1. Background
Three-quarters of the world's oceanic volume, represented by deep and bottom water masses, has its properties set at its last contact with the atmosphere in extremely limited areas, where deep convection occurs (Gascard 1991). While the importance to thermohaline circulation and global climate of this substantial volume is difficult to overestimate, the processes responsible are not well understood, mostly because deep convection occurs for limited times in remote locations with bad weather. As observations have increased, so has the understanding of processes involved (Gascard 1991). A hierarchy of scales involved in the process of convection has become recognized (Killworth 1983; Clarke and Gascard 1983; Schott et al. 1993): the gyre scale O(100 km), the mesoscale O(10 km), and the plume scale O(1 km).
On the mesoscale, Killworth (1983) postulated that homogenized patches of water of order 10 km extent could be preconditioned to experience deep convection by having circulation with cyclonic shear and the associated doming of the isopycnals that would encourage deep convection to drive through remaining stratification, producing a well-mixed, homogeneous “chimney” of water. The mesoscale is also understood to be of critical importance for exporting the convected product away from its point of origin and for the eventual capping (restratification) of the mixed layer, effectively ending convection (Marshall and Schott 1998).
The small size of plumes [1 km (Gascard and Clarke 1983) or 500–800 m as estimated by Visbeck (as cited in Marshall and Schott 1998)] has made direct observation difficult. In theory (Maxworthy and Narimousa 1994; Jones and Marshall 1993), high surface heat flux causes the formation of a thin thermal boundary layer that feeds rapidly descending plumes of dense water. The upward return flow consists of broad areas of slowly ascending warm water.
The interaction between the plume scale and the mesoscale has been the focus of increasing study and scrutiny. The “upright” convection described above, with buoyancy fluxes driving purely vertical motion, assumes no horizontal variability in the background stratification. While a region of cyclonic shear on the mesoscale may act to contain a volume of water with low horizontal and vertical variability, its boundary will naturally be a place of horizontal density gradients and stratification.
The interplay of horizontal gradients and high heat flux yields several interesting results theoretically: slantwise convection (Haine and Marshall 1998; Straneo et al. 2002), the arresting of mixed layer depth increase by lateral heat flux through the boundary (Legg and Marshall 1993; Send and Marshall 1995; Visbeck et al. 1996; Legg et al. 1996, 1998; Legg and McWilliams 2001), and the eventual break-up of the mesoscale feature through baroclinic instability (Marshall et al. 1993; Legg et al. 1998; Legg and McWilliams 2001). In addition, the release of potential energy during convection provides kinetic energy for both vertical and horizontal motions. Legg and McWilliams (2001) showed that with convection, a preexisting mesoscale field will become energized. This tendency toward greater mesoscale kinetic energy during convection has been noted in observations (Lilly et al. 1999; Prater 2002; Brandt et al. 2003, hereinafter BMS).
The energizing of the mesoscale field provides a further mechanism for water property modification: stirring of convectively modified water with the surrounding fluid. Legg and McWilliams (2001) noted that a wintertime cruise (during active convection) revealed substantial spatial inhomogeneity, yet a summer cruise found a homogeneous water mass. They concluded that vigorous stirring provided by convection-energized eddies mixed the water to a homogeneous mass. Thus, to account for and predict the properties of the final product, one must include not only the effects of vertical buoyancy flux on a preconditioned patch of water, but also the incorporation laterally of the properties of less well ventilated water.
Observationally, variability is seen on all scales in the wintertime Labrador Sea. On the basin scale, O(100 km), Pickart et al. (2002) found a 2°C temperature range (2.6°–4.6°C), and 0.1 psu (34.8–34.9 psu) salinity range, similar to those reported by D'Asaro and Steffen (1998) (2.5°C, 3°–5.5°C and 0.15 psu, 34.74–34.91 psu). On the mesoscale (∼20 km) Lilly et al. (1999, 2003, hereinafter LRSLL) reported variability of 0.3°C and 0.03 psu. On the plume scale O(1 km) Steffen and D'Asaro (2002) found 0.02°C temperature difference between upwelling and downwelling water parcels. This study will focus on variability between the plume and mesoscales.
Clearly eddies are extremely important to the modification and exportation of convected water. Unfortunately, given their small scale, observations of eddies in convective regions have been extremely limited. A recent dataset from the long-term mooring occupation of former Ocean Weather Station Bravo (Lilly et al. 1999; Lilly and Rhines 2002; J. Lilly 2002, unpublished manuscript, hereinafter LIL; LRSLL) has allowed resolution of mesoscale eddies of about 20-km radius at times when convection is not occurring. Studies in the Greenland Sea (Gascard et al. 2002; Wadhams et al. 2002) have found long-lived eddies of 5-km radius.
Satellite imagery can provide synoptic views occasionally (given cloud cover) during convection. However, as reported in Prater (2002) and BMS, only moderately strong signatures can be discerned with radii of about 10 km and strong SSH displacement or temperature departures. Utilizing these data combined with data from floats, Prater (2002) and BMS located a region of eddy formation off the west coast of Greenland. Prater found eddy pairs injected into the central Labrador Sea, with the anticyclone being warmer and the cyclone being colder. However, LRSLL noted a strong asymmetry in the data at Bravo, with only 2 of 33 eddies found exhibiting cyclonic rotation. An explanation of this phenomenon may be the preferential breakup of cyclonic structures through the development of baroclinic instabilities when exposed to convection (Legg and McWilliams 2001).
This paper will describe data collected through intensive field work in the Labrador Sea northwest of the Bravo site during the convective season of 1998 and will focus on observations spanning the mesoscale to the plume scale. Both cyclonic and anticyclonic features were recorded, with anticyclones being larger and showing evidence of expanding during the observations. These small-scale, O(1–10 km), features will be found to be very efficient mixers and to exert a weak influence on convection.
2. Data and processing
The low stratification and large-scale cyclonic gyre of the central Labrador Sea precondition the area to experience wintertime deep convection as cold, dry air sweeping off the North American continent causes strong air–sea heat fluxes. In light of its proclivity toward deep convection, the Labrador Sea was selected as the focus of a large, multidisciplinary study, the Labrador Sea Deep Convection Experiment (Lab Sea Group 1998). Mooring deployments, intensive CTD surveys (some during winter), a variety of float types, focused modeling efforts, and extensive meteorological studies were coordinated during the winters of 1997 and 1998 (the 1996/97 and 1997/98 seasons) in an effort to gain a better understanding of deep convection. This study will utilize data collected as part of this experiment by two types of floats (DLF and PALACE) as well as shipboard measurements (intake and CTDs) taken during the winter of 1998 to investigate the interaction of convection and horizontal variability on scales from the mesoscale, O(10–100 km), down to the plume scale, O(<1 km).
a. The deep Lagrangian float in the Labrador Sea
A total of 24 fully Lagrangian floats were deployed in the Labrador Sea during the winters of 1997 and 1998 in a region thought to experience the deepest wintertime convection (see Fig. 3a and Lab Sea Group 1998). This location was selected as having an optimal combination of low stratification, intense heat flux, and enough distance from the boundary current to prevent the floats from being ejected from the region. Indeed, isobaric drifters and hydrographic surveys found this area experienced the deepest mixed layers in 1997 (Lavender and Davis 2002; Pickart et al. 2002). Steffen and D'Asaro (2002) focused on the vertical aspects of convection experienced by these floats. The current work will focus on data from the 1998 deployment (26 January–24 March) of these floats, 5 of which obtained high quality horizontal tracking (accuracy of <1 km) that allows the complex eddy field in this region to be resolved.
Deep Lagrangian floats (DLFs) are designed to follow water motions in three dimensions and thus be close to fully Lagrangian (see D'Asaro et al. 1996; D'Asaro 2003; and Fig. 1). This is accomplished by having the same buoyancy as the surrounding water and by having a high drag. The floats have a specially designed aluminum hull, which survives pressures up to 2000 dbar and has almost the thermal expansion rate of seawater and nearly the same compressibility. Final adjustments to the compressibility are made using a motor-driven piston capable of up to a 25-cm3 change in volume (∼1.6 kg m−3 change in density). A 1-m2 cloth drogue provides drag, keeping the float in phase with the surrounding water. The drogue is initially folded, in a low drag configuration, and is opened after a weeklong autoballasting routine is completed. The floats sample temperature and pressure at 5-min intervals, measuring temperature with millidegree accuracy and pressure to about a decibar. A RAFOS receiver (Rossby et al. 1986) allows determination of horizontal position at 4-h intervals. The RAFOS processing for this study, described by Steffen and D'Asaro (2003), has been slightly modified from that described by Steffen and D'Asaro (2002). This modification uses more of the raw sound source delays (the data are subjected to a less stringent rejection criteria) and a slightly varying speed of sound (just over 2 m s−1 in 30 days). The resulting horizontal float positions have a relative accuracy of about 500 m, with errors dominated by the digitization accuracy of the RAFOS receiver. The mission length is approximately 2 months after which the floats surface by dropping a weight and transmit their data via the Argos satellite system.
An example of records from one 1998 float is shown in Fig. 1. The float sank after deployment, recording a temperature profile as it did so. It then went through its weeklong autoballast cycle, lightened itself, and rose. Mixed layer depths were shallow in early 1998 (about 200 dbar), and the float stabilized below the mixed layer. It was eventually entrained into the convecting layer, evidenced by the substantial reduction in temperature at about yearday 50 and subsequent, nearly continuous vertical motions. As discussed by Steffen and D'Asaro (2002) these vertical motions are convective cycling since they are driven by the surface buoyancy flux. At the end of the mission, the float performed a (non-Lagrangian) profile. Comparison of DLF temperature data with near-simultaneous CTD measurements found agreement to within the 0.01°C statistical scatter.
b. PALACE data
Over 200 Profiling Autonomous Lagrangian Circulation Explorers (PALACE) were deployed (Lavender and Davis 2002) during the winters of 1997 and 1998. These isobaric floats drift for several days at a prescribed depth (400 or 700 m) and then surface, communicate via satellite, and return to depth, thereby constructing profiles of temperature and salinity on a preprogrammed cycle of 3.5–20 days. They were intensively deployed in the same area as the DLFs. Records from PALACE floats within an approximately 200-km box surrounding the 1998 DLF missions, for the first 100 days of 1998, were provided by K. Lavender. These data comprised 74 profiles of temperature and 69 profiles of salinity from 14 floats. PALACE measurements have an expected accuracy of 5 dbar in pressure, 0.005°C in temperature, and 0.01 psu in salinity (Davis et al. 2001). Steffen and D'Asaro (2003) describe corrections for instabilities in these PALACE salinity measurements: a drift in the sensors of ∼0.06 psu in 80 days was isolated and removed. In addition, the first profiles of each PALACE float (when the salinity sensor was found to behave erratically) were discarded (along with profiles from PALACE floats that recorded two or fewer profiles). After this manipulation, 49 profiles of corrected salinity from seven PALACE floats remained for use in this study. PALACE temperature data agreed with nearby CTD measurements to within the 0.02°C statistical scatter.
c. Ship-based observations
Intensive CTD surveys were conducted by the R/V Knorr from 25 January to 12 February 1998. As the Knorr performed its surveys, intake temperature and salinity values were recorded. These can provide a near-synoptic estimate of sea surface variability. Knorr engine seawater intake logs were at times simultaneous with CTD profiles. Comparison with the corresponding CTD records in the region of interest (see Fig. 2) reveals the intake to register 0.03°C warm and 0.22 psu salty, and as such, corrections of these magnitudes were applied to the intake records. Accuracy of the CTD measurements was expected to be 0.0005°C and 0.001 psu (D'Asaro and Steffen 1998).
3. Analysis
Comparison of these diverse data, collected over a period of months, required three horizontal coordinate systems: physical space (no adjustments to position), drift adjusted position (positions after removal of the mean advective velocity), and eddy position (positions relative to the centers of eddylike features). In addition, because these data were collected during active convection, significant entrainment occurred during the record with a generally deepening, warming, and salifying mixed layer. This temporal evolution needs to be removed to address horizontal variability and, as described in section 3b, was used to construct coordinates detrended by mean mixed layer properties (depth, temperature, salinity, and density).
a. Horizontal reference frames
1) Real space
As seen in Fig. 3a, the DLFs drifted generally to the west during their 1998 deployment, moving on average about 100 km from their deployment location. In this coordinate system, patterns in the trajectories are difficult to discern.
2) Advected coordinates
Comparisons between the PALACE, ship-based, and DLF observations will be made in an advected coordinate system. Assuming a constant rate of advection (excluding the premission autoballast cycle) yielded a DLF mean velocity (drift rate) of 1.1 ± 0.3 cm s−1 west and 0.1 ± 0.3 cm s−1 south for the DLFs. Removing this velocity causes the tangled DLF tracks visible in Fig. 3a to collapse and reveal a coherent anticyclonic eddy as seen at the center of Fig. 2. A slight adjustment to this mean velocity, to 1.1 cm s−1 west and 0.2 cm s−1 south, causes alignment of a density anomaly and the anticyclone (without the adjustment, the anomaly is centered about 2 km from the center of the anticyclone). Attempts to confirm the rate of average drift with the PALACE data failed because the PALACE velocity errors for this relatively short time (as compared with their cycling period) were several centimeters per second.
3) Eddy coordinates
High-pass filters applied to the DLF tracks localized eddylike features of varying size, rotation rate, and duration (their characteristics are summarized in Fig. 3b). The position of the “eddy core” is defined as the running mean position of the DLF, with an averaging window the length of one eddy period. For example, DLF 22 (shown in red in Fig. 3a) encountered an anticyclone with a 5-day period (loops associated with this anticyclone can be found at 170 km west, 350 km north at about yearday 55 and at 140 km west, 345 km north at about yearday 70). The “core,” the 5-day running mean position for DLF 22, can be seen in Fig. 3b as the red “x”s. This core track aligns with the tracks of anticyclone(s) defined by the running mean positions of several other DLFs (blue and magenta “o”s, cyan, blue, and magenta “x”s), implying that these floats encountered the same anticyclone. An average anticyclone core track can be constructed by combining these individual tracks (Fig. 3c, shown in red). Using this path as the translating origin (Fig. 4a), the DLF tracks clearly show a circular anticyclonic eddy. Similar application of running-mean portions allows definition of the other eddy cores seen in Fig. 3c and resolution of the features seen in Figs. 4c,e and 5a,c,e. This method has limitations; particularly problematic is definitive establishment of eddy size and temperature signal relative to its surroundings since eddy “cores” could only be defined when DLFs were clearly in an eddy. It is therefore difficult to study times when a float entered or left an eddy.
Average paths for four other features (all cyclones) encountered by the DLFs are also shown in Fig. 3c, with circles representative of the eddy size shown along each track. As will be discussed later, A1's size increased during the observations, thus A1's track has two representative eddy sizes on its path. An eddy from the Labrador Sea boundary current, (float 404 in Prater 2002), a mooring-observed central Labrador Sea eddy (Lilly and Rhines 2002), Greenland Sea eddies (Gascard et al. 2002), and a meddy (D'Asaro et al. 1994) are included for comparison.
b. Mixed layer evolution
The horizontal structure of the mixed layer can only be resolved for these data after the removal of the large temporal variations caused by air–sea fluxes and convection. Average or “background” values for mixed layer temperature, salinity, density, and depth were therefore calculated (Steffen and D'Asaro 2003). Average temperature, salinity, and density within the convecting mixed layer were calculated for the CTD and PALACE data at the time of each profile, defined as the average from the water column 50 dbar below the surface to 30 dbar above the mixed layer base (Steffen and D'Asaro 2003). The (non-Lagrangian) profiling portions of the DLF data were treated similarly to obtain estimates of average temperature at the beginning and end of the DLF records. These profiling data were also used to estimate the depth of the mixed layer at the times of the profiles. Mixed layer depths were also estimated from the Lagrangian portions of the DLF records by excluding data from outside the mixed layer (determined by temperature) then selecting the maximum pressure recorded in a running 5-day window. For example, mixed layer depth estimates from DLF 33's record are shown in Fig. 1 (see also Steffen and D'Asaro 2003).
Figure 6 shows the evolution of mixed layer properties derived from all data available within a 200-km box surrounding the 1998 DLF missions. Data of each type were averaged in 5-day bins. The average values from all data sources were then averaged. For example, in a 5-day bin with 4 PALACE profiles, 200 intake data points, and 3 CTD casts, an average PALACE, intake, and CTD value were determined. These three (equally weighted) values were then averaged. For mixed layer depth, the trend so determined is shown as the solid line in Fig. 6a.
c. Departure coordinates
A new depth coordinate zd is defined as depth departure from background, zd = z −
This coordinate transformation assists in sorting the structure of A1. For example, when float 22 first encountered A1 (days 43–50) it was below the mixed layer base. In Fig. 7a this is seen as the 3°C water at 400 dbar. In real space this overlies data from later in the record (days 58–59) when the water was 2.85°C. In addition, temperatures at the core of the eddy (data collected around day 65) seems to be much colder than data collected during the floats' return to the core (after day 78). By looking at departures instead (Fig. 7b), patterns of eddy temperature distribution are much clearer. The very warm water encountered prior to day 50 is now seen to be below the mean mixed layer depth (0 dbar). The cold core of the eddy is also much clearer since the general warming of the mixed layer has been accounted for.
4. Results
a. Mesoscale O(10 km)
Figure 8 shows the departures of observed mixed layer properties from the background values defined by the fits in Fig. 6. Data prior to 26 January were excluded because of a lack of reliable definitions of “background” during this period.
Mixed layer depth departure (Fig. 8d) shows substantial (>300 dbar) variability, but does not have enough resolution to associate this variability with mesoscale structures. In contrast, the temperature, salinity, and density fields clearly show the 20-km warm core eddy explored in LRSLL at 30 km east and 30 km south in the advected coordinate system. This strong signal, 0.47°C warmer and 0.059 psu saltier than background levels, makes the region of the DLF missions seem homogeneous by comparison. However, there is structure on a smaller scale (both in horizontal extent and magnitude) within this relatively homogeneous region.
1) Mesoscale velocity structures
Eddy A1 was the dominant mesoscale velocity feature resolved. Its properties appear to have varied with time. It was encountered by all the DLFs (Fig. 4a). DLF 22, (shown in red) remained within A1 until the end of its mission. Figure 9a shows A1's azimuthal velocity as a function of radius. Prior to yearday 60, the azimuthal velocity can be seen to reach a maximum of 4–6 cm s−1 at 5–6 km. After yearday 61, only DLF 22 remains in A1. During this period the eddy appears larger and has a higher velocity (the maximum velocity is not definitively reached at 10–12-km radius). We therefore denote the later stages of the anticyclone A1′.
The transition from A1 to A1′ occurs after DLF 33 (shown in black in Fig. 4a), which had been rotating at a radius of 5 km, exits A1. DLF 22 (shown in red in Fig. 4c) then makes two circuits at 10–12-km radius. During the same period (yeardays 68–72) the temperature increased by 0.07°C. The fact that this warming occurred on a time scale similar to the eddy rotational period suggests that warmer water was stirred into the eddy, perhaps because of merging with another anticyclone.
Figure 12 divides the azimuthal velocity data from each eddy into two categories (above and below the mean mixed layer depth). The same fits to azimuthal velocity with radius found in Fig. 9 are plotted on both panels for each eddy in Fig. 12. Within the scatter of this data, no dependence on depth is apparent. It is important to note that these velocities are smoothed over one-half of an eddy period (which ranges from 0.75 to 3 days). Because a mixed layer transit averaged 0.8 days (Steffen and D'Asaro 2002), any vertical variability would also be averaged out. The data were found to be noisy enough that no patterns were discernible without applying the running mean.
Anticyclone A1 appears to be one of several similar features. As seen in Fig. 10 the sea surface density shows a substantial number of features with density anomalies of order 0.005 kg m−3 and horizontal extents of 5–10 km, marked with L for light core features and D for dense core features. In this advected coordinate system, the light feature at the origin (L1) aligns with the core of A1, and while this specific alignment may be coincidental, the association of A1 with such a light core feature is plausible.
The anticyclone A1 was thus likely associated with a density signal (or one similar to one of those) recorded by the shipboard survey. That survey revealed the region was populated by similar scale features, some with oppositely signed density anomalies, and each of which likely had a velocity signal similar to that directly measured by the DLFs.
b. Submesoscale O(1–10 km)
The DLFs encountered four small-scale (1–4 km) features for more than one eddy period (Figs. 4d and 5a,c,e). These features were all cyclonic with vorticities ranging from 0.2f to 1.3f (see Figs. 9b,c). Two of these small-scale features were found to be less than 2 km in horizontal extent (P1 and P2 for plumes 1 and 2) while the other two were 3–4 km in horizontal extent (C1 and C2 for cyclones 1 and 2). DLFs experienced three of these four cyclones both within and beneath the convecting layer (the fourth, P2, with the strongest vorticity sampled, was only experienced within the convecting layer). Despite the near–plume scale of the small cyclones, no association of the rotation with convective cycling was found.
c. Dispersion
This study reveals a rich eddy field on scales of 1–10 km, composed of both cyclones and anticyclones. Wadhams et al. (2002) and Gascard et al. (2002) hypothesize that Greenland Sea convective product is trapped within 5-km anticyclonic eddies. Similarly (though the eddies are larger) LRSLL and Lilly and Rhines (2002) find anticyclones with convected water seemingly trapped in their cores. Unlike these examples, the DLF data reveals an easy communication between and within small-scale structures. All the DLFs entered A1, all but one exited, and all but that one encountered other eddies (several encountered multiple other eddies). Water also seems free to move within the structures: DLF 22 was at the edge of A1 during yeardays 45–55 (Fig. 4a), found itself in the middle of A1 about yearday 60, circuited the edge of A1′ during yeardays 67–75 (Fig. 4c) and migrated back to the middle of A1′ just before the DLF mission ended.
These small features produce ample horizontal mixing. After accounting for mean drift, water parcels tagged by Lagrangian floats spread an average 26 km from their mission starting positions and an average 30 km from one another in 40 days, implying a horizontal diffusivity rate of 230–350 m2 s−1. This estimate agrees well with the range of 300–600 m2 s−1 deduced using tracer data from the central Labrador Sea by Khatiwala et al. (2002), implying that the majority of the mixing could be accomplished by small-scale features.
d. Slant convection
Figures 4f and 5 show the float trajectories in radial coordinates for the four small cyclonic eddies encountered. In three of the four (P2, C1, and C2, all in Fig. 5) the floats move outward from the eddy center as they move upward, then downward toward the eddy center. These convective trajectories therefore move along a circular cone, avoiding the top of the center of the eddy. In the fourth, P1 (Fig. 4f), little convective motion was recorded. Since the periods of vertical motion, driven by convection and horizontal motion circulating in the eddy, are comparable, the trajectories make one to two vertical transits during an eddy circulation period, resulting in three-dimensional tracks similar to those sketched in Fig. 11. Although the data are limited, this suggests an interaction between the eddy and the convection. One possibility is that light water is trapped on the surface in the eddy core. This would inhibit vertical convection in the core, forcing the trajectories into the conical shape and thus also allowing the fresh lens to persist.
This motion is similar to that predicted by theories of slant convection. In the presence of velocity shear or a horizontal component of the Coriolis force, Haine and Marshall (1998) and Straneo et al. (2002) predict that the plumes will align along nonvertical absolute momentum surfaces. This mechanism provides for deeper mixed layers than simple one dimensional models. Straneo et al. (2002) find about 15% error in depth estimates for values comparable to conditions in the Labrador Sea during convection between a model including slantwise convection and one where convection aligns entirely vertically. This difference could imply a substantial error in estimating the overall volume of water ventilated by a given meteorological forcing.
5. Discussion
a. Features
A variety of features were encountered in this study, from cyclones with 1-km radius to an anticyclone with 20-km radius (the warm core eddy explored in LRSLL). Previous studies have found few cyclones (LRSLL), and those cyclones encountered were much larger structures (with radii on the order of 10 km; LRSLL; Prater 2002; BMS). Four cyclones with 1–5-km radius were identified in this study.
Convective plumes are thought to be cyclonic in structure (Marshall and Schott 1998) and O(1 km) in size. Despite the small size of cyclones P1 and P2 (the smallest features resolved), these were not plumes because both downwelling and upwelling occurred in both features. It is unclear if the DLFs could resolve true plume vorticity given the 4-h interval for horizontal positioning. The time scale of convection is about 0.8 day to traverse the mixed layer each way (see Steffen and D'Asaro 2002), providing, at best, five positions as parcels ascend or descend. With a tracking accuracy of 500 m, a rich eddy field masking smaller features, and plumes apparently smaller than 1 km it would be difficult to resolve the expected vorticity signature of plumes with this instrumentation.
1) Origin of A1
LRSLL identifies two origins for anticyclones in the central Labrador Sea. Features containing convected water, generated locally, have a temperature signal on the order of 0.3°C colder than ambient water, while eddies shed from the Irminger Current on the west coast of Greenland are on the order of 1°C warmer than the ambient water. The anticyclones described in this study had significantly weaker signals than those seen at the Bravo mooring. However, LRSLL find the 20-km-radius warm core eddy seen in Fig. 8a likely to be of Irminger origin.
The anticyclone A1 does not fit either pattern. As seen in Fig. 4d, A1′ has a noticible rim temperature anomaly 0.05°C warmer than the core of the eddy. However, as can be seen in Fig. 4b, when DLF33 exited A1 (shown in black, around yearday 61), there was little temperature signal (perhaps there is a slight suggestion of a warm rim 0.01°C warmer than the core of the eddy). Both A1 and A1′ show a very slightly cold core (0.02°C and 0.01°C colder than background, respectively). The 5-km anticyclones described by Wadhams et al. (2002) and Gascard et al. (2002) in the Greenland Sea were determined to contain convected product and were 0.1°C colder than the surrounding water, again with no suggestion of a rim with a different character than the rest of the eddies. Anticyclone A1 does not behave quite like previously described eddies in this region. Perhaps part of the difference is that A1 underwent significant changes within the period of study.
Wadhams et al. (2002) and Gascard et al. (2002) find long-lived 5-km radius anticyclones in the Greenland Sea to persist nearly unchanged for months. Unlike these, A1 at least doubled in size, warmed, and sped up in a few days time, perhaps because of merging with another anticyclone. The late stage warm rim of A1′ may be a vestige of an Irminger ring that collided with a locally generated cold eddy, mixing incompletely before the end of this record.
b. Depth scale of eddies
All but one of the eddies had velocity signatures both within and below the convecting layer (P2 was sampled only within the convecting layer), and therefore these features extended across the convecting layer and into the underlying ocean. No variability of velocity with depth was detectable with this data. Simple thermal wind estimates found a substantial barotropic component of the velocity, consistent with other observations of anticyclonic eddies. Wadhams et al. (2002) find a Greenland Sea anticyclone to penetrate deeper than 2000 m. Similarly, LRSLL find anticyclones advected by the Bravo mooring to have expressions deeper than 2500 m.
c. Eddies and convection
Although convection has been theorized to occur preferentially in regions of cyclonic shear (Killworth 1983; Schott et al. 1993; Marshall and Schott 1998), the DLFs in this study experienced convection within both cyclones and anticyclones. The deepest mixed layers encountered by the DLFs occurred within an anticyclone; however, the DLF missions ended before the end of the convective season.
All DLFs, as seen in Fig. 4a, encountered A1. However, the location at which the floats encountered this feature had a significant impact on the timing of their first experience of convection. DLF 33, which happened to encounter the center of A1, was entrained first. DLFs 23, 27, and 30 circuited A1 3 km from the center, exited A1, and waited 15 days longer before they were all entrained. Thus a separation of a few kilometers resulted in 15 extra days of convective modification for one parcel of water due to the placement of its encounter with A1.
d. Where do all the cyclones go?
Remote generation of eddies, with the exportation of Irminger water into the central Labrador Basin has been observed to be a symmetric process, with cyclone/anticyclone pairs ejected from the west coast of Greenland (Prater 2002). Localized generation of eddies, through the action of convection, has also been theorized (Julien et al. 1999; McWilliams 1984) to be a symmetric process, generating equal numbers of cyclones and anticyclones.
However, previous studies (LRSLL; Legg and McWilliams 2001) have noted an asymmetry in the distribution, with anticyclones strongly outnumbering cyclones. BMS, combining satellite and surface drifter data, found 79 eddies in a 3-yr Labrador Sea record, 30 of which are cyclones. They find that the cyclones tend to be weaker (in azimuthal velocity) and smaller than the anticyclones. (Of the six eddies found with diameters less than 20 km, four are cyclones.) One proposed mechanism to explain this discrepancy is the preferential breakup of cyclones exposed to convection due to baroclinic instabilities (Legg et al. 1998; Legg and McWilliams 2001). Despite a paired formation they predict that cyclonic eddies become baroclinically unstable under the action of convection, causing them to break apart into progressively smaller cyclones, eventually dissipating (leading to a lack of observations). Anticyclones, however, remain stable, free to merge and grow as in Legg and McWilliams (2001) and Arai and Yamagata (1994). Such a mechanism could explain a paired formation and the lack of observations.
Our observations support the Legg and McWilliams (2001) hypothesis. The rapid warming, expansion, and deepening of A1/A1′ may have been caused by the merging of two anticyclones. If so, this may be an example of the predicted preferential expansion of anticyclones and demise of cyclones. Several small-scale cyclones were found, all much smaller than the anticyclones encountered. Observations at the Bravo mooring could not have detected these small cyclones because of their small size and because the mooring record is too noisy during convection to detect eddies (LRSLL; Lilly et al. 1999; Lilly and Rhines 2002; LIL).
6. Summary
Variability occurred on all scales in the Labrador Sea during the winter of 1998. The basin scale had a 2.5°C temperature range (3–5.5°C) and a salinity range of 0.15 psu (34.74–34.91 psu) (D'Asaro and Steffen 1998). On the mesoscale, the 20-km eddy explored in LRSLL was 0.47°C warmer and 0.059 psu saltier than background levels. Smaller features with radii of 1–10 km showed much weaker temperature signatures, on the order of 0.1°C.
Several small-scale eddies were resolved in this study with radii from 1 to 10 km. These eddies extended throughout the depths sampled (deeper than 700 m in some cases), but their full vertical extent could not be resolved with this data. The velocities sampled were not depth dependent, within the resolution of the data. Although cyclones of O(1 km) were resolved in this data, none were consistent with being convective plumes.
This study found only a weak interaction of convection with the small eddy structures resolved. Convection was not confined to preconditioned patches within cyclones. In fact, the deepest mixed layer depth departures resolved occurred within an anticyclone (A1′). In addition, water encountering the core of A1 underwent 15 days of convective modification not experienced by water at its periphery. At least in these very weak structures, convection is not trapped within cyclones.
Eddies were found to evolve during the convective season. Whether it was produced by the convection, energized by it, or just passing through, anticyclone A1 (the only feature tracked for more than 20 days) experienced a significant degree of modification, even when the generalized effects of convective modification were accounted for. It warmed and deepened faster than the trend in the region sampled: A1 also grew horizontally from a radius of 5 km to at least 12 km and sped up from 5 cm s−1 to 18 cm s−1 in less than a week.
This behavior (growth of an anticyclone) and the numerous small cyclones found are consistent with expectations that anticyclones will grow during convection while cyclones will break into progressively smaller pieces, eventually dissipating (Legg and McWilliams 2001; Arai and Yamagata 1994).
Previous studies of small scale anticyclones (Gascard et al. 2002; Wadhams et al. 2002) found them to persist relatively unchanged over months, trapping convected water within them. In contrast to these previous observations, this study found an easy communication within and between eddies. In particular, all five DLFs encountered A1 and, rather than becoming trapped within it, all but one DLF escaped. This communication allows the water to effectively mix at small scales, providing horizontal diffusion estimates (250–350 m2 s−1) with the same magnitude as basin-scale estimates.
Horizontal variability has been predicted to produce slant convection (Haine and Marshall 1998; Straneo et al. 2002). At least two of the four small cyclones found here appeared to distort the convective trajectories from their normal, upright alignment to a slanted orientation (Fig. 11) similar to that predicted. However, present theory required an unrealistic velocity shear in order to reproduce the observed slope of the trajectories. Whether another process was responsible for the nonvertical convective pathways is unclear.
Variability on scales larger than plume scale have been long predicted to have substantial influence on the creation and exportation of convected water masses. The small, weak, horizontal structures explored here produce substantial horizontal mixing, but show little capacity to confine either convection or water to particular structures. Some evidence of slant convection was found; however, theoretical predictions were inconsistent with these observations.
Acknowledgments
The authors express their appreciation to Fiamma Straneo and Sonya Legg for helpful advice on theoretical interpretation; to Kara Lavender for PALACE data; and to Tom Sanford, Kathie Kelly, and two anonymous reviewers whose recommendations significantly improved the manuscript. The authors also gratefully acknowledge support from the University of Washington Department of Oceanography, as well as ONR Contracts N00014-94-1-0930, N00014-94-1-0024, and N00014-94-1-0025.
REFERENCES
Arai, M., and T. Yamagata, 1994: Asymmetric evolution of eddies in rotating shallow water. Chaos, 4 , 163–175.
Brandt, P., C. Martins, and F. Schott, 2003: Seasonal to interseasonal and variability in the Labrador Sea from satellite altimetry and surface drifters. J. Geophys. Res., in press.
Clarke, R., and J. Gascard, 1983: The formation of Labrador Sea water. Part I: Large-scale processes. J. Phys. Oceanogr., 13 , 1764–1778.
D'Asaro, E. A., 2003: Performance of autonomous Lagrangian floats. J. Atmos. Oceanic Technol., 20 , 896–911.
D'Asaro, E. A., and E. Steffen, 1998: KN156 CTD and float deployments January 26–Febrary 8, 1998. WHOI Tech. Rep.
D'Asaro, E. A., S. Walker, and E. Baker, 1994: Structure of two hydrothermal megaplumes. J. Geophys. Res., 99 , 20361–20373.
D'Asaro, E. A., D. Farmer, J. Osse, and G. Dairiki, 1996: A Lagrangian float. J. Atmos. Oceanic Technol., 13 , 1230–1246.
Davis, R. E., T. Sherman, and J. Dufour, 2001: Profiling ALACEs and other advances in autonomous subsurface floats. J. Atmos. Oceanic Technol., 18 , 982–993.
Gascard, J., 1991: Open ocean convection and deep water formation revisited in the Mediterranean, Labrador, Greenland, and Weddell Seas. Deep Convection and Deep Water Formation in the Oceans, P. Chu and J. Gascard, Eds. Elsevier, 157–182.
Gascard, J., and R. Clarke, 1983: The formation of Labrador Sea water. Part II: Mesoscale and smaller-scale processes. J. Phys. Oceanogr., 13 , 1779–1797.
Gascard, J., A. J. Watson, M. Messias, K. Olsson, T. Johannessen, and K. Simonsen, 2002: Long-lived vortices as a mode of deep ventilation in the Greenland Sea. Nature, 416 , 525–527.
Haine, T., and J. Marshall, 1998: Gravitational, symmetric, and baroclinic instability of the ocean mixed layer. J. Phys. Oceanogr., 28 , 634–658.
Jones, H., and J. Marshall, 1993: Convection with rotation in a neutral ocean: A study of open-ocean deep convection. J. Phys. Oceanogr., 23 , 1009–1039.
Julien, K., S. Legg, J. McWilliams, and J. Werne, 1999: Plumes in rotating convection. Part 1: Ensemble statistics and dynamical balance. J. Fluid Mech., 391 , 151–187.
Khatiwala, S., P. Schlosser, and M. Visbeck, 2002: Rates and mechanisms of water mass transformation in the Labrador Sea as inferred from tracer observations. J. Phys. Oceanogr., 32 , 666–686.
Killworth, P., 1983: Deep convection in the world ocean. Rev. Geophys. Space Phys., 21 , 1–26.
Lab Sea Group, 1998: The Labrador Sea deep convection experiment. Bull. Amer. Meteor. Soc., 79 , 2033–2058.
Lavender, K., and R. Davis, 2002: Observations of open-ocean deep convection in the Labrador Sea from subsurface floats. J. Phys. Oceanogr., 32 , 511–526.
Legg, S., and J. Marshall, 1993: A heton model of the spreading phase of open-ocean deep convection. J. Phys. Oceanogr., 23 , 1040–1056.
Legg, S., and J. McWilliams, 2001: Convective modifications of a geostrophic eddy field. J. Phys. Oceanogr., 31 , 874–891.
Legg, S., H. Jones, and M. Visbeck, 1996: A heton perspective of baroclinic eddy transfer in localized open ocean convection. J. Phys. Oceanogr., 26 , 2251–2266.
Legg, S., J. McWilliams, and J. Gao, 1998: Localization of deep convection by a mesoscale eddy. J. Phys. Oceanogr., 28 , 944–970.
Lilly, J., and P. Rhines, 2002: Coherent eddies in the Labrador Sea observed from a mooring. J. Phys. Oceanogr., 32 , 585–598.
Lilly, J., P. Rhines, M. Visbeck, R. Davis, J. Lazier, F. Schott, and D. Farmer, 1999: Observing deep convection in the Labrador Sea during winter 1994/95. J. Phys. Oceanogr., 29 , 2065–2098.
Lilly, J., P. Rhines, F. Schott, K. Lavender, J. Lazier, U. Send, and E. D'Asaro, 2003: Observations of the Labrador Sea eddy field. Progress in Oceanography, in press.
Marshall, J., and F. Schott, 1998: Open-ocean convection: Observations, theory and models. MIT Center for Global Change Science Tech. Rep. 52, 158 pp.
Marshall, J., J. Whitehead, and T. Yates, 1993: Laboratory and numerical experiments in ocean convection. Ocean Processes on Climate Dynamics: Global and Mediterranean Examples, P. Malanotte-Rizzoli and A. Robinson, Eds., Kluwer Academic, 437 pp.
Maxworthy, T., and S. Narimousa, 1994: Unsteady, turbulent convection into a homogeneous, rotating fluid, with oceanographic application. J. Phys. Oceanogr., 24 , 865–887.
McWilliams, J., 1984: The emergence of isolated coherent vortices in turbulent flow. J. Fluid Mech., 146 , 21–43.
Pickart, R., D. Torres, and R. A. Clarke, 2002: Hydrography of the Labrador Sea during active convection. J. Phys. Oceanogr., 32 , 428–457.
Prater, M., 2002: Eddies in the Labrador Sea as observed by profiling RAFOS floats and remote sensing. J. Phys. Oceanogr., 32 , 411–427.
Rossby, T., D. Dorson, and J. Fontaine, 1986: The RAFOS system. J. Atmos. Oceanic Technol., 3 , 672–679.
Schott, F., M. Visbeck, and U. Send, 1993: Open ocean deep convection, Mediterranean, and Greenland Seas. Ocean Processes on Climate Dynamics: Global and Mediterranean Examples, P. Malanotte-Rizzoli and A. Robinson, Eds., Kluwer Academic, 203–225.
Send, U., and J. Marshall, 1995: Integral effects of deep convection. J. Phys. Oceanogr., 25 , 855–872.
Steffen, E., and E. D'Asaro, 2002: Deep convection in the Labrador Sea observed by Lagrangian floats. J. Phys. Oceanogr., 32 , 475–492.
Steffen, E., and E. D'Asaro, 2003: Data processing for central Labrador Sea winters of 1997–98: PALACE, CTD, ship intake, and DLF records. University of Washington Applied Physics Laboratory Tech. Memo. TM4-03, 16 pp.
Straneo, F., M. Kawase, and S. Riser, 2002: Idealized models of slantwise convection in a baroclinic flow. J. Phys. Oceanogr., 32 , 558–572.
Visbeck, M., J. Marshall, and H. Jones, 1996: Dynamics of isolated convective regions in the oceans. J. Phys. Oceanogr., 26 , 1721–1734.
Wadhams, P., J. Holfort, E. Hasen, and J. P. Wilkinson, 2002: A deep convective chimney in the winter Greenland Sea. Geophys. Res. Lett., 29 , 1–4.
(top) Potential temperature record with time for DLF 33. (bottom) Pressure record with time for DLF 33 (solid) with mixed layer depth estimate (dashed). Shaded gray regions indicate non-Lagrangian, end-of-mission profile. The mixed layer depth estimate was made by excluding data collected outside the mixed layer (utilizing temperature as an indicator) and then selecting the maximum pressure recorded within a 5-day window. Inset shows a DLF; hull diameter is 19 cm.
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Advectively adjusted positions for data used in this analysis shown using grayscale based on the time of data collection. PALACE data are shown as stars, CTD data are squares, Knorr intake are dots (combining to form grayscale lines), DLF data collected within the mixed layer are diamonds, and DLF tracks are solid lines. The arrow shows the direction of the drift (mean) velocity. This drift rate, 1.1 cm s−1 west and 0.2 cm s−1 south, was removed from the data, adjusting positions to yearday 35. A record collected in real space along the arrow on the day denoted by the numbers would appear at to (yearday 35) in the advected reference frame. Dashed circle shows the position of the mesoscale eddy discussed in LRSLL
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Maps of DLF trajectories; numbers are yearday color coded to tracks. (a) Float tracks, each of five DLFs shown in different color. Day ticks are shown as small diamonds; day labels are color coded to floats. The right panel shows the location of the Bravo mooring (dot) and the area shown in the left panels (black rectangle). (b) Tracks of eddy centers as described in text. Each float's encounter with an eddylike feature is shown as a different symbol color pair as outlined in the legend. Column of the legend are DLF number (float), yearday of first encounter (start), yearday of exit (stop), and period of feature in days (period). (c) Corresponding centers of eddies to be explored further, with circles representative of the size of the eddies. Several other features are included for comparison: “Lab Sea Bound” is a cyclone from the Labrador Sea Boundary current described by Prater (2002). “Cent Lab Sea” is from mooring data in the central Labrador Sea (LRSLL). “Green Sea” is from the Greenland Sea (Wadhams et al. 2002; Gascard et al. 2002). “Meddy” is a meddy (D'Asaro et al. 1994). Here ζ/f is defined as 2υ/rf, where υ is the maximum azimuthal velocity and r is the radius at which that velocity occurs
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
(left) DLF tracks relative to translating centers of eddies shown in Fig. 3c, with colors and day labels as in Fig. 3a. (right) Vertical sections radially through the eddies, with positions shown as dots of color representing temperature departure from background. Thin lines connecting the positions and yearday labels are shown in the color of the DLF as in the left-hand panels (and Fig. 3a). Vertical coordinate is shown as departure from background mixed layer depth, i.e., depth below mean mixed layer base
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
As in Fig. 4
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Mixed layer properties. As in Fig. 2, dots are from Knorr intake, stars are from PALACE floats, squares are from CTDs, crosses mark data collected during the Lagrangian portion of DLF mission (within the mixed layer), and diamonds are from DLF (non-Lagrangian) profiles. Solid line is “background” from which departures will be calculated, constructed as described in section 3. (a) Mixed layer depth, (b) mixed layer potential temperature, (c) mixed layer salinity, and (d) mixed layer density
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Radial and depth positions of DLF 22 with temperatures in grayscale. Yeardays shown as numbers. (a) Vertical coordinate is pressure, and grayscale is potential temperature; (b) vertical coordinate is depth departure from “background” mixed layer depth, and grayscale is temperature departure from “background.”
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Departures from background levels at advectively adjusted positions. Data are marked as in Fig. 2. Mesoscale eddy described in LRSLL is shown as a dashed ellipse. (a) Temperature anomalies, (b) salinity anomalies, (c) density anomalies, and (d) mixed layer base anomalies.
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Velocity, smoothed over one-half eddy cycle, as a function of radius, color coded by depth departure from background mixed layer depth. Thin colored lines and small yearday numbers are color coded by DLF as in Fig. 3a. Dashed black lines are fits assuming solid body rotation (the same lines are shown in Fig. 12). (a) Anticyclones' velocity: A1 (yeardays 42–63.5) shown as circles and A1′ (yeardays 58–79) shown as triangles. (b) Small cyclones' velocity: P1 shown as squares and P2 shown as stars. (c) Larger cyclones' velocity: C1 shown as triangles and C2 shown as diamonds
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Density departures on the submesoscale (a portion of Fig. 8c); gray arrow indicates the magnitude of advective drift removed. Examples of light and dense features are marked L and D, respectively. Shaded blue region is A1
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Schematic diagram of idealized three-dimensional convective trajectories, with a small cyclonic eddy the size of P2. Here water parcels make two mixed layer transits in one eddy revolution. An idealized convective trajectory with no eddy is shown to the left and right of the lens of light water. Note that the aspect ratio of this figure is approximately true
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
Velocities by radius for all DLFs, divided into within and below the mixed layer. Visually selected simple fits for velocity appear on both panels of each eddy. These fits are the same as those found in Fig. 9. Data are coded by symbol to the DLFs
Citation: Journal of Physical Oceanography 34, 1; 10.1175/1520-0485(2004)034<0044:MASSOA>2.0.CO;2
