35. Black, thin lines show the orientation of the isobaths at the mooring location inferred from bathymetry. pairs found in B24, calculated from the across-slope velocity component, are clustered at wavelengths from 200 to 300 km. The along-slope wavelengths of the across-slope velocity component in B35 are significantly shorter, ranging from 60 to 100 km with phase speeds from 47 to 85 cm s 1.The wavelengths and phase speed in the along-slope component range from 100 to 370 km and from 76 to 320 cm s 1, respectively. For both components, the frequency decreases with increasing wave number and the group velocity is thus in the opposite direction of the phase velocity. [19] The coherence and phase differences are also calculated for different vertical separations at a given mooring and at a given height between moorings separated in across-slope direction. In the vertical, at moorings M3 and M4, the squared coherency is high, approximately 0.8, and decreases with distance from the reference level (Figure 9). The across-slope component is more coherent than the along-slope component at both moorings. At mooring M3, there is a general increase of phase difference with height in the depth range covered by the mooring. The maximum phase offset is observed at 220 m depth, where the acrossslope velocity lags that in the deeper layers with 16. In the upper half of the water column at mooring M4, the measurements are in phase to within 65, with no significant vertical phase offset. The observation of coherent and inphase across-slope component is also representative of M1 (not shown). [20] There is also coherence between across-slope velocity measurements at moorings situated in a line across- Table 2. Angle (in Degrees) Between the Major Principal Axis and the Orientation of the Local Isobath Mooring Depth (m) Yearly Average Summer Event Winter Event M M M M M Figure 7. (a) Squared coherency, 2, and (b) phase difference, D, in across (U 0 ), and along-isobath (V 0 ) velocity component between moorings M1 and M4 at 25 m.a.b. A positive D indicates that mooring M4 leads mooring M1. Only results above the 97.5% confidence level are shown. 4261

7 Figure 8. The wave number and frequency pairs calculated from observed velocity between moorings M1 and M4 at 25 m.a.b. for the across (U 0 ) and along-isobath (V 0 ) velocity components. The dispersion relations from the numerical code are plotted as black lines. slope from each other (Figure 10). The coherence and phase are calculated at heights above bottom where joint measurements are available, at roughly 400 m.a.b. between moorings M3, M4, and M5. The phase difference in the across-isobath component is close to zero between moorings M3 and M4, whereas it increases to 55 between moorings M3 and M Seasonal Variations [21] The amplitudes of the across-slope velocity oscillation are generally larger during austral summer than austral winter (Figure 3). The development of EKE with time for B24 and B35, normalized by their respective maximum, is shown in Figure 11. Both the EKE 35h and EKE 24h show an increase in EKE during austral summer. The maximum value is found during December in both frequency bands. During austral winter, the EKE 35h is low, generally between Figure 10. B35 band-averaged (a) squared coherency, 2, and (b) phase difference, D, in across-isobath (U 0 ) velocity between moorings placed along the mooring line M3 M5. A positive D indicates that the reference mooring, M3 leads. Only results above the 97.5% confidence level are shown. 5% and 20% of its maximum value. The EKE 24h is generally high, exceeding 20% of its maximum value, throughout the year. [22] A couple of 2 week long oscillation events are compared; one during austral summer and one during austral winter. The summer event is from 16 to 28 December 2009, and the winter event is from 2 to 14 July The oscillations at the shelf are relatively more energetic during the summer event (Figure 12) and the velocity oscillations are larger by a factor of two. Temperature anomalies of 0.5 C reach up to 0.28 of the total depth during summer, Figure 9. B35 band-averaged squared coherency, 2, in the across-isobath (U 0 ) and along-isobath (V 0 ) velocity component for moorings (a) M3 and (c) M4, and band-averaged phase difference, D, for moorings (b) M3 and (d) M4, relative to a reference level of 123 and 442 m.a.b. for M3 and M4, respectively. A positive D indicates that the reference level leads. Black lines mark the zero phase difference. Only results above the 97.5% confidence level are shown. 4262

8 Figure 11. Normalized vertical average of eddy kinetic energy, neke, in (a) B24 and (b) B35 at moorings M1, M3, and M4. Each EKE curve is normalized by its maximum value, neke ¼ EKE/EKE max, and EKE max is indicated in cm 2 s 1 next to the corresponding curve. Gray patches mark the summer and winter events analyzed in detail. compared to 0.14 during winter. Temperatures above 0 C are observed only in 1 day during the winter event. During both events, the current vectors alternate between converging vectors, i.e., crossing each other, and diverging vectors, i.e., spreading out. The periods of oscillations are longer during the summer event; the mean period of the temperature oscillations is 34 h, as opposed to 26 h during the winter event. [23] The substantial difference in the apparent period of the oscillations between the summer and winter event is a general pattern, not limited to the chosen events. Acrossslope velocity spectra, derived from 3 month long records in winter and summer periods, show higher energy densities during summer months (December to February) for longer periods than during winter months (July to September) (Figure 13). During the winter event, the EKE 24h is large, 60% of its annual maximum value, whereas EKE 35h is low, 5% of its annual maximum value (Figure 11). During the summer event, the EKE in both frequency bands is at its maximum value. The coherence between the velocity measurements generally increases during the summer event and decreases during the winter event (not shown). 4. Numerical Calculations [24] The observations are compared with the idealized numerical code of Brink [2006], run with a topographic profile representative of the area and summer stratification Figure 12. Low-passed temperature (contours), mean velocity (white arrows), and velocity anomalies (white sticks) at mooring M3 during the summer and winter event. Velocities directed in positive y axis are directed downslope. 4263

9 Figure 13. Across-slope velocity variance spectra at mooring M1, at 921 m depth, for summer and winter months. Vertical lines mark the frequency bands B35 and B24. The error bars show the 97.5% confidence level. inferred from a Conductivity, Temperature, and Depth (CTD)-survey taken in February 2009 from the Royal Research Ship (RRS) Ernest Shackleton. The topographic profile is the average of six adjacent across-slope profiles with approximately 20 km separation at the measurement site extracted from ETOPO2. The ice shelves are represented as a coast that is approximately 300 km away from the shelf break. [25] The 2-D code varies only in the across-slope direction and solves the CTW problem [see, e.g., Brink, 2006] numerically by resonance iteration and calculates the dispersion relations and modal structures for stable, inviscid CTWs. The bottom friction is assumed to be negligible and the resolution is chosen to be 120 grid points in the acrossslope direction and 20 grid points in the vertical. The resulting dispersion relations for the first three modes are presented in Figure 8, and compared with the wave number-frequency pairs inferred from observations at moorings M1 and M4 (section 3.1.3). Mode 1 is the only mode with frequencies in the bands of interest in the wave number range presented. Modes 2 and 3 correspond to lower frequencies observed farther down the continental slope. Mode 0 is too fast to be shown in the wave number range presented. [26] The mode 1 dispersion curve intersects the frequencies in band B35 at both small (k < m 1 ) and large (k > m 1 ) wave numbers, with the latter covering a broad range of wavelengths from 75 to 200 km. The phase speed in band B35 ranges from 56 to 164 cm s 1. The observed wave number-frequency pairs compare fairly well with mode 1 in the frequency band of interest (Figure 8a). Although the pairs in the along-slope component adhere to the dispersion line in B35, the pairs in the across-slope component are about rad s 1 above the dispersion line. At wavelengths less than 70 km, the along-slope data points deviate toward higher modes. [27] The across-slope and along-slope mode 1 velocity structures at four different wavelengths with frequencies close to, or within, band B35 are compared with the observations. Generally, the oscillations are enhanced close to the shelf break with a weak but notable bottom trapping of the oscillations that decreases with increasing wavelength (Figure 14). At the shelf proper and off the continental slope, the signal is barotropic. The width at the shelf break where the across-slope oscillations are enhanced is narrower and locked at the shelf break for shorter waves. [28] Current ellipses inferred from the modeled velocity structures at the positions of the moorings are compared with the observed current ellipses (Figure 15) to aid in the comparison of observations and the waves modal structure. For the longest wavelength (Figure 15b), the modeled oscillations are largest in the along-slope component, and the wave is close to barotropic. This does not compare well with the observed current ellipses, where the across-slope component dominates and varies with depth. For the shorter Figure 14. Normalized mode 1 perturbation velocities at different wavelengths. White circles mark the mooring positions. The velocities are normalized with the maximum negative value found in either the across-slope or the along-slope velocity component for the given wavelength. 4264

10 Figure 15. Current ellipses calculated from (a) observations, and model (b) mode 1, ¼ 3141 km, (c) mode 1, ¼ 196 km, (d) mode 1, ¼ 137 km. The y axes represent across-slope direction. The current ellipses from the numerical code are inferred from the normalized velocities shown in Figure 14 at the positions of moorings M1, M3, and M4. Black, horizontal lines in Figure 15a mark the bottom. wavelengths (Figures 15c and 15d), the modeled oscillations resemble the observations in that ellipses are elongated in the across-slope direction and vary with depth. However, the middepth increase in energy at M3 and the enhancement of the oscillations at mooring M1 compared with mooring M3 are not seen in the numerical calculations. [29] The results from the numerical code capture the enhanced oscillations in the across-slope component (at wavelengths shorter than 200 km) and at the upper slope and shelf break. The reduction in oscillation amplitude toward moorings M2 and M5 is seen in both observations and in the CTW model results (Figure 14) Sensitivity to Changes in Mean Velocity [30] The numerical code is run with different mean alongshore velocities to study the westward flowing ACoC s impact on the dispersion relation and the structure of the waves. Alongshore velocities are introduced, directed westward varying from a maximum of cm s 1. The position (x ¼ 340, 310, 280 km, corresponding to on slope, shelf break, shelf proper, respectively), width (W ¼ 10, 50 km), and vertical structure (barotropic, surface enhanced) of the core of the alongshore flow are varied. The structure of the flow centered at the shelf break can be seen in Figure 16 for a core velocity of 25 cm s 1. The structure is similar for different core velocities and for flows centered on the slope or the shelf. The observed mean alongshore current, averaged over the year, has a maximum speed of 16 cm s 1 at the shelf break and decays rapidly downslope (not shown). Monthly mean velocities of 25 cm s 1 are only observed during winter months. [31] For all cases, the impact of a westward flow on the dispersion relation is the decrease in the period for a given wave number (Figure 17), that is, an increase in the phase Figure 16. The mean alongshore velocity structure for a flow centered on the shelf break at x ¼ 310 km with V max ¼ 25 cm s 1 and width (a) W ¼ 10 km and (b) W ¼ 50 km. The structure is similar for different V max and for flows centered on the slope or shelf. 4265

11 Figure 17. The modeled dispersion relation for different mean alongshore velocities. Figures in the legend correspond to the core of the current, x (km), the horizontal width, W (km), and the maximum velocity, V max (cm s 1 ), respectively. x ¼ 310, 340, 280kmisacurrentcenteredonthe shelf break, upper slope, and on the shelf proper, respectively. speed. For strong and wide flows, waves in B35 do not occur. For all other cases, however, B35 oscillations are supported. When the core of the flow is located at the shelf break, the most energetic oscillations shift 10 km further offshore compared to when there is no flow (Figure 18). When the core is offshore or on the shelf proper, there is no discernible change in the lateral position of the maximum oscillations. [32] A strong alongshore flow generally leads to slightly more bottom trapping. Waves in B35 are generally found at higher frequencies, corresponding to shorter waves that are typically more bottom trapped. Runs with a baroclinic flow show that adding a surface-enhanced mean alongshore flow produces more bottom-trapped oscillations than with a barotropic flow or without a flow (not shown) Sensitivity to Changes in Stratification and Topography [33] An increase in stratification increases the frequencies of the mode and the baroclinity of the waves [e.g., Huthnance, 1978], whereas increasing cross-shelf scale decreases the frequencies of the mode. To see the effect of stratification and topography at our study site, the CTW numerical code is run with vanishing stratification and a new, smooth topography. The topography is adapted from Middleton et al. [1982] and includes a less distinct shelf break and a smaller slope gradient. [34] Sensitivity runs with both topography representations, with and without stratification, show that the dispersion curves are relatively steeper for the cases without stratification (Figure 19). Adding stratification shifts the dispersion relation toward higher frequencies within B35, i.e., increases the phase speed for a given wave number. Over smooth topography with a larger horizontal scale of the shelf break, the frequencies decrease, and B35 is only found at wavelengths exceeding 400 km. The mode 2 dispersion curves do not cross the observed dominant frequency bands in the mooring records, either after changing the stratification or the topography. [35] With the smooth topography representation, the relative difference between across-slope and along-slope oscillation strength increases, with larger across-slope oscillations (Figure 20). In addition, the lateral extent of the oscillations increases for both components and the position of the most energetic oscillations moves down the continental slope. 5. Discussion 5.1. Observations [36] We observe energetic subinertial oscillations on the continental slope of the southern Weddell Sea, which vary Figure 18. Normalized mode 1 perturbation velocities for a wavelength of order 100 km in B35, for different mean alongshore currents centered on the shelf break. The velocities are normalized with the maximum negative value found in either the across-slope or the along-slope component for each model run. 4266

12 Figure 19. The modeled dispersion relation for different representation of topography for mode 1 and mode 2. The dispersion relation for average topography is the same as in Figure 8. with time and season. The amplitude of the oscillations decays rapidly away from the shelf break, and the observed variability is hypothesized to be caused by CTWs. We concentrate on the velocity oscillations with a period of approximately 35 h in the band B35, which have previously been observed farther west in the Weddell Sea by Darelius et al. [2009]. [37] Alternating converging and diverging current vectors are observed on the upper slope (see Figure 12). Foldvik et al. [1988] showed that converging current vectors correspond to CCW eddies passing a moored instrument, and diverging vectors to CW eddies. The observed pattern can thus be explained by the movements of a wave that contains alternating CW and CCW rotating zones. There is high and significant coherence in B35, suggesting waves with the same frequency generated from the same source. Our analyses imply alongshore propagation with phase velocity to the west, i.e., with shallow water to the left, consistent with the CTW theory in the Southern Hemisphere [Mysak, 1980]. The alongshore propagation of across-slope velocity at the 1000 m isobath has wavelengths ranging from 60 to 100 km and phase speeds ranging from 47 to 85 cm s 1 in frequency band B35. For these short waves, the group velocity is directed opposite to the phase velocity. [38] The velocity measurements used in the analysis (moorings M1 and M4) are at 25 m.a.b. A rough estimate of the bottom boundary thickness is h ¼ G=f, where G is the magnitude of the geostrophic current outside the bottom boundary layer [Weatherly, 1975]. Using f ¼ s 1, and typical average velocities above the boundary layer of 0.2 m s 1, we estimate the bottom boundary layer thickness to be about 17 m. The velocity measurements are thus close to the bottom boundary layer, and the results should be interpreted with caution. [39] The analysis presented in section also implies offshore phase differences suggesting offshore propagation. For idealized CTWs, only alongshore propagation occurs. The observed offshore phase differences may be because of bottom-layer friction [Brink and Allen, 1978] or uncertainties with the isobath orientation where across-shelf depth variation results in phase lag for barotropic CTWs. [40] Independent CTW characteristics inferred from the idealized numerical code with observed summer stratification and topography representative of the measurement site bear comparison with the observations. Mode 1 is the only mode compatible with the observations and the only mode allowing waves in the frequency band of interest. The observed wave number-frequency pairs are scattered near the mode 1 dispersion line. However, the observed acrossslope and along-slope velocity components follow different curves. As noted in Middleton et al. [1982], coherence and phase lag calculations using Cartesian components of the velocity might produce biased phase lags through a Figure 20. Normalized mode 1 perturbation velocities for a wavelength of 137 km for different topographies and stratifications. The velocities are normalized with the maximum negative value found in either the across-slope or the along-slope component for each model run. 4267

13 realignment of current fluctuations with topographic features. The mean direction of the 1000 m isobath changes from approximately 35 near M4 to 10 near M1, which can have a significant effect on the resulting phase lags. Modes 2 and 3 are compatible with the longer period motions observed farther down the continental slope. [41] The Burger number is a measure of the relative importance of shelf geometry and stratification; where N is the buoyancy frequency, B u ¼ ðnhþ2 ðflþ 2 ; ð2þ N 2 ¼ ; and L and H are the length and height scales, respectively, of the shelf. Flow tends to be baroclinic for large B u ; the stratification becomes important when the shelf is narrow. For B u of O(1) or less, the flow tends to be barotropic [Brink, 2006]. [42] At the southern Weddell Sea shelf, B u using L ¼ 300 km, H ¼ 600 m, and N ¼ s 1, inferred from 40 CTD stations collected in February 2009 in the vicinity of the moorings. For long waves, this implies a barotropic flow with no bottom intensification. According to Wang and Mooers [1976], CTWs should approach barotropic continental shelf waves for such weak stratification. In this study, the velocity oscillations vary with depth in the water column close to the shelf break (moorings M1 and M3) with close to barotropic waves at the deeper moorings, similar to observations made farther west [Darelius et al., 2009]. At mooring M1, which has poor vertical resolution, we observe bottom trapping. A weak bottom trapping close to the shelf break is consistent with the theoretical study of Wang and Mooers [1976]. At mooring M3, however, the energy increases to a maximum at middepth, typically observed from September to January. This is not captured with the numerical code, or observed in Darelius et al. [2009] or in the theoretical study of Wang and Mooers [1976]. [43] The model dispersion curve intersects the frequency band B35 at wavelengths shorter than 200 km and longer than about 3000 km. Apart from the middepth enhancement at mooring M3, most of the vertical and lateral structures of velocity compare fairly well between observations and modeled mode 1 waves with wavelengths shorter than 200 km; both the numerical code and the observations show the largest oscillations in the across-slope velocity component and the energy is concentrated close to the shelf break. The observed increase in oscillation amplitude at mooring M1 compared with mooring M3 is not captured by the model. This can be attributed to the lack of a representation, in the model, of the processes which lead to wave decay. Energy loss from friction and from scattering by, for example, irregularities of topography and other mean and winddriven flows will lead to wave decay in the direction of energy propagation, i.e., to the east (see section 5.3.3). Hence smaller amplitudes at the eastern section (e.g., M3) compared with the western section (e.g., M1) might be ð3þ expected. Friction can also cause a wave to be more energetic over the slope than over the shelf break [Huthnance, 1995], and should be included for a more accurate representation of the CTWs. The presence of a mean alongshore flow, centered on the shelf break, leads to larger oscillation amplitudes located 10 km off the shelf break. In addition, a topography with a less distinct shelf break has been shown to support waves with maximum oscillations further down the slope, demonstrating that other processes can shift the position of the maximum oscillation amplitudes. [44] The lateral and vertical structures of velocity for wavelengths of approximately 3000 km do not compare well with the observations; the oscillations are largest in the along-slope component. The observation and model comparison leads us to conclude that mode 1 CTWs with wavelengths less than 200 km on the continental slope of the southern Weddell Sea are the primary source of variability at B35. [45] Our observations indicate short CTWs with negative group velocity (opposite to the phase velocity). Observations of similar short CTWs in nature are scarce [e.g., Gordon and Huthnance, 1987]. In their observations of stormdriven CTWs, Gordon and Huthnance [1987] identified lowest mode CTWs from different parts on the dispersion curve: a quasi-steady, low wave number response resulting in an along-isobath current for the duration of long wind events, and oscillatory responses forced by relatively short winds. The latter occurred at a frequency approximately corresponding to the peak of the dispersion curve where the group velocity is zero. Although the 3 6 day period oscillations discussed in section 5.2 can be because of low frequency and low wave number CTWs, the peak in the mode 1 dispersion curve is in the diurnal frequency band B24. Here, resonance can be expected because of forcing by the diurnal tide and nonpropagating energy of CTWs Synthesis with Earlier Observations [46] Low frequency variability on the continental shelf and slope of the Weddell Sea has previously been observed west of our study site [Middleton et al., 1982; Darelius et al., 2009]. Our results are consistent with both studies that reported decreasing frequency of oscillations down the slope and a barotropic structure on the shelf. The observations of Darelius et al. [2009] were made roughly 100 km west of our measurement site, on the continental slope from depths of 650 to 2100 m and is the only study that reported oscillations with periods in the frequency band B35. A 35 h period was generally found on the upper slope, whereas longer periods, of 3 and 6 days, were generally observed at deeper moorings. Less energetic oscillations with periods of more than 3 days were observed at the deeper moorings in our data set. [47] Our results support the suggestion of Darelius et al. [2009] that the Filchner area is home to mode 1 CTWs. The oscillations with periods of approximately 35 h investigated in Darelius et al. [2009], generally had the largest variations in the across-slope direction, consistent with our study. No middepth enhancement of energy was reported; the oscillations were neither attenuated nor strengthened with depth, agreeing with observations off the shelf break in this study. Their analysis suggested a link between the tides and the oscillations whereby the 35 h oscillation was 4268

14 Figure 21. Current ellipses calculated from observations (O), modeled mode 2 (2) with ¼ 125 km and T ¼ 3 days, and mode 3 (3) with ¼ 55 km and T ¼ 3 days. Vertical lines represent across-slope direction. All figures have identical y axis range with tick marks at 200 m intervals. The depth of the uppermost level is shown on top right of each panel. The current ellipses from the numerical code are inferred from the normalized velocities from model runs at the depths of moorings F1 F4. The observed current ellipses are calculated in a band centered at 3 days. When the angle between the major axis and the orientation of local isobath orientation is larger than 20 away from the along-slope or across-slope direction, the ellipse is rotated and the local isobath is marked with a horizontal line. modulated by the neap-spring cycle of the tides, presumably because of the influence of tidal motions on the stratification in the region, and thus the transmission properties of CTWs. In our study, the neap-spring modulation is not observed. The study by Darelius et al. [2009] did not, however, conclude on the existence of CTWs causing the oscillations and further questioned the longer period oscillations in the area that had maximum amplitudes over the slope instead of over the shelf break. Higher modes, however, have more complicated structures, often with a maximum over the slope [Huthnance, 1995]. Stratification and friction may also affect the position of the strongest oscillations. The modeled dispersion relations for modes 2 and 3 reported here are consistent with the longer periods observed farther down the continental slope and are further examined to explain the oscillations with longer periods. The results from the numerical code show a second maximum in both velocity components over the slope for mode 2 waves. [48] We calculated the current ellipses in the 3 and 6 day band using historical data from the current meter records over four moorings on the slope, F1 F4, reported in Darelius et al. [2009]. The mooring array is oriented in the southwest-northeast direction across the slope with F1 at the 650 m isobath and F4 at the 1980 m isobath. The observations are compared with the model results from mode 2 and 3 waves (Figures 21 and 22). The comparison shows that mode 2 CTWs bear resemblance to the oscillations at 3 and 6 day periods to a larger extent than mode 3 waves. The structure of the 3 day oscillations with wavelengths less than approximately 150 km agrees with the salient features of the observations showing dominant along-slope oscillations at mooring F2 and across-slope oscillations at F3. This is not the case for mode 3 results. The oscillations are most energetic at moorings F2 and F3, a feature not entirely captured by the numerical code. Sensitivity studies have shown that the lateral extent of the oscillations changes with topography and as the above mentioned comparison is performed using a topography profile representative of farther east, the change in topography can partly account for the discrepancy. Our analysis suggests that mode 2 waves with wavelengths of about 100 km lead to the 3 day oscillations observed by Darelius et al. [2009]. [49] For the 6 day period, both the numerical code and observations show the smallest oscillations at the upper slope. Mode 2 waves have the largest oscillations in the along-slope velocity component, consistent with the observations. Mode 3 waves with large across-slope oscillations at F3 are inconsistent with observations. The cross-isobath tilt of the ellipses at mooring F4 is not seen in the model results. Nevertheless, we propose that mode 2 waves with wavelengths exceeding 1000 km are responsible for the 6 day oscillations Forcing Mechanisms Energy Variations [50] The evolution of the subinertial oscillations in time shows seasonal and intraseasonal variations; for example, the EKE is stronger during austral summer and larger in July than in August. The energy in the 24 h band, EKE 24h, is generally large throughout the year (Figure 11a), always greater than 20% of the annual maximum. Because the tidal forcing is continuous, the presence of substantial background energy suggests forcing by diurnal tides. There are pronounced maxima in June to July and in December to January when the energy increases from the background values by a factor of approximately 3 and 5, respectively. Astronomical forcing of the diurnal tide is typically weak near the spring and autumnal equinoxes at a time when the sun and moon are both close to the equator. The diurnal 4269