During a recent research cruise to investigate the nature and continuity of the Mozambique Current, we observed that the flow in the Mozambique Channel is dominated by a train of large anti-cyclonic eddies (diameters >300 km) that reach to the channel bottom and propagate southward. At a frequency of 4 per year they cause a net poleward transport of about 15 Sv (1 Sv = 106 m3/s). In the deep sea, a Mozambique Undercurrent flows equatorward along the continental slope. Using a lowered acoustic Doppler current profiler maximum observed velocities are about 0.2 m/s around 2400 m with another current core around 1000 m. It carries about 5 Sv of intermediate (AAIW) and deep waters (NADW) of Atlantic origin into the Channel. Subsequently, the equatorward flowing AAIW is largely entrained by the eddies and, while mixing with intermediate water from the North Indian Ocean in the eddy core, returned to the Agulhas Retroflection region.
 As part of the warm upper limb of the global thermohaline ocean circulation the Pacific-Indian Ocean flow through the Indonesian Passages propagates westward in the South Equatorial Current [Gordon, 1986]. A discussion is ongoing about its fate. The transport in the SE Madagascar Current is too small to accommodate both the wind-driven transport of the subtropical gyre and the Pacific-Indian Ocean throughflow. Based on this, indirect estimates have been derived for the transport through the Mozambique Channel, ranging from 6 [Fu, 1986; Swallow et al., 1991] to 15 [Ganachaud et al., 2000] Sv. Direct observations have been lacking [Schott and McCreary Jr., 2001]. The flow structure of the Mozambique Current has been estimated from surface drift [Saetre, 1985; Lutjeharms et al., 2000], and sparse hydrographic observations [Saetre and Da Silva, 1984; Donguy and Piton, 1991]. The results seem to be conflicting, probably due to the highly varying flow in the Channel. This is supported by eddy-permitting model simulations [Biastoch and Krauss, 1999] that give a series of shallow mesoscale eddies moving poleward in the upper layer. To investigate the nature of the Mozambique Current—coherent current, collection of eddies or both, the first ACSEX (Agulhas Current Sources Experiment) cruise was designed to intercept the purported trajectory of the Mozambique Current at a series of locations (Figure 1) and, to investigate circulation anomalies in the sector known for high mesoscale variability. The latter were identified by real-time altimetric data. Currents were observed over the full water column using a Lowered Acoustic Doppler Current Profiler (LADCP).
2. The Mozambique Channel Eddies
Figure 1 shows the observed near-surface currents in relation to concurrent altimetric sea surface height (ssh) anomalies at four cross-shore hydrographic sections. These transects were connected by XBT-lines to provide further evidence on the geometric characteristics of the flow structure. Wherever positive anomalies of ssh were intersected, their predictive skill for surface currents appeared very good. Ship drift and XBT-observations indicated that their negative counterparts, suggesting cyclonic motion, were non-existent or much weaker. This is probably due to the averaging process to define a mean ssh-field: a negative anomaly merely means the absence of an anticyclonic eddy. At the northern transect, an along-slope southward boundary current was observed (Figure 1) with a total transport of about 30 Sv: 15 Sv in the upper 450 m layer of Indian Central Water (T > 10°C, S > 34.9, sigma theta <26.8 kg/m3), 11 Sv in the Intermediate Water layer, and 4 Sv of Circumpolar Deep Water. Temperature and salinity characteristics of this boundary current (the light-green TS curve in Figure 4) are similar to those in the East African Coastal Current [Schott and McCreary Jr., 2001], pointing to a common origin in the northern limb of the East Madagascar Current. This section was too short to determine whether the current was part of an eddy, as the altimetry might suggest (Figure 1), or a western boundary current.
 The observations in the other transects confirm that anti-cyclonic eddies are prevalent in the Mozambique Channel. Moreover, drifters launched during the cruise described circular paths around the periphery of the eddies (Figure 2). They appear to be over 300 km wide and penetrate to the bottom (Figure 3), where swirl velocities are still around 0.1 m/s. They propagate southward [Schouten et al., 2001b] at an average speed of about 4.5 km/day. Consequently, a train of 3 anti-cyclonic eddies is mostly present in the Channel. This produces a rectified southward current at the Mozambique side of the Channel and northward towards the Madagascar side. The eddies are surface intensified but also have a significant barotropic component (Figure 3). Their water mass and velocity structure indicate trapping of water to below the intermediate water mass level. With a frequency of 4 per year [Schouten et al., 2001b] and water trapping down to 1500 m an eddy induced volume transport results of about 15 Sv.
3. The Mozambique Undercurrent
 Below the thermocline, our direct current observations have revealed a Mozambique undercurrent flowing along continental slope (Figure 3a). Inshore of the eddies it flows equatorward. It has two cores. One is at intermediate level, where speeds range between 0.1 and 0.2 m/s. It weakens northward (Figure 3b), while carrying Antarctic Intermediate Water (AAIW) along the continental slope. At 24°S it stands out by a marked salinity minimum (Figures 4 and 5). The other core is below 2000 m, where it has maximum equatorward speeds near 0.2 m/s (Figure 3a). This narrow jet carries North Atlantic Deep Water (NADW) northward, as can be deduced from its relatively high salinity and oxygen concentrations, together with low silica levels (not shown here). The deep undercurrent is still observed in the narrow section of the Channel around 2500 m (Figure 3b), but much weaker. During this cruise, the total equatorward transport in the Mozambique Undercurrent is, roughly estimated, 5 Sv at 24°S, of which some 2 Sv is in the NADW-core.
4. Pathways of Intermediate Water
 The Mozambique Channel eddies carry salty and warm tropical upper layer and thermocline water southward. In their cores, at intermediate level, they contain relatively salty, low oxygen (not shown here) water of Red Sea origin (RSW, Figure 4). Along isopycnal mixing is observed between the AAIW from the northward undercurrent and this RSW in the eddy cores (Figure 5). Interleaving of these two water masses has been observed at all hydrographic sections taken in the Mozambique Channel. Apparently, the Mozambique eddies are effective transporters of RSW and its associated salt towards the Agulhas Current. Moreover, they provide a short cut for the AAIW by entraining this northward flowing water mass and subsequently carrying it back southward.
5. Discussion and Conclusion
 The Mozambique Undercurrent is probably a continuation of the Agulhas Undercurrent [Beal and Bryden, 1997]. The rectified southward transport by the Mozambique Channel eddies is estimated at 15 Sv in the upper 1500 m. This agrees with recent inverse modelling estimates [Ganachaud et al., 2000]. The poleward transport of thermocline and surface waters by the Mozambique Channel eddies is large enough to provide for a significant part of the connection in the global thermohaline circulation, mentioned earlier. Its magnitude (of about 5 Sv) is about one half of that estimated for the Indonesian throughflow [Fieux et al., 1994] and the leakage of Agulhas water into the South Atlantic [De Ruijter et al., 1999]. The regular formation of Mozambique Channel eddies at the frequency of four per year is probably due to a train of baroclinic Rossby waves that travels westward around 12°S at a frequency of four per year [Schouten et al., 2001b]. A lagged correlation exists between the appearance of these waves in the eastern Indian Ocean and the formation of Mozambique Channel eddies little over a year later. However, mooring, altimetric and shipboard observations north and east of Madagascar [Schott et al., 1988; Quadfasel and Swallow, 1986] have shown dominant variability in this region at a period around 50 days, related to locally generated barotropic Rossby waves. Whether and how these different waves interact to generate four eddies per year in the Mozambique Channel is unknown. The eddies propagate south into the Agulhas Retroflection where they trigger the shedding of Agulhas Rings [Schouten et al., 2001b]. If the Red Sea Water which is mixed into the interior of the Indian Ocean is eventually exported at the western boundary [You, 1998; Beal et al., 2000], then the eddies from the Mozambique Channel provide a crucial link. The same holds for the connection they establish between the warm and salty tropical surface layer and thermocline waters of the Indian Ocean and the South Atlantic.
 From this ACSEX cruise we can conclude that a coherent and persistent Mozambique Current along the shelf edge off Mozambique does not exist. Instead, a significant link in the global ocean circulation system is brought about by the regular train of poleward propagating Mozambique Channel eddies. Intermediate and Deep Atlantic waters are carried equatorward by the Mozambique Undercurrent.
 This investigation was funded by: Netherlands Organisation for Scientific Research via its CLIVAR(NET) programme; Netherlands Institute of Sea Research; Graduate School COACH via the Institute for Marine and Atmospheric research of Utrecht University, South African Foundation for Research Development, South African Water Research Commission and the University of Cape Town. We thank Martin Visbeck for the latest LADCP-analysis software, Peter Niier for providing the drifters, and the crew of the RV Pelagia and all who sailed on the ACSEX I cruise.