During the SEMANE 2000 experiment southwest of Portugal, two meddies were found in near contact. These meddies had hydrological radii of about 20 and 30 km, thickness of 900 m, maximum temperatures of 12.45°C and 13.45°C, and maximum salinities of 36.52 and 36.78. The smaller meddy with more pronounced thermohaline anomalies was clearly double cored (at 750 and 1300 m depths) while the wider one was more diffuse and more homogeneous. The associated geostrophic velocities (referenced at 2000 m) locally reached 0.5 m/s in the smaller meddy, and 0.2 m/s in the wider one. Three RAFOS floats and two deep-drogued surface drifters, seeded in the two meddies, rapidly gathered in the more intense meddy. This meddy trajectory, revealed by the float motion, was first eastward, then southward. Maps of sea level anomaly indicate that this motion did not correspond to the long-term evolution of the initial positive sea level anomaly signature of the meddies, and that neighboring cyclones must have played a role in the meddy evolution. To determine the role of each eddy in the observed evolution, several scenarios were studied with a three-layer quasi-geostrophic numerical model. The interaction of two meddies in isolation did not result in the observed meddy trajectories on the long term. The interaction of these two meddies with successive neighboring cyclones provided a more realistic trajectory of the meddy containing the floats.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The Mediterranean Sea undergoes intense warming and strong wind bursts which result in substantial evaporation. Warm and salty water masses, such as the Levantine Intermediate Water, mix with Upper Deep Western Mediterranean Water, to flow out of the basin into the Atlantic Ocean, through the Strait of Gibraltar; fresher Atlantic water enters the Mediterranean Sea at the surface to close the mass budget. This export of warm and salty water into the Gulf of Cadiz is first materialized by a flow exiting the Straits and cascading down the continental slope. It then veers right along the Iberian continental slope under the influence of the Coriolis force [Madelain, 1970; Zenk, 1975]. As mixing with surrounding waters acts, this water which becomes Mediterranean Water in the Atlantic Ocean (hereafter MW), finally stabilizes as two deep currents near 8°W, 36°30′N; they flow westward at 850 and 1250 m depth. Downstream, these two currents encounter the Portimão Canyon (near 8°30′W, 36°30N) and then Cape Saint Vincent and its canyon (near 9°30′W). From there, the undercurrents head north. Topographic accidents are known to generate perturbations on these two currents, often leading to the formation of mesoscale eddies of MW, called meddies [Armi and Zenk, 1984; Armi et al., 1989; Prater and Sanford, 1994; Bower et al., 1995, 1997; Sadoux et al., 2000; Chérubin et al., 2000; Serra and Ambar, 2002]. As the two deep currents of MW proceed north, they encounter the Estremadura Promontory where they also form meddies [Käse et al., 1989; Zenk et al., 1992] and further north, Cape Ortegal which is presently the northernmost identified meddy formation site along the Iberian Peninsula [Paillet et al., 1999, 2002].
 Observations of deep eddy interactions in the ocean, and in particular of like-signed eddies, are not numerous. The interaction of two eddies has been observed near the East Australian Current [Creswell, 1982]; these eddies having cores at different depths, this process was closer to vertical vortex alignment than to merger. A single detailed measurement of meddy merger was carried out in 1991 in the Iberian Basin [Schultz-Tokos et al., 1994]. During this experiment, two meddies with similar thermohaline properties, but slightly different sizes (radii of 25 and 35 km), were first surveyed hydrologically and then followed with RAFOS floats. The float trajectories indicated that these two meddies merged and migrated before colliding with the Josephine Seamount. Another observation of meddy interaction was obtained in 1999 when two meddies and a cyclone of Mediterranean water were observed to interact south of Portimão Canyon [Carton et al., 2002]. One meddy was strongly coupled with the cyclone, forming a coherent baroclinic dipole which propagated for several months (as tracked by deep-drogued surface buoys). The initial hydrological network evidenced that this dipole was tearing a long filament of warm and salty water from a second meddy located west of this dipole and drifting southwestward. This interaction was also modeled numerically to determine the origin and fate of the three eddies.
 From a theoretical point of view, the interaction of two like-signed eddies, and their possible merger, has been studied at length, both analytically and numerically, in 2D and in stratified rotating flows [see, e.g., Wu et al., 1984; Melander et al., 1988]. The end product of such an interaction is a larger vortex surrounded by a smaller vortex, and/or by vorticity filaments.
 Here, we investigate the structure of two meddies, and their probable interactions with neighboring cyclones, southwest of Portugal, during the SEMANE 2000 experiment; we compare these interactions to previous observations and to model results. This comparison will reveal several novel features.
 The paper is organized as follows: section 2 recalls the data collection and processing. Section 3 presents the meddy hydrological and dynamical structures. Section 4 details the meddy evolution via float trajectories. In section 5, model simulations investigate the mechanisms underlying this two-meddy interaction, and the subsequent interactions with neighboring cyclones.
2. Data Collection and Processing
2.1. Data Collection
 The SEMANE (Mediterranean Water Outflow in the North Eastern Atlantic Ocean) program was conducted by SHOM (French Navy Hydrographic and Oceanographic Service) to measure the currents of Mediterranean Water and their variability near the continental slope southwest of Portugal. Hydrological transects, float deployment and current measurements were performed across the continental slope and extending into the deep ocean (especially in the Gulf of Cadiz) to evidence meanders of the currents and meddies. The interaction between two meddies and a cyclone mentioned above was observed during SEMANE 1999.
 The SEMANE 2000 experiment (15–28 November 2000) was aimed at renewing four current meter moorings near Cape Saint Vincent and Portimao Canyon, at deploying two sound sources in the Gulf of Cadiz and at achieving a CTD-LADCP section along 8°20′W (a control section of the Mediterranean Water outflow). Bad weather conditions shortened the original program and the remaining ship time (25–28 November) was devoted to meddy search with satellite information, and to meddy survey with XBT and XCTD probes, RAFOS floats and deep-drogued surface drifters (Surdrifts). Search was guided by satellite measurement of sea surface height provided by the French Navy Oceanographic Center (SHOM/CMO). Figure 1 presents the mean sea level anomaly (SLA) on 22 November, based on the merger of Topex/Poseidon and ERS products. An anomaly of the sea surface with positive elevation extended from 35°30′N to 36°30′N and from 10°W to 11°30W with maximum elevation larger than 12 cm; therefore, hydrological search for meddies concentrated in this area (see the map of hydrological measurements on Figure 2).
 A preliminary transect was achieved to verify the presence of Mediterranean Water under the positive SLA area (stations DTX338 to 346, performed on 25 November). From the point of maximum temperature and salinity anomalies on this section, a long section was achieved, oriented WNW-ESE in accordance with the maxima and minima of SLA at 36°N–11°W and at 37°N–13°W. On this section, alternating XBT and XCTD casts were achieved on average every 7 NM, from 35°25′N, 9°55′W to 36°8′N, 11°10′W (stations DTX344 to DTX372 on 26 November). This section evidenced two meddies, hereafter called NW (northwest) and SE (southeast); two perpendicular sections with XBT and XCTD casts every 5 NM were then achieved across each meddy center (soundings DTX369 to DTX378, and DTX382 to DTX394, on 27 November). Within 3 days, 42 casts (19 XCTD and 23 XBT) were performed in and near the meddies.
2.2. Data Processing
 Hydrological profiles were all calibrated against in situ measurements (surface temperature and salinity from chemical analysis of water samples). Individual inspection and a median filter eliminated spurious points on the profiles, which then underwent a binomial Newtonian smoothing. The total number of points on the profile was finally reduced to 10 or 20% of the initial number under the constraint that the initial profile could be reconstructed from the final one with an error at most equal to the probe accuracy. With these expendable probes, the accuracy on temperature measurements is 0.018°C and that on salinity is 0.013 (the same accuracy is obtained for reconstructed salinity profiles). From a water mass diagram, salinity was reconstructed for XBTs using the method developed by Käse et al. . Hydrological data are presented here using the Ocean Data View software [Schlitzer, 2005]. The interpolating method is based on the DIVA algorithm with 6 km characteristic horizontal scale [Alvera-Azcarate et al., 2005]. This radius was found to avoid the bull's-eye artifact [see, e.g., Waller and Gotway, 2004]; horizontal maps of temperature and salinity were found mostly insensitive to this scale in the 6–10 km range.
 Three isobaric RAFOS floats were released, a shallower one, SV101 (at 850 dbar), in meddy NW and two deeper floats, SV106 and SV120 (at 1000 and 1300 dbar) in meddy SE (see Table 1). This choice of seeding was justified by the hydrological structure of the meddies. These floats were tracked with 3 local sources (IL1, a recent Portuguese source and M1, M2, two older German sources); ambiguities on the trajectories (in the case of small angles) were solved using more distant sources in the North Atlantic Ocean. A gross estimate of the positioning error, based on the distance between the last acoustic triangulation and the first surface location given by Argos is 5 km, but relative errors within a trajectory are likely to be much smaller.
Table 1. Characteristics, Position, and Date of Release and of Mission Termination for RAFOS Floats and Surdrift Buoys During the SEMANE 2000 Experiment
Date of Release
Date of Effective Termination
26 Nov 2000
17 Jul 2001
28 Nov 2000
17 Jul 2001
26 Nov 2000
17 Jul 2001
26 Nov 2000
7 May 2001
26 Nov 2000
4 Jun 2001
 Two surface drifters (Surdrift buoys), drogued at 900 and 1100 m, were released in the meddies, the shallower one in meddy NW, the deeper one in meddy SE (see dates and positions of release in Table 1). They were programmed for a 180 day mission and tracked via Argos. Only the one released in meddy NW became trapped. Their trajectories were resampled every 8h. A test on relative acceleration was performed to detect a possible loss of the drogue; in that case, their mission was terminated (further data were discarded).
3. Hydrographic and Dynamical Structure of the Two Meddies
3.1. Long Transect
Figure 3 shows the vertical sections of temperature, salinity, density anomaly (referenced at 1000 m) and planetary potential vorticity anomaly (the difference of (f/ρ) dρ/dz in and out of the meddies) along the main hydrological transect. Two meddies are distinctly visible, one with two tall and narrow cores to the northwest (“meddy NW”) and a wider one to the southeast (“meddy SE”). Meddy NW has temperature above 13°C at 700 m depth, while it is only above 12°C in its lower core (at 1300 m depth). The salinity structure evidences a maximum above 36.4 at 1300 m, and a secondary peak of 36.4 near 750 m. This double structure and the values of these maxima are characteristic of the Mediterranean Water undercurrents near Portimão Canyon [Chérubin et al., 2000]. The corresponding temperature and salinity anomalies, with respect to North Atlantic waters out of the Mediterranean Water tongue, are 2°C and 0.8 at 750 m, and 2.5°C and 0.6 at 1400 m depth.
 On the contrary, meddy SE has more diffuse temperature and salinity cores, with a temperature above 12°C between 700 and 1200 m and a salinity maximum above 36.4 between 950 and 1400 m. The temperature and salinity anomalies with respect to North Atlantic waters are 2°C and 0.5 at 1000 m, and 2°C and 0.4 at 1500 m depth. The thermohaline characteristics of meddy SE are found in the Mediterranean Water Undercurrents near Cape Saint Vincent [Daniault et al., 1994; Chérubin et al., 2000], but they can also result from intense spreading and mixing. A bridge in temperature connects the two meddies at 750 m depth and a bridge in salinity is present near 1300 m depth. This indicates that the two meddies interact. Meddy SE also exhibits a sharp thermal gradient on its southwestern flank indicative of a strong shearing activity.
 Based on isotherm 11.5°C at 1250 m depth, meddy NW has a diameter of about 45 km and meddy SE of about 60 km. These small values, compared to those of other meddies reported in the literature, are believed to result from erosive processes (meddy-eddy or meddy-undercurrent interaction, in particular). Isopycnal deviations below meddies are slightly larger than those above meddies. But the ambient stratification, and in particular the pycnocline near 600–700 m depth, renders isopycnals more densely packed vertically above than below these meddies (see again Figure 3). Consequently, planetary potential vorticity anomalies are larger above the eddies than below. Potential vorticity anomalies in the meddy cores are not as large as in other meddies previously observed [see, e.g., Carton et al., 2002]. Note that we have not included relative vorticity in potential vorticity anomalies, because its relative error is too important due to the small number of measurement points. Nevertheless, since the Rossby number of meddy NW is large (about 0.75, see subsection 3.4), we can expect the total potential vorticity anomaly in the meddy core, to be nearly twice as large as shown here. The vertically tripolar structure in potential vorticity anomaly, already observed before, is present for both eddies.
3.2. Perpendicular Cross Sections
 Perpendicular sections across each meddy are shown in Figures 4 and 5 to provide a more detailed view of their internal structure. Figure 4 shows meddy NW. Its salinity maximum is 36.78. Its temperature peak is 13.45°C at 750 m depth, and temperature at 1300 m depth reaches 12.5°C. The compact core of the meddy is well defined by isotherm 12°C and by isohaline 36.4; isotherms 11.5°C and isohaline 36.2 evidence substantial extrusion of warm and salty water near 750 and 1300 m depths, in particular northward. This suggests that this meddy may have interacted with another structure lying farther north or with the Gorringe Bank. We also note that the vertical deviations of isopycnals have their maxima near 600 m and 1600 m depths.
Figure 5 presents meddy SE. Again, stronger homogeneity in temperature between 600 and 1300 m depths, and of salinity between 700 and 1500 m depths is evidenced. Extrusions of warm and salty water are seen near 1200 m depth on both ends of the section (northward and southward). A cold filament (10°C) borders the southern flank of meddy SE between 800 and 1200 m depths; similar waters were observed at 1000 m depth north of meddy NW; they may result from the tearing of colder water patches around the meddies under the action of velocity shear.
3.3. Horizontal Maps of Temperature and Salinity
Figure 6 presents the horizontal maps of salinity and temperature at 700 m and at 1300 m depths, computed from all measurements. These maps illustrate the continuity of thermohaline anomalies between the two meddies, even if T and S maxima are well defined at each meddy center. These maps also evidence that the thermohaline gradients are stronger along the transverse (NNE-SSW) axis of the meddies and that warm and salty filaments seem to extend west of meddy NW and east of meddy SE. These plots are similar to vorticity maps showing the interaction of two like-signed vortices, where filaments trail and detach behind the corotating vortices [see, e.g., Melander et al., 1987]. The salinity structure of meddy NW clearly evidences a vertical tilt of the core between 700 and 1300 m which has often been noticed during like-signed vortex interaction in stratified flows [e.g., Dritschel, 2002].
 The salinity maps at 700 and 1300 m depths, and the temperature map at 700 m depth, indicate that warm and salty water extends northward from meddy NW (a possible interaction with another eddy or with topography, as mentioned above). Meddy SE is surrounded by fresher and colder waters at 1300 m depth, which amplify the thermohaline contrasts (this also occurs, but less distinctly at 700 m depth). Also, the temperature and salinity contours indicate that the water masses are well defined in the vortex cores.
3.4. Geostrophic Velocities
 Geostrophic velocity of the meddies is computed via thermal wind balance with reference to 2000 m depth (Figures 7 and 8). Vertical cross sections of velocity confirm the conclusions drawn from the temperature and salinity sections: meddy NW is more intense than meddy SE.
 In meddy NW, the velocity peak lies at about 8 km from the meddy axis and the maximal velocity reaches 50 cm/s (see Figure 7). Vertically, the positive and negative peaks do not lie exactly at the same depth (1000 m and 1150 m); this may be related to the vertical tilt of the meddy, observed in the salinity structure. The asymmetry between the negative and positive lobes can also be related to larger-scale advecting velocities (regional currents or influence of other eddies). For the velocity in meddy NW, the e folding scale is about 8 km horizontally and 300 m vertically.
 Based on these values, meddy NW has a Rossby number Ro = U/f0L = 0.75, a Burger number Bu = NH/f0L = 0.89 and hence Ro/Bu = 0.84. Theory [see, e.g., Carton, 2001] states that, via the hydrostatic and geostrophic balances, the ratio Ro/Bu is a measure of the relative vertical deviation of isopycnals ΔH/H, where H is the fluid thickness between two isopycnals out of the meddies and H + ΔH is the vertical distance between the same two isopycnals in the meddy core. Here ΔH/H is computed by measuring the vertical distance between isopycnals σ1 = 31.85 and σ1 = 32.25 which have the maximal vertical deviations; we obtain ΔH/H = 0.94. This fits the theoretical relation.
 Meddy SE has a more complex velocity pattern (see Figure 8). As anticipated from the hydrological structure, this pattern is wider than that of meddy NW (the dynamical radius is close to 18 km instead of 8, and the maximal velocity closer to 20 cm/s instead of 50). On the southwestern flank of meddy SE, a seemingly cyclonic velocity pattern with two narrow poles reaching 25 cm/s appears where the cold patch was noticed in the hydrological section. This cyclonic pattern is also related to the squeezing of isopycnals at these location and depth.
 Note that the geopotential anomalies at the surface have been calculated from hydrological measurements yielding a maximum of 9 cm. The observed peak of SLA (see Figure 1) is 13.5 cm. We attribute the difference to noise in the near surface hydrological structure and/or to an underestimation of the deep velocities.
4. Meddy Evolution and Trajectories From Observations
4.1. Analysis of RAFOS Float and Surdrift Buoy Trajectories
Figure 9 presents the RAFOS float trajectories colored by temperature over 7.5 months (for floats 101 and 106) and over 4 months (for float 120). Mission termination dates are given in Table 1.
 First, all floats initially followed similar paths: a slow cyclonic motion to the Northeast (first to the East, then to the North). Then a faster drift to the Southeast is visible for floats 101 and 106. This trajectory is called the Y-shaped trajectory, below. Clearly, the correlation between all float trajectories shortly after their release indicates that they have all become trapped in the same meddy. The initial drift of float 106 from its seeding position is westward-northwestward, i.e., toward the center of meddy NW. This seems to indicate that the meddy trapping floats afterward, was meddy NW, or the end product of a rapid interaction between meddy NW and meddy SE. This will be shown below via numerical modeling.
 Secondly, the temperature recorded by the floats during their progression is displayed on Figure 9. RAFOS 106 (at 1000 dbar) which circled closer to meddy NW center, experienced weak variations in temperature: a progressive decrease in the northern branch of the Y-shaped trajectory, but an increase on the northeastern side of its trajectory, at late stages. This increase suggests the existence of a warm eddy east of the meddy at that time (a Mediterranean water cyclone, for instance). On the contrary, RAFOS 101 trajectory (at 850 dbar) saw a rapid increase in its radius, correlated with a decrease in temperature. The radius of RAFOS 120 trajectory was 25 km since its first loop, which is likely to reflect that it circled meddy NW in its periphery. RAFOS 120 did not experience much change in temperature, except after expulsion from meddy NW.
 The trajectories of Surdrifts 25466 and 25467, drogued at 1100 and at 900 m, are presented on Figure 10. Buoy 25466, released in meddy SE, circled meddy NW only once before complete ejection. This shows that it drifted to the far periphery of meddy NW, where the influence of surrounding flow can dominate the internal rotation of the eddy. This also supports our hypothesis of a fast initial interaction between the two meddies. On the contrary, buoy 25467 achieved more than ten loops (with increasing radii) before ejection. It clearly followed meddy NW. After these ten loops, buoy 25467 was ejected but evidenced a southward drift, correlated with the meddy, at late times. The trajectory is not colored any more, when the buoy is not clearly looping around a meddy.
 In summary, the global drift of the meddy trapping floats and buoys was cyclonic in both stages, first toward the east-northeast, then to the south-southeast. The change of heading, the sudden acceleration, and the variations in recorded temperature are likely to reflect successive interactions of this meddy with neighboring eddies.
 The RAFOS float trajectories were then analyzed to determine the azimuthal velocity and rotation rate, the rotation period and the distance of the floats to the center of the meddy (see Paillet et al.  for details on the computation of these velocities and distances). Figures 11–13 indicate that after 40–50 days, floats 101, 106 and 120 stabilized respectively at 22, 12 and 25 km distance from the center (though with variations). The rotation velocity (or rotation rate) in this meddy varies from 1.8 rad/day at 8 km radius to 1.0 rad/day at 25 km from the center, on Figures 11–13. Floats 101 and 120 indicate azimuthal velocities on order of −0.3 m/s at 25 km radius. Maximum azimuthal velocities in the meddy, as measured by floats, was −0.45 m/s, close to the Eulerian value of −0.5 m/s, but at larger distances from the center (which may be the result of eddy interactions).
 The analysis of the rotation of Surdrift buoy 25467 indicates weaker velocities, a priori due to the drag of the currents on the cable (Figure 14). For RAFOS floats, the rotation period was fairly uniform, around 6 to 8 days, except near days 130–150. At that time, a strong perturbation occurred in the rotation period and radii of all floats: their rotation slowed down and they drifted away from the meddy center. At day 140, the trajectory accelerated southward. Again, these observations support the hypothesis that the meddy containing the floats was successively influenced by two eddies.
4.2. Joint Analysis of Float and Buoy Positions and of Altimetric SLA
 To explain further the meddy motion observed with floats, and in view of Figure 1, which evidences cyclonic structures north, northwest, west and south of the meddies initially, we superimpose the meddy positions, as determined by the float trajectories, on maps of sea level anomaly (SLA) obtained from satellite altimetry, from the beginning of the experiment (22 November 2000) until the end of June 2001 (i.e., for the duration of the RAFOS float mission). Figures 15–17 present the time evolution of SLA correlated with the RAFOS and Surdrift trajectory centroids over 7.5 months.
 Initially, a single positive SLA is observed above the two meddies. From 29 November to 6 December, all float trajectories collapse on the northernmost location (figure not shown) indicating that all floats and buoys are then trapped in a single anticyclone, a priori meddy NW, as stated earlier. From 29 November to 20 December, the positive SLA, initially centered at 35°50′N–10°50′W becomes considerably elongated in a NW-SE orientation. This orientation is compatible with a shear created by two cyclones north and south of the meddies (the observed negative SLAs). In January, the positive SLA follows a clockwise trajectory.
 But the floats and buoys do not follow this positive SLA: they migrate eastward and then northward, for a month (7 January to 10 February). This migration can be related to the negative SLA (a priori a cyclone) initially located north of the meddy. Such a meddy-cyclone interaction would advect the meddy eastward (or cyclonically, if the cyclone is intense). Only in early February do the floats recover a positive SLA that they will not leave until May. This temporary absence of positive SLA above the floats may be due to a “shielding” of the meddy surface signature by the surface signature of a strong neighboring cyclone.
 From January to February, the original positive SLA has drifted northwestward, and by mid-February, it lies between 11 and 12°W. Most likely, this positive SLA reflects the trajectory of meddy SE. From 14 March, till end of June, the positive SLA above the floats follows a cyclonic path, a priori related to a negative SLA (another cyclone) near 35°30′N, 9°W.
 In summary, the motion of the meddy containing the floats (a priori meddy NW) does not follow the drift of its initial surface signature. This signature drifts northwestward, most likely with meddy SE, advected clockwise by meddy NW. On the contrary, meddy NW follows a Y-shaped trajectory, most likely related to interactions with neighboring cyclones. Such cyclones are supplied from the eastern boundary of the domain, where they can be generated by the Mediterranean Water Undercurrents [Serra and Ambar, 2002; Serra et al., 2005]. Furthermore, anticlockwise motion is not uncommon for meddies generated south/southwest of Cape Saint Vincent, as Carton et al.  and Ambar et al.  already noticed.
 Some theoretical explanation for these observations can be found in the work by Nof , which shows that nonlinear cyclones can break into smaller eddies and thus provide the energy for the merger of anticyclones nearby. In our observations, there is little doubt indeed (as numerical modeling will show) that neighboring cyclones must have played an important role in the evolution of meddy NW. Baey and Carton  have studied the stability of nonlinear eddies of both polarities (cyclones and anticyclones) with similar velocity profiles; their analytical and numerical results indicate that anticyclones are more stable than cyclones, for Rossby numbers larger than one half. Carton and Bertrand  have also shown that anticyclones merge more easily than cyclones when the Rossby number is also about one half.
 Now, to explain why the two meddies have not completely merged, another study must be recalled [Perrot and Carton, 2009]: two vortices embedded in opposite shear or strain merge less easily than in the absence of externally imposed deformation. This positive shear is created here by the neighboring cyclones. To support this interpretation based on SLA maps, a simple numerical model is now used.
5. Modeling the Meddy Evolution
5.1. Model Equations and Configuration
 We use a three-layer quasi-geostrophic model for eddy evolution in a large, open ocean domain. Though the Rossby number of our meddies is large (and comparable to their Burger number), quasi-geostrophy is expected to provide a correct first-order description of their dynamics, as many studies showed [e.g., Käse and Zenk, 1986; Carton et al., 2002]. As recalled above, nonlinear eddies are different from their quasi-geostrophic counterparts in that anticyclones are more stable and merge more easily that cyclones. But the cyclone observed from April to June 2001 must have been stable enough to induce the second branch of the Y-shaped trajectory.
 The main objective of this modeling section is to show that two meddies cannot have drifted as SLA maps and RAFOS float trajectories indicate, without the influence of neighboring cyclones. The dynamical equations of the quasi-geostrophic model are [see Pedlosky, 1979]
where ψj is layer-wise stream function and the two-dimensional advecting velocity is j = ∢ ψj. The layer-wise potential vorticity is
with flat ocean surface and bottom.
 Note that if the shallow water model was used to incorporate all nonlinear effects, for a given velocity distribution, due to cyclogeostrophic balance, the pressure gradient would be modified by a factor 1 + Ro where Ro is the Rossby number. The shallow water potential vorticity anomaly would also be qualitatively modified by this factor compared to the quasi-geostrophic potential vorticity. The numerical scheme and model parameters have already been described at length by Carton et al.  and only the main features are recalled here: the first and second internal radii of deformation are 30 and 15 km, the layer thicknesses are from top to bottom, 600, 900 and 2500 m. Neither topography, nor solid boundaries are included in the model since the area under consideration is an abyssal plain. The dimensionless hyperviscosity coefficient is ν = 10−7. The model domain is square with 754 km side length. The horizontal resolution is 256 × 256. Simulations performed at the higher resolution of 512 × 512 did not evidence substantial differences with the lower resolution. The model variables are scaled with L = 30 km and T = 0.33 days (model velocities thus read in m/s).
5.2. Evolution of the Meddies and Cyclones
 The first experiment (E1) is initialized with only the two meddies: in the middle layer, meddy NW has a velocity maximum of −0.5 m/s at 10 km radius while meddy SE has a velocity maximum of −0.2 m/s at 20 km radius (see Table 2 for a list of parameters for each experiment). Their radial profile of isopycnal elevation is Gaussian and they are separated by a 50 km distance.
Table 2. Characteristics and Initial Positions of the Vortices in Each Numerical Experimenta
The position is given with respect to the domain center.
E1 (meddy NW)
E1 (meddy SE)
E2 (meddy NW)
E2 (meddy SE)
E2 (cyclone N)
E3 (meddy NW
E3 (meddy SE)
E3 (cyclone N)
E3 (cyclone S)
E4 (meddy NW)
E4 (meddy SE)
E4 (cyclone N)
E4 (cyclone S)
E4 (cyclone W)
Figure 18 presents the time evolution of potential vorticity in the intermediate layer, every 18 days. The stronger meddy advects and deforms the weaker one. The periphery of the weaker meddy is absorbed by the stronger meddy. This bears strong similarity with the initial observations where the floats seeded in the weaker meddy rapidly drifted toward the stronger one. This initial stage of meddy evolution is comparable to observations, but their long-term trajectories are not. The model shows a southward drift of meddy SE and stationarity of meddy NW, whereas in observations, meddy NW heads eastward and then northward. Therefore, the sole meddy-meddy interaction is not sufficient to explain the observed meddy trajectories.
 This result and the observation of SLA maps lead us to add a cyclone north of the two meddies (experiment E2). The velocity structure of this cyclone is derived from an in situ observation during the SEMANE 1999 experiment: a cyclone was observed with a velocity maximum of 0.35 m/s at 20 km radius, near 1000 m depth. We assume that the cyclone located north of the two meddies is comparable to this observation (we choose here a velocity maximum of 0.4 m/s for this cyclone), and we initialize a second experiment with the two meddies and this cyclone, located 90 km north of the meddies. A test was made on the sensitivity of model results to the velocity structure of the cyclone. Qualitatively comparable results were obtained for moderately different velocity distributions.
Figure 19 shows the time series of potential vorticity for E2. One can clearly see that the evolution during the first two weeks is comparable to that without the cyclone. But after three weeks, the influence of the northern cyclone is important. Meddy NW now drifts eastward much faster than in the absence of the cyclone. At the end of the simulation, it also veers northeastward in agreement with the observations. Meddy SE also drifts westward more markedly than in experiment E1, but unfortunately too far southward.
 Considering again the SLA maps, we add a second cyclone, about 150 km south of the meddies. Several tests showed that this cyclone must be weaker than the northern one. Figure 20 shows that this cyclone pairs with meddy SE after the meddy-meddy interaction. This southern dipole heads west and its final position is in agreement with the western SLA anomaly mid-April 2001.
 Finally, we add a third cyclone 180 km east of these eddies (Figure 21). This cyclone is introduced to model the positive sea surface anomaly seen in the SLA maps from April to June 2001, but it is initialized southeast of its observed position (to account for its subsequent northwestward drift due to beta effect). Clearly, meddy NW exchanges partner as it reaches its northeasternmost position, and then couples with the eastern cyclone. Their trajectory is southward (cyclonic), and follows the second part of the observed Y-shaped trajectory of the RAFOS floats.
6. Discussion and Conclusions
 The analysis of the SEMANE 2000 data set, and a simple series of quasi-geostrophic model simulations, lead to the following conclusions (and interpretation of the observations): Two meddies came in close contact, southeast of Cape Saint Vincent, end of November 2000. Considering the SLA maps, it is likely that they have drifted to this observed position, accompanied by cyclones. The two meddies were asymmetric, the northwestern one being narrower, more marked in temperature and salinity, than the southeastern one. Deep-drogued Surdrift buoy, and RAFOS float trajectories indicate that the interaction of these two meddies was brief and intense, since all floats were drawn toward the larger meddy. Its subsequent trajectory was Y shaped. The simple interaction of these two meddies is not sufficient to explain the surface signatures of eddies on the successive SLA maps. Indeed, the corresponding model simulations show that such an interaction is an incomplete merging process (the reason why, being discussed below), whereby a stronger meddy emerges without major displacement and a smaller anticyclonic fragment is released and drifts southwestward. On the contrary, the SLA maps indicate that the stronger meddy underwent a strong eastward displacement and that the smaller meddy finally drifted westward. Several model simulations indicate that at least two cyclones, one initially north and one initially south of the meddies, must have played a critical role in their motion.
 Furthermore, a third cyclone, entering the observation domain from the East, must have paired with meddy NW at the end of the first part of its Y-shaped trajectory. This pairing leads to the change in direction, evidenced by the second part of this trajectory. Thus, the cyclones have played an essential dynamical role in the evolution of these meddies southwest of Cape Saint Vincent. This role is not negligible since other MW dipoles have been observed south of Portimão Canyon, with a clear origin near this canyon [Carton et al., 2002]. RAFOS float trajectories of the MEDTOP experiment [Ambar et al., 2008, Figure 9] also show these cyclonic motions. The interpretation by Ambar et al. , that anticyclones follow a path around cyclones in the Gulf of Cadiz, is therefore strengthened by our observations. Thus, this recirculation in the Gulf diverts part of the southwestward salt flux of MW. This eddy circulation may also be related to the cyclonic surface circulation found by Peliz et al. .
 The second important point here is that the meddy-meddy interaction was brief and incomplete. Several dynamical elements can explain this evolution. First, the two meddies were asymmetric. Many studies have shown that for two unequal vortices, vortex merger is not a completely efficient mechanism and two vortices can remain as end product of the process (a so-called “partial merger” process [Melander et al., 1987; Dritschel and Waugh, 1992; Yasuda and Flierl, 1997]). Secondly, the meddy interaction took place in a stratified ocean. Studies have shown that, in a stratified rotating fluid, the ability of two vortices to merge strongly depends on their potential vorticity distribution [Verron and Valcke, 1994; Valcke and Verron, 1997]. Indeed, both vertical and horizontal shielding effects can be due to the presence of opposite sign vorticity above, below or around the vortex core; merger is then inhibited. Thirdly, the beta effect is known to favor vortex drift and dispersion against their merger [Carton and Bertrand, 1993]. Finally, the neighboring cyclones may have hindered the meddy interaction by inducing zonal advection, or a velocity shear, on these meddies. This explains why observations of meddy interactions are rare in the ocean.
 The third important point is that SLA maps did not show a signature of meddy NW continuous in time, and that the westward drift of the positive SLA corresponded to meddy SE, and did not follow the RAFOS float trajectories. Our interpretation is that the cyclones, which are intensified at shallower levels than meddies, may have canceled the surface signature of meddy NW at times. A consequence of this is that SLA maps do not accurately reflect meddies in this generation region.
 Obviously, our data set is incomplete for a detailed study of meddy-cyclone interactions, and we had to hypothesize how strong cyclones were for model simulations. The model simulations were relatively robust to variations in cyclone strengths, nevertheless, and qualitatively similar evolutions were obtained for cyclones with slightly different strengths or initial positions. To obtain more precise simulations, both the use of a primitive equation model and the availability of hydrological data in a larger domain, to calculate the cyclone strengths and the large-scale currents, would be necessary. A synoptic description of all eddies in an ocean domain escapes hydrological measurements by a single ship. Gliders, or seismic reflection measurements [see, e.g., Gonella and Michon, 1988] may also help locate and characterize deep eddies, with great precision, in an oceanic domain several hundreds kilometer wide. This will be necessary to provide an accurate correlation of deep eddy dynamics with the time evolution of their surface signature, measured by satellite altimetry, and also to characterize regional dynamics.
 The authors express gratitude to two anonymous referees whose comments clearly improved the paper (and in particular for pointing the 1991 Nof paper to us). SHOM/CMO participation in cruise preparation and initial data processing was crucial for the program. The Captain and crew of R/V D'Entrecasteaux are thanked for their help in data collection. We express deep gratitude to Alain Faisant for his thorough processing of the float data. This work is a contribution to the SEMANE program.