Abstract

A 550-day record of Agulhas Undercurrent transport between March 2003 and August 2004 is constructed from five deep moorings placed on the continental shelf off South Africa at nominally 32°S. The vertical and lateral scales of the undercurrent are estimated to be 2000 m and 40 km, respectively, using the average of seven direct velocity sections, predominantly taken in austral autumn over a 10-yr period from 1995 to 2005. Peak speeds in the undercurrent are some of the greatest ever seen at depth: over 90 cm s−1 at 1400 m. The undercurrent has a transport of 4.2 ± 5.2 Sv (1 Sv = 1 × 106 m3 s−1), in close agreement with a previous estimate from a single current meter record during 1995 of 4.2 ± 2.9 Sv. Records below 1800 m, within the North Atlantic Deep Water (NADW) layer of the undercurrent, show a higher level of coherence and less variance than those at intermediate depths. On average, 2.3 ± 3.0 Sv of NADW is carried northeastward within the undercurrent, an amount similar to that estimated previously by analyses of deep water mass characteristics around South Africa.

Short-term variability in the undercurrent peaks at the semidiurnal period, at a local shear-adjusted inertial period (21.6 h), and at 4.5, 6.5, and 9.5 days. The latter may be associated with topographic Rossby waves, although no evidence for enhanced onshore velocities was found at these periods. The variability of the undercurrent is highly topographically controlled, strongly aligned in the along-stream direction, with significant variance in cross-stream velocity only at 2-day periods or less and isotropic variance only at the (effective) inertial period. For the mesoscale, the deeper current meters within the NADW layer all exhibit broad peaks at 50–60 days, which matches the periodicity of solitary meanders of the Agulhas Current (so-called Natal pulses) presented previously in the literature. The results of this study demonstrate that these meanders are highly barotropic in nature.

1. Introduction

The Agulhas Undercurrent was first measured in 1995 and at that time it was not clear whether it is a continuous feature carrying deep waters northward below the Agulhas Current, or whether it is more locally driven, by interaction of the Agulhas Current or incoming Rossby waves with topography, for instance (Beal and Bryden 1997). Perhaps there was more evidence for the latter, since maximum velocities were found up at 1200-m depth, rather than within the incoming North Atlantic Deep Water (NADW) layer, and water mass characteristics were not significantly different from those outside the undercurrent (Beal and Bryden 1999). Coincident with this first synoptic measurement of the undercurrent, a current meter array was deployed at 32°S to measure the transport of the Agulhas Current over a period of 1 yr. This array was not designed with the undercurrent in mind and only one instrument, at 2000-m depth and 32.4 km offshore, showed significant northward velocities on average. There were no time series measurements within the NADW layer of the undercurrent. A mean undercurrent transport of 4.2 ± 2.9 Sv (1 Sv = 1 × 106 m3 s−1) was estimated (Bryden et al. 2005). The undercurrent appeared to vary independently of the Agulhas Current above it, particularly with respect to meander events with periods of 50–70 days, which dominate the variance of the Agulhas Current, but did not appear significant in the undercurrent record.

Since the discovery of an Agulhas Undercurrent, a few investigators have proposed that it carries NADW from the southeast Atlantic slope current into the interior of the Indian Ocean, via the Mozambique Channel (Bryden and Beal 2001; Arhan et al. 2003: van Aken et al. 2004). Bryden and Beal (2001) postulated that the undercurrent could represent 40% of the Indian Ocean overturning transport, which is estimated at about 10 Sv (e.g., Ganachaud et al. 2000; McDonagh et al. 2008). Certainly the Natal Valley, the basin between South Africa and the Mozambique Plateau (see Fig. 2), is filled with NADW rather than Circumpolar Deep Water, which dominates only farther to the east (Wyrtki 1971). With careful water mass analysis, van Aken et al. (2004) has shown that NADW can be found in the narrows of the Mozambique Channel. However, the Natal Valley is closed to the north below 2250 m (Smith and Sandwell 1997), with a break in the plateau allowing waters above 2700 m to pass to the east at 31.5°S and into the Mozambique Basin. Hence, while some upper NADW [defined as waters above a neutral density surface of 28.08 at about 3000 m; Beal et al. (2006)] appears to reach the Mozambique Channel via the Agulhas Undercurrent, most of the undercurrent’s lower NADW must recirculate to the south [with some possible vertical mixing; Beal et al. (2006)]. Arhan et al. (2003) estimate that from 11 Sv of NADW exiting the South Atlantic around South Africa, only 2 Sv flow northward into the Natal Valley and the rest flows eastward underneath the Agulhas Return Current.

Fig. 2.

The average velocity structure of the Agulhas Undercurrent (enlarged from Fig. 1) with positions of the instruments on all AUCE moorings (M2 in gray was not recovered). Velocities to the southwest are shown in shades of yellow through red. Velocities in the undercurrent are shown in shades of white through blue. Black contours every 5 cm s−1. Bathymetry is depicted by the thick gray line. The inset map shows the geographical position of the AUCE moorings (black dots), on the eastern continental slope of South Africa. The bathymetries of the Natal Valley and Mozambique Plateau are shown in color contours: red through blue, 1000 to 5000 m, in intervals of 1000 m.

Fig. 2.

The average velocity structure of the Agulhas Undercurrent (enlarged from Fig. 1) with positions of the instruments on all AUCE moorings (M2 in gray was not recovered). Velocities to the southwest are shown in shades of yellow through red. Velocities in the undercurrent are shown in shades of white through blue. Black contours every 5 cm s−1. Bathymetry is depicted by the thick gray line. The inset map shows the geographical position of the AUCE moorings (black dots), on the eastern continental slope of South Africa. The bathymetries of the Natal Valley and Mozambique Plateau are shown in color contours: red through blue, 1000 to 5000 m, in intervals of 1000 m.

In 2003 the Agulhas Undercurrent Experiment (AUCE) was designed to collect sections of hydrographic and direct velocity data across the Agulhas Current and Undercurrent at four different latitudes (nominally 30°, 32°, 34°, and 36°S) and to measure a time series of undercurrent transport from a small moored array of deep instruments at the historical 32°S section. From these data, Casal et al. (2009) show that an undercurrent of similar structure to that measured in 1995, although generally weaker, was found along both sections to the south (upstream) of the 32°S section (as well as at that historic section) during AUCE. However, at the northernmost section (downstream), where the continental slope is broadened, depths are shallower, and the Agulhas Current is weakest, the deep flow structure was different. Here, direct velocities suggest an undercurrent core offshore of the Agulhas Current rather than over the slope, and at the depth of upper NADW. These results show that the Agulhas Undercurrent is a continuous feature within the Natal Valley, but is blocked at the head of the valley by topography, as expected.

Casal et al. (2009) constrained geostrophic transports from the quasi-synoptic AUCE data, using an inverse model initialized with direct velocity data, to find values spread between 1.5 and 4.2 Sv of upper NADW flowing northward across each of the AUCE sections. They found onshore–offshore recirculations 2–3 times stronger than this and also large transports of upper NADW carried back southward by the deep reaches of the Agulhas Current. Hence, this “snapshot” is dominated by variability and leaves the picture of the net flow of NADW through the region unclear, with very different values of undercurrent transport found from one section to the next.

Here, we report on the AUCE mooring data from 32°S, which yield a 552-day-long record of transport between March 2003 and August 2004. Of five current meters recovered from the undercurrent, three are within the NADW layer. We are interested in such questions as: What is the total transport and variability of the undercurrent? Is there a net northward transport of NADW? Is the undercurrent bottom intensified? What are the dominant periods of its variability? Is there evidence for a barotropic component to the meander mode of the Agulhas Current?

2. Mean undercurrent structure

To date, seven direct velocity sections have been occupied across the Agulhas Undercurrent at the historic 32°S section, as shown in Fig. 1. All were collected with a high pressure–rated, 150-kHz acoustic Doppler current profiler (ADCP), which was lowered to the seabed as part of a conductivity–temperature–depth (CTD) package. Three of these were occupied during the Agulhas Undercurrent Experiment (AUCE) in February and March 2003, and in March 2005. The data show that the undercurrent is present during every occupation, with speeds varying between 10 and 40 cm s−1. Transports in the undercurrent were calculated by first extrapolating horizontally to bathymetry with constant velocity. The minimum transport is 0.6 Sv, from AUCE_R (AUCE recovery cruise; Fig. 1, bottom left), while the maximum is 7 Sv, from I5W (Fig. 1, top right).

Fig. 1.

Seven direct velocity sections across the Agulhas Current at 32°S occupied between 1995 and 2005 (chronologically, panels from left to right; top to bottom). The final section in the bottom-right panel is the average velocity structure of the Agulhas Current and Undercurrent from these seven sections. Velocities to the southwest within the Agulhas Current are shaded yellow through red; velocities to the northeast within the undercurrent are shaded white through blue. Black contours mark the zero, −50, and −100 cm s−1 contours. Bathymetry is shaded gray. Dotted lines show station positions: ACE [principal investigator (PI), Bryden]; I5W = WOCE section I5W (PI, Toole); I5_R = I5 repeat (PI, Bryden); and AUCE(PI Beal).

Fig. 1.

Seven direct velocity sections across the Agulhas Current at 32°S occupied between 1995 and 2005 (chronologically, panels from left to right; top to bottom). The final section in the bottom-right panel is the average velocity structure of the Agulhas Current and Undercurrent from these seven sections. Velocities to the southwest within the Agulhas Current are shaded yellow through red; velocities to the northeast within the undercurrent are shaded white through blue. Black contours mark the zero, −50, and −100 cm s−1 contours. Bathymetry is shaded gray. Dotted lines show station positions: ACE [principal investigator (PI), Bryden]; I5W = WOCE section I5W (PI, Toole); I5_R = I5 repeat (PI, Bryden); and AUCE(PI Beal).

To treat each crossing consistently and to merge them to obtain a mean velocity section, all occupations were interpolated onto a 2 km × 10 m grid. The mean was then found by averaging the occupations, on a gridpoint basis, ignoring missing data. This is shown in the bottom-right panel of Fig. 1. The depth of the seabed is taken from bathymetry data collected during AUCE. On average, the undercurrent appears as a slug of northeastward flow hugging the continental slope between 1000-m depth and the foot of the slope at 2900 m, and between 11 and 60 km offshore. Mean maximum velocities are rather small, order 15 cm s−1, with a core at intermediate depth and the suggestion of a second core below 2000 m. The mean undercurrent transport from these seven sections is 2.4 Sv. An enlargement of this mean undercurrent structure, together with a map showing the AUCE mooring locations, is shown in Fig. 2. Note that there has been no spatial smoothing of this merged velocity field and no extrapolation.

3. Mooring data

The undercurrent array consisted of three moorings deployed in March 2003 along the historic 32°S section across the Agulhas Current (Fig. 2, subpanel). The moorings were placed at distances of 18.2 (M1), 26.6 (M2), and 39.9 km (M3) from the coast (coastal origin at 30.32°E, 31.00°S), projected on a line perpendicular to the local continental slope topography (130°T). All instruments were Aanderaa rotary current meters (RCMs), consisting of a paddle rotor to measure speed and a large vein to align the instrument with the current and provide direction. Speed, direction, pressure, and temperature were measured every hour. The moorings were recovered in March 2005, although none of the instruments from mooring M2 were found. On mooring M1 were two RCM7s at nominally 1060- and 1365-m depths; these instruments will be referred to as m1_1100 and m1_1400. On mooring M3 were three RCM8s at nominally 2130-, 2445-, and 2775-m depths; these will be referred to as m3_2200, m3_2400, and m3_2800, respectively. The positions of the moorings and their instruments are shown in Fig. 2, against a background of the mean Agulhas Undercurrent structure.

Blow-down effects were greatest for instrument m1_1100, with four instances of 80-m dips over 3–4 days, corresponding to maxima in northward flow. Changes in temperature associated with these events are of the order of 1°C. For M3, blow-down events are generally much shorter (1–2 days) and smaller in amplitude (mode = 20 m) and associated with temperature changes of 0.2°C. From here on, we assume constant depth for each instrument and discuss only the records of speed and direction. The duration of the records from each instrument varied greatly and none lasted the entire 24 months of deployment. The shallowest instrument, m1_1100 had the shortest record (only 192 days). Both m1_1400 and m2_2200 collected data for over 700 days, while the two deepest instruments, m3_2500 and m3_2800, failed 536 and 436 days after deployment. Figure 3 shows a time series of raw (i.e., no despiking, no filtering, no extrapolation) velocity from each instrument, rotated 50° from north, that is, perpendicular to the orientation of the mooring line. Notice the increase in velocity with depth on each mooring and also the strong vertical coherence, particularly between all records on the M3 mooring. Except for two stalls in the m1_1400 record in August and December of 2004, the records appear clean.

Fig. 3.

Raw records of velocity, rotated 50° from north (alongstream), from each instrument in the undercurrent. Mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths.

Fig. 3.

Raw records of velocity, rotated 50° from north (alongstream), from each instrument in the undercurrent. Mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths.

Table 1 summarizes the statistics from these raw records (with an additional column in brackets for the corrected m1_1400 data, which will be described in the next section). The principal axis of the flow, θ, at each instrument was calculated such that tan2(90 − θ) = 2Puυ/(PuuPυυ), where Puu, Pυυ are the variances of the eastward and northward velocity anomalies (e.g. u′ = uu), respectively, and Puυ is the cross covariance (at zero lag). This gives the principal axis as an angle rotated clockwise from north. The three M3 instruments have principal axes close to 50°, while the flow at the M1 moorings is rotated farther to the east. The variance of the velocity along the principal axes in each case is 5–10 times greater than the cross-stream variance, leading to the conclusion that flow in the undercurrent is strongly aligned in the direction of the local topography, as might be expected. For consistency, and to facilitate the construction of a transport time series later, from hereon we define alongstream as the component of velocity rotated 50° from north, with the cross-stream component 90° to the right of this. With this definition we note that the alongstream velocities for instruments on mooring M1 will slightly underestimate the principal flow (<3% in the mean), as defined by the principal axis. The record-length mean, maximum, and variance of the alongstream component of the flow from each instrument are displayed in Table 1.

Table 1.

Statistics of the velocity records in the Agulhas Undercurrent. Approximate depths of the current meters are given in their names. The second column for m1_1400 shows the statistics for the corrected record, as explained in section 4.

Statistics of the velocity records in the Agulhas Undercurrent. Approximate depths of the current meters are given in their names. The second column for m1_1400 shows the statistics for the corrected record, as explained in section 4.
Statistics of the velocity records in the Agulhas Undercurrent. Approximate depths of the current meters are given in their names. The second column for m1_1400 shows the statistics for the corrected record, as explained in section 4.

Every instrument records a northeastward velocity in the mean, consistent with an undercurrent. Maximum velocities are found close to the seabed, but relatively high on the slope, at m1_1400. Here, the peak speeds are over 90 cm s−1 and the average flow is 20 cm s−1. These are some of the highest speeds ever measured at depth. For instance, peak speeds in the deep western boundary current at 26.5°N in the Atlantic are less than 50 cm s−1 (e.g., Lee et al. 1990; Johns et al. 2008). The weakest velocities are found at m3_2200, 740 m off the bottom, where the average is 5 cm s−1. Clearly, the undercurrent is bottom intensified, but with the strongest flows higher on the slope in the intermediate water layer, while weaker flows are found at the foot of the slope in the deep water layer.

Topographic steering of the flow and its variability is emphasized in Fig. 4, which shows the power spectral densities of the along- and cross-stream velocity variances from each record. Variance is isotropic only in the inertial band and out to a maximum period of 3 days (at m1_1100). At the semidiurnal tidal period, cross-stream variance is less than one-half that of the alongstream. At all periods greater than 3 days the cross-stream variance is negligible. Considering the spectral peaks in Fig. 4, the semidiurnal tidal peak is prominent, although harmonic analysis shows only small tidal currents of 1–2 cm s−1. This is in agreement with estimates of tidal currents by Bryden et al. (2005). The observed inertial peak for all instruments is at 21.6 h (Fig. 4), while the local inertial period at this latitude is 23.2 h. Hence, the effective inertial frequency, feff, is 15% greater than f, evidence of a significant shift in the near-inertial wave band by large, local relative vorticity. Assuming the simplification of Kunze (1985) holds, for example, feff = f + ζ/2 (for ζ ≤ 0.2 f ), then the shift in the inertial peak implies a local horizontal vorticity field of order −1 × 10−5 s−1, which is equivalent to a cross-stream shear of 1 m s−1 over 100 km. This is consistent with the large shears on the cyclonic side of the Agulhas Current core (Beal et al. 2006), where the undercurrent resides.

Fig. 4.

Power spectral density of the along- (black) and cross-stream (gray) velocity variances (unfiltered) at mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths. The alongstream component is rotated 50° from north.

Fig. 4.

Power spectral density of the along- (black) and cross-stream (gray) velocity variances (unfiltered) at mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths. The alongstream component is rotated 50° from north.

Figure 4 shows little evidence for topographic Rossby wave variability, which should be revealed by enhanced cross-stream (perpendicular to topography) variance in the 8–20-day band (Thompson and Luyten 1976). This was further investigated by examining the orientation of variance ellipses for different signal periods (or frequencies) to look for an onshore shift in the principal axis toward shorter periods. And by calculating the lag of the cross-covariance between moorings M1 and M3 to find evidence of features propagating up-slope between moorings. However, we find no clear evidence for topographic Rossby waves in our data, perhaps partly because the moorings are only a small distance apart. There are distinct peaks in alongstream variance shared by all instruments at 4.5, 6.5, and 9.5 days, with the latter being most energetic. These peaks are consistent with the Eady time scale for baroclinic instabilities. The deepest moorings, m1_1400 and m3_2800, also show a peak in variability at time scales of 25 days. A 25-day peak is also exhibited by the Agulhas Current above (Bryden et al. 2005).

At longer periods, all M3 moorings share a dominant peak at 60 days, which is consistent with the meander period found in the Agulhas Current by Bryden et al. (2005). Mooring m1_1400 also has 60-day periodicity, but variabilities at 40- and 25-day periods are of similar power. In the case of m1_1100, peaks at longer time scales are unresolved owing to the short data record. Hence, the velocity field of the undercurrent largely shares the same dominant periodicity as the Agulhas, which corresponds to propagating solitary meanders called Natal pulses (Lutjeharms and Roberts 1988). This result undermines Bryden et al.’s suggestion that the undercurrent appeared to vary independently of the Agulhas. It also implies, if surface and deep flows vary in concert, that Natal pulses must be strongly barotropic in nature, as surmised previously by Lutjeharms et al. (2001) using middepth floats. Results from a regional simulation of the Agulhas corroborate this idea, showing that solitary meanders of the Agulhas are highly barotropic (Biastoch et al. 2009, hereafter BBLC).

4. Correcting m1_1400

Looking closely at the variance of the mooring records, it became clear that there was a problem with the velocity record from instrument m1_1400 (Fig. 3b). The variance over the first half of the record is more than double that over the second half, with the mean flow also dropping off considerably. A tidal analysis of 29-day pieces of the data was conducted using the T_TIDE program (Pawlowicz et al. 2002). This revealed an unnatural downward trend in the amplitude of the M2 tide, with a reduction from about 1.4 cm s−1 at the beginning of the record to 0.75 cm s−1 at the end. However, the trend was not statistically significant, because the results are noisy about such a small signal. Moreover, temporal scales are not well enough resolved by this approach to discern between a slow, steady decrease in amplitude and a more sudden drop. It is also possible that the malfunctioning instrument responded differently at different frequencies of variability. To address these problems, we conducted a wavelet transform of the data.

In a Fourier analysis there is an implicit assumption that signals of variability in a dataset are stationary over time, in regard to both period and amplitude. A wavelet transform expands the analysis to allow variabilities that may evolve over time. In our case, an examination of how the wavelet power over all periods, but particularly at tidal periods, is changing over time will provide a diagnosis of our instrument malfunction. Since the gravity of the moon and the sun is unchanging, power at tidal periods (averaged over the usual cyclical signals) should remain constant throughout our record. Therefore, any persistent changes in amplitude at these periods can be attributed to instrument error. We utilized the wavelet package described by Grinsted et al. (2004) and Moore et al. (2005). Qualitatively, results show a rather abrupt, visible reduction in wavelet power over all periods less than 4 days, about 400 days into the record (not shown). Furthermore, the wavelet pattern of variability appears to be similar throughout the record, despite the drop in power, implying that real signals are still being measured. More quantitative results are shown in Fig. 5.

Fig. 5.

Time-varying wavelet power at the M2 tidal period. Wavelet power is normalized by the time-mean variance of the M2 tide and has been smoothed with a 29-day box filter. Results for m1_1400 are shown in black; results for m3_2200 are shown in gray, for a null comparison. Here, U is the cross-stream component and V is the alongstream component of the flow. An 11-th order polynomial fit for the m1_1400 wavelet power time series is shown in green. The mean tidal power for the first 300 days and final 300 days of the time series are shown in magenta, with error bars given by the sensitivity of means to ±100 days.

Fig. 5.

Time-varying wavelet power at the M2 tidal period. Wavelet power is normalized by the time-mean variance of the M2 tide and has been smoothed with a 29-day box filter. Results for m1_1400 are shown in black; results for m3_2200 are shown in gray, for a null comparison. Here, U is the cross-stream component and V is the alongstream component of the flow. An 11-th order polynomial fit for the m1_1400 wavelet power time series is shown in green. The mean tidal power for the first 300 days and final 300 days of the time series are shown in magenta, with error bars given by the sensitivity of means to ±100 days.

The black lines in Fig. 5 illustrate the wavelet power at the M2 tidal period for m1_1400, with respect to time. There is a distinct reduction in power somewhere between decimal days 300 and 500. We surmise that friction was somehow increasing on the current meter rotor over this period. Both the along- (dashed line) and cross-stream (solid line) components of the flow are similarly affected, indicating that the current meter was able to orient itself with the flow correctly. In contrast, the data from m3_2200 (in gray) appear normal, showing no persistent changes in tidal power over the same period. An 11th-order polynomial fit (in green) to the m1_1400 time series of wavelet power indicates a power drop of approximately 100-day duration, between decimal days 350 and 450. The magnitude of the power reduction is estimated by calculating the mean before and after the drop, with error bars given by changing the period over which the mean is calculated, by ±100 days (Fig. 5, magenta lines). In this way, the power in the first half of the record is 3.2 ± 1 times greater than that in the second half, corresponding to a velocity factor of 1.8 ± 0.2.

Hence, our “best guess” correction for the m1_1400 velocity data is to apply a factor that begins at 1, linearly increases from 1 to 1.8 between decimal days 350 and 450 of the record, and then remains at 1.8 for the rest of the series. The corrected m1_1400 data show some significant changes from the statistics of the raw data, as shown in Table 1. Specifically, the mean velocity at m1_1400 increases by 5 cm s−1 and the variance by 200 cm2 s−2.

Similar analyses of wavelet power at the inertial and subinertial periods gave amplitude factors within ±0.1. The sensitivity of the final undercurrent transports to the m1_1400 correction was investigated by carrying through four different corrections, including our best guess, and calculating the difference in the mean transport between each case. (The method for calculating transport from the current meter data is given in the next section.) Case 1 is a step function at decimal day 400, case 2 is our best guess, case 3 is a longer transition period of 200 days, and case 4 is as in case 2, but with the correction factor increasing to only 1.6, corresponding to the minimum difference from the standard deviation envelopes of the means (dotted magenta lines in Fig. 5). Table 2 gives a summary of the cases and their effects on the undercurrent transports. Transports are given as the difference in transport from our best guess (case 2). The maximum error for any reasonable corrections of m1_1400 is order 0.1 Sv. This includes errors carried through from the extension of the m1_1100 record, which was based on regression with m1_1400 data (see next section).

Table 2.

Sensitivity of undercurrent transports to different corrections for m1_1400.

Sensitivity of undercurrent transports to different corrections for m1_1400.
Sensitivity of undercurrent transports to different corrections for m1_1400.

5. Constructing a time series of undercurrent transport

To combine the mooring data into a transport time series, the data were first cleaned, filtered, and extrapolated in time, where necessary. Gaps in the m1_1400 record (August and December 2004) were linearly interpolated. All records were smoothed with a 40-h, low-pass Butterworth filter to remove tides and inertial oscillations, and resampled at 12-h intervals. For extrapolation of the shortest record, m1_1100, a linear regression with the two longest records, m1_1400 and m3_2200, was performed with a good level of variance explained (regression coefficient, r2 = 0.72). For extrapolation of m3_2500, both m3_2200 and m1_1400 records were used, with the second giving only an incremental improvement to the fit. Finally, m3_2800 was extended using both of the other M3 instruments. In both of these last cases, the variance explained by the linear regressions was high, with r2 > 0.9. The mean velocity at m1_1100 drops to half that given by the short, raw record. Given that the extension of m1_1100 relies on data from m1_1400, this could indicate that the amplitude factor applied as a correction for the latter may be too small. However, we interpret that the undercurrent was simply more intense over the first 6–9 months of the record, as reflected also by a drop in the mean velocity of all three instruments on M3, although to a lesser degree (up to 20% reduction). The resulting filtered and extended time series are shown in Fig. 6. Although each was extended to February 2005, we note that the majority of the records were lost by August 2004 and we use this as our cutoff for calculating a reliable transport time series.

Fig. 6.

Filtered and extended records of velocity, rotated 50° from north (alongstream), for each instrument in the undercurrent. Note that m1_1400 has been corrected, as described in the text: mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths.

Fig. 6.

Filtered and extended records of velocity, rotated 50° from north (alongstream), for each instrument in the undercurrent. Note that m1_1400 has been corrected, as described in the text: mooring 1 at (a) 1100- and (b) 1400-m depths, and mooring 3 at (c) 2200-, (d) 2500-, and (e) 2800-m depths.

Figure 6 further illustrates (as previously seen in Fig. 4) that site M1 appears to have greater variance at shorter periods than M3. The M3 velocities clearly exhibit time scales of order 2 months or so, similar to the meander mode of the Agulhas Current above (Bryden et al. 2005), as mentioned previously. On the other hand, M1 appears to have stronger variability on shorter time scales. This is reflected in the levels of cross covariance, which exhibit coefficients larger than 0.8 between instruments on the same mooring, but drop to about 0.5 between moorings. To speculate, the differences between moorings could be due to topographic (refractive and/or reflective?) effects on the incoming barotropic waves associated with the meander mode. Or to the superposition of baroclinic variability at M1, as a result of heaving of the high-shear interface between the undercurrent and the Agulhas Current above, onto the largely barotropic meander mode.

To calculate undercurrent transports, it was necessary to devise a method to combine the individual instrumental records into a coherent cross section of the undercurrent. This was not a simple matter, owing to the loss of M2, which leaves a rather sparse array where no single depth horizon has greater than one data point. First, a “fake” instrumental record was created 100 m off the bottom at M1, assuming a data record exactly the same as that from m1_1400. Since we have seen that the undercurrent is bottom intensified, this represents a conservative estimate of the peak speeds here. Next, the current meter measurements at each site, M1 and M3, were vertically interpolated using a shape-preserving, cubic spline (Akima 1970). These vertical profiles were then placed on a 1 km × 20 m grid with bathymetry masked, similar to the grid created previously to obtain a mean lowered acoustic Doppler current profiler (LADCP) section across the undercurrent, as shown in Fig. 2. The zero isotach from this mean section, which depicts the average position of separation between the undercurrent and the Agulhas (as best as we can tell from all the available LADCP data), was smoothed to give an undercurrent “envelope.” In the region outside this envelope, the grid was filled with a static field of (smoothed) mean velocities as seen in Fig. 2. Finally, inside the undercurrent envelope, the field was allowed to fluctuate at each time step, using objective mapping to fill the grid points between the vertical current profiles at each mooring site and the zero isotach envelope. In this way, a cross-sectional snapshot of the undercurrent was created at each time step, as shown in Fig. 7. Transports are then calculated as the flow integrated over the mean undercurrent envelope (between the bathymetry and the blue line in Fig. 7).

Fig. 7.

Three snapshots of the gridded, objectively mapped current meter data in the Agulhas Undercurrent. Velocities above the blue line are essentially a frozen field, although small adjustments can occur with the objective analysis. Below the blue line, the velocity field changes at each time step based on the current meter data, whose position and intensity are illustrated by the shading in the blue circles. (top) A typical undercurrent structure of average intensity (14 Jan 2005). (middle) The case of a strong undercurrent (maximum transport; 29 Mar 2004). (bottom) A case when the undercurrent is strongly reversed (minimum transport; 21 Apr 2004). Southwest velocities are in shades of yellow through red. Northeast velocities with the undercurrent are shown in shades of blue. Bathymetry is shaded in gray.

Fig. 7.

Three snapshots of the gridded, objectively mapped current meter data in the Agulhas Undercurrent. Velocities above the blue line are essentially a frozen field, although small adjustments can occur with the objective analysis. Below the blue line, the velocity field changes at each time step based on the current meter data, whose position and intensity are illustrated by the shading in the blue circles. (top) A typical undercurrent structure of average intensity (14 Jan 2005). (middle) The case of a strong undercurrent (maximum transport; 29 Mar 2004). (bottom) A case when the undercurrent is strongly reversed (minimum transport; 21 Apr 2004). Southwest velocities are in shades of yellow through red. Northeast velocities with the undercurrent are shown in shades of blue. Bathymetry is shaded in gray.

The correlation length scales used for the objective mapping were 40 km in the horizontal and 1500 m in the vertical: the scale of the undercurrent. Mapping was conducted using the influence of the 100 nearest points, to reduce the size of the inversion matrix. Errors were set low (1 cm s−1) for the current meter data and unrealistically high (10 ms−1) for the static LADCP data, to allow the frozen field to adjust at each time step. Even so, as the second panel in Fig. 7 shows, the Agulhas field is too fixed and does not allow the undercurrent to push the zero isotach up and down enough to approximate the true velocity field well. Nor, in the other extreme, does it allow the Agulhas to penetrate into the undercurrent envelope and merge with the current meter data during periods when there is no undercurrent. The movie snapshots are inevitably limited in there realism of the complete velocity field, because we have only five data points at every time step. However, since we are confining ourselves to calculating transport within the average undercurrent envelope, the variability of transport is captured reasonably well. The robustness of the transports estimated in this way was tested by constructing another, very simple, transport time series by integrating the current meter velocities over boxes of height equal to the distance between instruments and of width equal to the distance between moorings. The patterns of variability were the same, but the mean of the “simple” transport was larger by almost a Sverdrup, because of the overestimate of the height of the undercurrent above m1_1100 and some overlap with bathymetry in the m1_1400 box.

The resulting transport time series is given in Fig. 8 for the period February 2003–August 2004, a duration of 552 days. The mean Agulhas Undercurrent transport and its standard deviation is 4.2 ± 5.2 Sv. We estimate the standard error for this mean as 0.6 Sv, by estimating the independence time of the transport time series. We use the integral of the autocorrelation coefficient out to a lag of 20 days (the position of the first zero crossing), giving an independence time of 7 days. Standard error is then given by the standard deviation divided by the square root of the number of independent samples in our record (552/7). Hence, although the variability is high, the transport of the mean undercurrent is highly significant. Our undercurrent transports show variability that is considerably larger than estimated by Bryden et al. (2005), but the mean transport is the same. This agreement is remarkable when one considers that Bryden et al. (2005) summed transports only to 2400-m depth and included northward transport anywhere across their Agulhas array. Here, we define the undercurrent as only flow over the slope, inshore of the Agulhas. The maximum undercurrent transport was 13.9 Sv in May 2003 and the minimum was −11.4 Sv in April 2004.

Fig. 8.

Time series of Agulhas Undercurrent transport (black), of northeastward transports (gray), and of transport below 2000 m (blue). The latter curve depicts the transport of NADW within the undercurrent. Black and gray dots show the corresponding transports from coincident LADCP sections across the undercurrent. Values for the mean and standard deviation of undercurrent and NADW transports are shown at bottom left.

Fig. 8.

Time series of Agulhas Undercurrent transport (black), of northeastward transports (gray), and of transport below 2000 m (blue). The latter curve depicts the transport of NADW within the undercurrent. Black and gray dots show the corresponding transports from coincident LADCP sections across the undercurrent. Values for the mean and standard deviation of undercurrent and NADW transports are shown at bottom left.

Defining the undercurrent as only northeastward (positive) flow gives the gray line and a mean transport elevated by 1–2 Sv. This is a typical definition for the undercurrent when using one-time data, such as the historical LADCP sections (Fig. 1). This gray curve shows that although the variability of the undercurrent is large, there are only 11 occasions when there is no northward flow at any of the current meters. This occurs for less than 10% of the time in our record, with each event typically lasting 5 days or less, except for two events that lasted 8 and 13 days in December 2003 and April 2004, respectively.

The gray circles in Fig. 8 are the northeastward transports from two LADCP occupations of the undercurrent in February 2003 (AUCE and AUCE2 from Fig. 1). Each of these occupations resolves the undercurrent better than the array, with three or four stations showing northward flow at depth. The northward LADCP transports match the array transports to within a few tenths of a Sverdrup. In black are the transports from these occupations calculated in the exact same way as for our transport time series—that is, velocity integrated over the region of the mean undercurrent (as constrained by the mean zero isotach; shown in blue in Fig. 7). In this case there is some discrepancy between the array and AUCE transports, we think owing to the vertical movement of the high shear interface between the Agulhas and the undercurrent, which is not resolved by the current meter array. In this way, the array will tend to underestimate undercurrent variability. The transport time series is cut off, owing to instrument failure, before the reoccupation of the section during the mooring recovery cruise in March 2005; hence, there are no comparisons to be made with LADCP data toward the end of our time series.

The blue curve in Fig. 8 shows the transport within the NADW layer of the undercurrent, that is, the transport below 2000 m. This is of particular interest because of the implications for the ventilation of the deep Indian Ocean via a direct influx of NADW around South Africa and northward via the Agulhas Undercurrent. The transport and standard deviation is 2.3 ± 3.0 Sv, suggesting that the undercurrent does provide a route for the influx of deep water (the standard error of this mean is 0.4 Sv). The magnitude is about half that postulated by Bryden and Beal (2001), who lacked mooring data below 2400 m, but is in agreement with the hydrographic investigations of Arhan et al. (2003) and van Aken et al. (2004). To the north of our array, shoaling topography means that this water must upwell or turn to the east if it is to reach the Mozambique Basin and continue into the interior of the Indian Ocean.

A histogram of daily undercurrent transports shows that the distribution is strongly skewed and bimodal (Fig. 9). An undercurrent of 10–11 Sv or a weak undercurrent of 1 Sv or less is a more common occurrence than the “average.” This bimodal distribution is repeated when transports are defined as northeastward only. However, for NADW transport, the distribution is also skewed to high transports, but with only one peak or mode at 5 Sv. To speculate, the bimodal distribution could be a result of the dominant meander mode. Low transports occur when the Agulhas meanders onshore and “squeezes” the undercurrent, while high transports occur when it meanders offshore. The deep transport could be less affected by the meandering, since the Agulhas, and hence the horizontal velocity shears, are much weaker at depth. These kinds of motions are studied carefully in a companion paper of a regional model that simulates the Agulhas and its undercurrent (BBLC).

Fig. 9.

Histogram of the distribution of daily transport values. Undercurrent transports are shown in cyan, while NADW transports are in blue.

Fig. 9.

Histogram of the distribution of daily transport values. Undercurrent transports are shown in cyan, while NADW transports are in blue.

The spectrum of variability of undercurrent transport is shown in Fig. 10. Spectral averaging tends to shift long periodicities to higher frequencies and so we show results from three different spectral techniques. The smoothest is the Welch spectrum (Welch 1967), which indicates a dominance of mesoscale variability in the broad 40–80-day band. The multitaper method (Percival and Walden 1993) illustrates a similar story, while the raw periodogram suggests separated peaks at 40, 70, and 150 days, although these are less statistically robust. All methods show several peaks in the 4–20-day band, similar to those seen previously in the individual instrument spectra (Fig. 4) and a robust periodicity at 25 days. Overall, the meander mode centered at about 60 days dominates the transport variability in the undercurrent. This is the same mode that dominates the variability of the velocity field of the Agulhas Current, although Bryden et al. (2005) concluded that it was not a major contributor to the Agulhas transport variability.

Fig. 10.

Spectra of the Agulhas Undercurrent transport time series, calculated using three different methods: periodogram (gray), multitaper (dark gray), and Welch (black). All spectra are plotted in variance-preserving form, where the power spectral density is calculated as the frequency × spectral density.

Fig. 10.

Spectra of the Agulhas Undercurrent transport time series, calculated using three different methods: periodogram (gray), multitaper (dark gray), and Welch (black). All spectra are plotted in variance-preserving form, where the power spectral density is calculated as the frequency × spectral density.

6. Discussion and summary

In previous work, Beal and Bryden (1997), Donohue et al. (2000), and Beal et al. (2006) noted that water properties within the undercurrent are not detectably different from those outside in the Agulhas, at any depth. Yet, in this study we find a mean northward transport of waters at NADW depth within the undercurrent. Such a transport is justified, in the sense that NADW must exit the Atlantic as part of the global thermohaline circulation and evidence for a small portion of NADW to flow around South Africa and northward into the Natal Valley has been put forward from previous observations. Arhan et al. (2003) deduced a budget from a number of World Ocean Circulation Experiment (WOCE) sections, suggesting that 2 Sv flows northward (while another 9 Sv flows eastward as part of the Agulhas Return Current) and remnants of NADW have been observed within the Mozambique Basin and Channel farther to the east and north of our section (van Aken et al. 2004). Why, then, is there no water mass signature? The implication is that flow within the undercurrent is far more “leaky” than that within the Agulhas, where anomalous water mass signatures are clear. Indeed, cross-stream mixing has been examined by Beal et al. (2006), who found that at NADW depth there is not a strong enough dynamical barrier to trap water masses. Waters are constantly entrained and detrained into and out of the undercurrent, producing a relatively well-mixed environment along isopycnals.

A further impediment to the direct transport of NADW northward is the bathymetry of the Natal Valley, which blocks the flow of deep waters to the north below 2250 m. NADW in the undercurrent must either recirculate to the south within the Natal Valley, escape eastward into the Mozambique Basin through breaks in the Mozambique Plateau, or vigorously upwell. One LADCP section 200 km north of our array hints that a weak undercurrent still exists, but is displaced offshore of the Agulhas over the shoaling bathymetry (Casal et al. 2009). Ultimately then, the path of NADW toward the Mozambique Channel is likely to be convoluted and not via a continuous flow, trapped in the undercurrent.

At intermediate depths, a mean northward transport as found here is somewhat confusing, since at these depths there is a mixture of Antarctic Intermediate Water (AAIW) and Red Seawater (RSW) within the undercurrent (Beal and Bryden 1997; Donohue et al. 2000). AAIW circulates into the Agulhas from the north and east via the south Indian Ocean subtropical gyre, while RSW crosses the equator to enter the Agulhas from the north. Therefore, both of these water masses must be flowing toward the south within the Agulhas Current system on average. Indeed, analyzing three hydrographic sections of the Agulhas Current and Undercurrent, Donohue et al. (2000) found this to be the case: the net flow of both RSW and AAIW was southward. Hence, it must be the case that there is also a large amount of entrainment and detrainment of these water masses into and out of the undercurrent locally. Beal et al. (2006) found signatures of cross-frontal fluxes in the form of vigorous interleaving between AAIW and RSW, and deduced that stirring by large meanders (i.e., Natal pulses) is the dominant mechanism for detrainment at intermediate depth. A Lagrangian analysis of the undercurrent would clearly be instructive to investigate just how leaky the flow is. Such an analysis has been performed in a model study of the undercurrent and can be found in a companion paper (BBLC).

We have found that the dominant form of variability in the undercurrent matches that of the Agulhas: a 50–70-day meander mode related to so-called Natal pulses. Within the undercurrent, this mode is strongest at the deepest mooring at 2800 m. Hence, we can conclude that these meanders are highly barotropic, a feature previously alluded to by theory (de Ruijter et al. 1999) and middepth observations (Lutjeharms et al. 2001), but not measured until now. Interestingly, Bryden et al. (2005) found that while variance in the velocity field of the Agulhas was dominated by Natal pulses, its transport variability was not. Yet, for the undercurrent, we find that the meander mode dominates both in the individual velocity records and in the transport time series. The reason for this discrepancy could be owing to the nature of a Natal pulse. By examining the Agulhas movie of Bryden et al. (2005) we find that, during a meander event, the Agulhas moves first onshore, then offshore, and finally back onshore, with moderate weakening and then strengthening of its transport by 10–15 Sv (well within one standard deviation, which is 21.5 Sv). At the same time, the undercurrent is “squeezed out” at the slope as the Agulhas moves onshore, then strengthens and penetrates upward as the Agulhas moves offshore, finally returning to “normal” as the Agulhas comes back to its mean position. Hence, while the Agulhas meanders back and forth, largely maintaining its transport, the undercurrent is stemmed as northward velocities over the slope are replaced by the southward velocities of the Agulhas and then intensified as the “envelope” between the Agulhas and the slope opens up again. Unfortunately, we do not have time series that resolve both the Agulhas and the undercurrent simultaneously and cannot be sure about this description. However, previous LADCP sections, the Agulhas moorings, and the undercurrent measurements shown here point to this kind of picture. Once again, further study of these mechanisms was made by analysis of a model (BBLC).

In summary, we find that the Agulhas Undercurrent is a robust feature, carrying intermediate water and NADW northeastward in the mean, below the southwestward-flowing Agulhas Current. It exhibits some of the highest velocities measured at depth, with peak speeds of 90 cm s−1 below 1000 m. From a 552-day record, the undercurrent as a whole has a transport of 4.2 ± 5.2 Sv, while the NADW layer has a transport of 2.3 ± 3.0 Sv. Only less than 10% of the time is the undercurrent completely missing. For an Indian Ocean overturning transport in the region of 10 Sv (McDonagh et al. 2008), the Agulhas Undercurrent represents about 20% of the ventilation of the deep Indian Ocean.

Acknowledgments

We gratefully acknowledge the generosity of Harry Bryden and the mooring team at the National Oceanography Centre, Southampton, for the loan of current meters and the fabrication of the AUCE moorings, whose data are analyzed herein. Devising a correction for the current meter time series with a slowing rotor benefited from discussions with Bill Johns. We thank the officers and crew of R/V Melville for their assistance and professionalism. This work was supported by NSF Grant OCE-0244769.

REFERENCES

REFERENCES
Akima
,
H.
,
1970
:
A new method of interpolation and smooth curve fitting based on local procedures.
J. Assoc. Comput. Mach.
,
17
,
589
602
.
Arhan
,
M.
,
H.
Mercier
, and
Y-H.
Park
,
2003
:
On the deep water circulation of the eastern South Atlantic Ocean.
Deep-Sea Res. I
,
50
,
889
916
.
Beal
,
L. M.
, and
H. L.
Bryden
,
1997
:
Observations of an Agulhas Undercurrent.
Deep-Sea Res. I
,
44
,
1715
1724
.
Beal
,
L. M.
, and
H. L.
Bryden
,
1999
:
The velocity and vorticity structure of the Agulhas Current at 32°S.
J. Geophys. Res.
,
104
,
5151
5176
.
Beal
,
L. M.
,
T. K.
Chereskin
,
Y. D.
Lenn
, and
S. K. Y.
Elipot
,
2006
:
The sources and mixing characteristics of the Agulhas Current.
J. Phys. Oceanogr.
,
36
,
2060
2074
.
Biastoch
,
A.
,
L. M.
Beal
,
J. R. E.
Lutjeharms
, and
T. G. D.
Casal
,
2009
:
Variability and coherence of the Agulhas Undercurrent in a high-resolution ocean general circulation model.
J. Phys. Oceanogr.
,
39
,
2417
2435
.
Bryden
,
H. L.
, and
L. M.
Beal
,
2001
:
Role of the Agulhas Current in Indian Ocean circulation and associated heat and freshwater fluxes.
Deep-Sea Res. I
,
48
,
1821
1845
.
Bryden
,
H. L.
,
L. M.
Beal
, and
L. M.
Duncan
,
2005
:
Structure and transport of the Agulhas Current and its temporal variability.
J. Oceanogr.
,
61
,
479
492
.
Casal
,
T. G. D.
,
L. M.
Beal
,
R.
Lumpkin
, and
W. E.
Johns
,
2009
:
An inverse model of the Agulhas Current system during February–March 2003: Structure and downstream evolution.
J. Geophys. Res.
,
114
,
C03001
.
doi:10.1029/2008JC004954
.
de Ruijter
,
W. P. M.
,
P. J.
van Leeuwen
, and
J. R. E.
Lutjeharms
,
1999
:
Generation and evolution of Natal pulses: Solitary meanders in the Agulhas Current.
J. Phys. Oceanogr.
,
29
,
3043
3055
.
Donohue
,
K. A.
,
E.
Firing
, and
L.
Beal
,
2000
:
Comparison of three velocity sections of the Agulhas Current and Agulhas Undercurrent.
J. Geophys. Res.
,
105
,
28585
28593
.
Ganachaud
,
A. S.
,
C.
Wunsch
,
J.
Marotzke
, and
J.
Toole
,
2000
:
Meridional overturning and large-scale circulation of the Indian Ocean.
J. Geophys. Res.
,
105
, (
C11
).
26117
26134
.
Grinsted
,
A.
,
J. C.
Moore
, and
S.
Jevrejeva
,
2004
:
Application of the cross wavelet transform and wavelet coherence to geophysical time series.
Nonlinear Processes Geophys.
,
11
,
561
566
.
Johns
,
W. E.
,
L. M.
Beal
,
M. O.
Baringer
,
J. R.
Molina
,
S. A.
Cunningham
,
T.
Kanzow
, and
D.
Rayner
,
2008
:
Variability of shallow and deep western boundary currents off the Bahamas during 2004–05: Results from the 26°N RAPID-MOC array.
J. Phys. Oceanogr.
,
38
,
606
623
.
Kunze
,
E.
,
1985
:
Near-inertial wave propagation in geostrophic shear.
J. Phys. Oceanogr.
,
15
,
544
565
.
Lee
,
T. N.
,
W.
Johns
,
F.
Schott
, and
R.
Zantopp
,
1990
:
Western boundary current structure and variability east of Abaco, Bahamas at 26.5°N.
J. Phys. Oceanogr.
,
20
,
446
466
.
Lutjeharms
,
J. R. E.
, and
H. R.
Roberts
,
1988
:
The Natal pulse: An extreme transient on the Agulhas Current.
J. Geophys. Res.
,
93
,
631
645
.
Lutjeharms
,
J. R. E.
,
O.
Boebel
,
P.
van der Vaart
,
W.
de Ruijter
,
T.
Rossby
, and
H.
Bryden
,
2001
:
Evidence that the Natal pulse involves the Agulhas Current to its full depth.
Geophys. Res. Lett.
,
28
,
3449
3452
.
McDonagh
,
E. L.
,
H. L.
Bryden
,
B. A.
King
, and
R. J.
Sanders
,
2008
:
The circulation of the Indian Ocean at 32°S.
Prog. Oceanogr.
,
79
,
20
36
.
Moore
,
J.
,
A.
Grinsted
, and
S.
Jevrejeva
,
2005
:
New tools for analyzing time series relationships and trends.
Eos, Trans. Amer. Geophys. Union
,
86
,
24
.
Pawlowicz
,
R.
,
B.
Beardsley
, and
S.
Lentz
,
2002
:
Classical tidal harmonic analysis including error estimates in MATLAB using T_TIDE.
Comput. Geosci.
,
28
,
929
937
.
Percival
,
D. B.
, and
A. T.
Walden
,
1993
:
Spectral Analysis for Physical Applications: Multitaper and Conventional Univariate Techniques.
Cambridge University Press, 583 pp
.
Smith
,
W. H. F.
, and
D. T.
Sandwell
,
1997
:
Global seafloor topography from satellite altimetry and ship depth soundings.
Science
,
277
,
1957
1962
.
Thompson
,
R. O. R. Y.
, and
J. R.
Luyten
,
1976
:
Evidence for bottom-trapped topographic Rossby waves from single moorings.
Deep-Sea Res.
,
23
,
629
635
.
van Aken
,
H. M.
,
H.
Ridderinkhof
, and
W. P. M.
de Ruijter
,
2004
:
North Atlantic Deep Water in the south-western Indian Ocean.
Deep-Sea Res. I
,
51
,
755
776
.
Welch
,
P. D.
,
1967
:
The use of fast Fourier transform for the estimation of power spectra: A method based on time averaging over short, modified periodograms.
IEEE Trans. Audio Electroacoust.
,
15
,
70
73
.
Wyrtki
,
K.
,
1971
:
Oceanographic Atlas of the International Indian Ocean Expedition.
National Science Foundation, Washington, DC, 531 pp
.

Footnotes

Corresponding author address: L. M. Beal, Rosenstiel School of Marine and Atmospheric Science, University of Miami, 4600 Rickenbacker Cswy., Miami, FL 33149. Email: lbeal@rsmas.miami.edu