Based on these regional NA data, global distributions of SST from CLIMAP (1981) were updated in the northern NA and the Nordic Seas to form LGM and MWE SST distributions. Present day sea-surface salinity was increased everywhere by 0.8 psu (to compromise between the upper limit estimate of 1 psu and smooth connection of the low and high latitudes in the NA; see (Seidov et al. 1996)) and the regional NA LGM SSS replaced the modified SSS to the north of 40°N to form the global LGM and MWE SSS distribution. Modern mean annual SST and SSS are specified from new ocean climatological data sets (Levitus and Boyer 1994; Levitus et al. 1994), except for the area south of Greenland where some additional cooling of up to 2 °C mimicking cold spells from Greenland was needed to obtain present-day annual NADW production.

3 Numerical Models

3.1 OGCM

Any OGCM contains a procedure to allow convective mixing that develops due to hydrostatic instability if denser water is formed (or advected) over lighter water. This is the key process for deep water production and transport. As done commonly in an OGCM, this instability is accommodated by mixing water vertically until it regains complete hydrostatic stability (Cox 1984). This vertical adjustment takes place as a step process with several successive mixings of adjacent layers until complete hydrostatic stability is restored. As the water convects, the particles associated with the mixing volumes exchange their positions vertically, and hence the convection facilitates deep ventilation.

3.2 Ocean Sedimentation Model

The 3-D component of the sedimentation model simulates the lateral transport and the entry of sediment particles at the sea-surface, including the entry of ice-rafted sedimentary material. The supply of large-grain sediment at the sea-surface can be increased or decreased depending on the amount of icebergs injected into the open ocean due to decay of major ice sheets. The icebergs distribution over the NA in the past can be inferred from analyses of proxy data from foraminifera [e.g. Maslin et al. 1995). The mass of sediment covering the sea floor depends only on the balance of sources and sinks, whereas the spatial variation of sedimentation rates depends on the circulation pattern and particle grain size. The 2-D model is initialized at every time step by the exchange of sediment between the ocean body and the ocean floor. The sediment in the bottom layer is transported by a corrected benthic flow which is largely a projection of the OGCM velocity field onto the smoothed bottom layer (1 cm thick). Additionally, the near-bottom velocities are reduced to take bottom friction into account (Miller et al. 1977; Zanke 1978; Sündermann 1983).

The impact of convection was not included in the previous version of the sedimentation model (Haupt et al. 1994). In the updated version the convection is incorporated in the form of convection depths from the OGCM (Seidov and Haupt 1997). Vertical mixing, similar to that employed in the OGCM, is applied to concentration. The sediment transport model ?knows? when to mix water because we have encoded the convection depths in the velocity field as an additional parameter. When used in the sedimentation model, this information is decoded to enable vertical mixing in the grid points where convection occurs (see below in the discussion of convection patterns).

3.3 Particle-Tracing Model

As we have emphasized above, ventilating convection induced by hydrostatic instability is included in all three components of our simulations. The semi-Lagrangian trajectory-tracing model of Haupt et al. (1994) was upgraded in Seidov and Haupt (1997) to facilitate vertical ventilation in convective "chimneys" [Send and Marshall (1995) show that ventilating convection occurs as water mixing in water columns or "chimneys"]. Here the velocity field from the OGCM is supplemented by the convection depths showing where and to what depth the vertically mixing volume should be propelled in the turbulent chimney. A detailed discussion of the parameterization of chimney mixing in the particle-tracing model is given in Seidov and Haupt (1997).

More details of the sediment transport model and the particle-tracing model can be also found in Haupt et al. (1997; see this Volume).

4 The Setup of Numerical Experiments

The regional NA study includes modeling of sediment transport, that is, these simulations comprise a threefold approach. The global ocean circulation modeling at this stage employs only the OGCM and trajectory tracing models, and therefore forms a twofold approach to water transport and ventilation problem.

For the regional NA studies we used a 2°x 2° grid with 12 vertical levels. In view of uncertainties of the past sea-surface conditions, and in order to facilitate multiple runs extended over 10 years with different parameter settings, were employed a coarse resolution of 6°x 4° in longitude and latitude respectively with 12 vertical levels. Although all major currents, except for the largely wind-driven Antarctic Circumpolar Current (ACC), are two to three times weaker than the observed ones, most of these currents are still clearly seen on the vector maps. At the same time, the meridional thermohaline overturning, which largely depends on the deep convection and isopycnal outcrop, is modeled far better than the horizontal flows. For the present-day overturning we have obtained a value that agrees well with experiments of much finer resolution. The overturning in the NA, which comprises the NADW production, amounts to 23 Sv (1 Sv=10⁶ m³ s^-1), for comparison Antarctic Bottom Water (AABW) inflow into the Atlantic Ocean is about 9 Sv. Since we are mainly interested in the major changes in conveyor operation and rely only on the relative changes of the conveyor intensity, the coarse resolution is legitimate for global paleocirculation studies such as that presented here. We therefore consider it to be sufficient for the semi-Lagrangian simulations which are the core of our investigation.

In contrast to sensitivity studies which operate with so-called mixed boundary conditions (e.g., Manabe and Stouffer 1994; Rahmstorf 1995), we restore the upper-layer thermohaline fields to the specified sea-surface temperature (SST) and sea-surface salinity (SSS). The sea-surface boundary conditions are described in the previous section. In our study, the important reason for the chosen approach is the local nature of the freshwater driving. Indeed, the LGM sea-surface conditions were disturbed by the meltwater invasion only in the small area in the northern NA and Norwegian-Greenland Seas where reliable proxy data are available (see above). This restoring technique implies that the freshwater fluxes were those that maintained the reconstructed SSS in these areas. In fact, we merely diagnose the circulation regimes which would satisfy the "observed" sea-surface conditions.

5 North Atlantic Sediment Transport and Water Motion

5.1 OGCM Results

Note that although glacial NADW production is 8 Sv, much lower than today's value of 13 Sv, the intensity of the glacial conveyor is comparable to that of modern times. If we quantify the southward transport at 30°N, the Upper NADW outflow, or forward conveyor branch, does indeed rise to only 8 Sv (also see the inventory of the glacial water masses given by, e.g., Oppo and Lehman 1993, and Oppo et al. 1995). However, if we take into account the deep reverse branch comprising AABW and its mixture with lower NADW still produced in the central northern Atlantic (see below in the discussion of convection patterns), the total transport amounts to 12 Sv, which gives almost the same intensity of the deep water outflow as the present-day North Atlantic combined deep outflow. The intensity of this NADW-depleted conveyor, similar to present-day conveyors' intensity, seems to fit well the recent finding based on proxy data analysis (Yu et al. 1996). However, the modern overturning strength is too low. A slight cooling of the sea-surface to the southeast of Greenland and in the Nordic Seas may easily double the NADW production (see Section 5.2). This is because the sea-surface density taken as averages of summer and winter values is somewhat biased toward summer values, and therefore the convection is not deep enough to form the vigorous observed present-day overturning. Yet we may consider the numbers as a first approximation, keeping in mind that the glacial conveyor was probably relatively even weaker than our regional modeling indicates. Hence, we believe that the glacial conveyor was probably still noticeably weaker than the modern one, in contrast to what is suggested by Yu et al. (1996). The global circulation modeling with sea-surface cooled down by 2 °C to the east of Greenland between 60° and 65°N gave 22 Sv NADW production (see below), which is probably too high; if one mediates the regional and global results there would be about 18 Sv, a number most appropriate for a coarse resolution study. This would give an LGM forward conveyor twice as weaker but still comparable outflow which would fit the idea of still intensive glacial thermohaline circulation (Yu et al. 1996; Webb et al. 1997).

As we have already noted, the 3-D conveyor is essentially more complex than its 2-D image given by the total meridional overturning. The velocity fields are discussed in great detail in Seidov et al. (1996) and Seidov and Haupt (1997). Here we briefly overview how the glacial and MWE currents differ from the modern ones. The present-day, LGM, and MWE simulated velocity fields may be inspected in Seidov et al. (1996). The most noticeable feature is the deviation of the paleo-North Atlantic Drift from its modern northeastern path. Strong zonality of the subpolar front indicates a reduced supply of water that can be downwelled as NADW in the northern North Atlantic and the NGS. Moreover, the route of the return southward flow in the deep ocean changed radically, a robust feature emerging in all our glacial experiments regardless of complexity. In the eastern part of the basin, the incursion of AABW dominates the near-bed transport up to the Faeroe-Shetland Ridge, in agreement with the water mass contouring by Sarnthein et al. (1994). The most striking feature of the MWE currents in addition to the changes found at LGM is the reversal of the Norwegian Current and the inflow-outflow regime in the Norwegian-Greenland Seas (see Seidov et al. 1996 for details). Hence the redeposited sediment in the eastern mid to high latitudes might be of different origin. Today, sediment drifts there transport material largely from northeast to west and southwest. In contrast, during the LGM some sediment drifts could change direction of transport to redeposit the grains from south to north and northwest.

Today's southward deep return current is a deep ocean western boundary flow forming a strong countercurrent under the Gulf Stream. This western boundary current is well recognized as the most prominent feature of the thermohaline circulation (Stommel and Arons 1960). Though a noticeably weaker western boundary current still existed at the LGM and MWE, the descending branch returned to the southwestern basin largely as a broad zonal westward flow in the mid latitudes. The glacial countercurrent under the paleo-Gulf Stream was weaker, deeper, and farther eastward. At the LGM a southward deep ocean flow originated near the Rockall Plateau at a depth of about 2 km, occurring in the eastern part of the basin, rather than in the western part as today. This particular feature of the computer model in the eastern North Atlantic is again in good agreement with the contouring by Sarnthein et al. (1994). Using the trajectory-tracing model below, we demonstrate that the simulated glacial deep water indeed moved along the eastern flank of the Mid-Atlantic Ridge. Therefore as water in the western deep Atlantic contained more AABW at the LGM, the deep and abyssal water during the glacial time was older than at present.

To understand the mechanism of change of deep water formation, one should consider changes in the convection regime. The modern deep convection sites are found mainly in the NGS, east and south of Iceland, and in the Labrador Sea. The present-day convection pattern in our calculations is consistent with the patterns given by Maier-Reimer et al. (1993), Toggweiler et al. (1989), and Rahmstorf (1995). Shallow convection starts in the Gulf Stream area and marks a further progression of subtropical water to high latitudes carried by the North Atlantic Current becoming deeper as it progresses northeastward and westward around the Greenland coast.

The major feature of glacial convection, as compared to its modern state, was the far weaker glacial convection in the NGS and the southward shift of the main convection sites. The maximum ventilation at the LGM occurred in the mid-North Atlantic between 50° and 60°N. This result is well supported by proxy data (Duplessy et al. 1988; Sarnthein et al., 1995). A southward shift of both the convection sites and the polar front led to a decrease in the northward transport of warm and salty water. Hence a positive feedback worked to establish the glacial mode of circulation. An even farther southward shift of the convection was limited by Ekman pumping in the subtropics which protected the major anticyclonic gyre from shrinking even more, i.e., providing a negative feedback to balance the positive one. The MWE convection sites were shifted further southward and the convection was very shallow, penetrating no deeper than 600 m in the central North Atlantic. There was no convection to the north of 50°N and the Nordic Seas were completely convection-free during that time slice (Seidov et al. 1996).

5.2 Sediment Accumulation Rates and Sediment Transport

As the glacial North Atlantic Current diverged from today's northeastern path, the accumulation rate in the Iceland and Irminger Basins dropped, and most of the sediment mass was spread over the abyssal valley in the Canary Basin (Figure 2b). Note that although the sedimentation rate is far lower there than in the Iceland Basin, the sediment mass is roughly the same because bottom area in the latitude-longitude grid cells increases rapidly to the south. A relatively high glacial accumulation rate is found in the Newfoundland Basin. However, we do not see continuous southward sediment transport along the American coast, a signature of all our experiments based on the modern ocean surface climatology. Instead, a noticeable southward sediment transport is found at the eastern flank of the Mid-Atlantic Ridge.

Though an increased northward incursion of AABW at the LGM in the eastern North Atlantic is consistent with the contouring by Sarnthein et al. (1994), the conclusions about an increased reverse abyssal gyre cannot be attributed likewise to the western part of the basin. Indeed, the southward transport, though noticeably curtailed, still existed in the western deep ocean. Analysis of the velocity maps suggests (and trajectory analysis confirms; see below) that in the west the northward flow of AABW at LGM was probably the same as or weaker than today's. Simultaneously, the eastern flank of the glacial AABW intrusion (initially through the Vema Fracture Zone) was enhanced considerably. The exchange between the eastern and northern parts of the deep North Atlantic was stronger at the LGM than today.

As the meltwater North Atlantic Current diverged from today's northeastern path, the accumulation rate in the Iceland and Irminger Basins dropped and a significant part of the sediment mass was spread over the abyssal valley in the Canary Basin (Figure 2c). Note that although the sedimentation rate is far lower there than in the Iceland Basin, the sediment mass is roughly the same because bottom area in the latitude/longitude grid cells increases rapidly to the south. The sedimentation rate around Iceland decreased by over a factor of two as compared with its modern value (compare Figure 2a and 2b). This is mainly because of the absence of ventilating convection both in the NGS and in the Irminger Sea. At the LGM, a relatively high glacial accumulation rate still emerges in the model in the Newfoundland Basin (Figure 2b). However, we do not see continuous southward sediment transport along the North American coast, a signature of all our experiments based on the modern ocean surface climatology. Instead, a noticeable southward sediment transport is found at the eastern flank of the Mid-Atlantic Ridge. At the MWE this tendency is increased significantly. The accumulation rate in the Newfoundland Basin decreased in comparison with the LGM run, whereas deposition was stronger at the eastern flank of the Mid-Atlantic Ridge. The area between Newfoundland and the Caribbean is almost totally free of sediment deposition.

5.3 Trajectory Tracing

In our experiments, the particles illustrating the flow were deployed in different areas in the northern NA (Figure 3 in Seidov and Haupt 1997). In each of the areas, about 30 particles started to travel through the Eulerian velocity fields in all three cases - the HM, MWE and LGM. In the following maps (Figure 3 and Figure 4) we employ two different techniques to show the trajectories. In Figure 3 the trajectories are colored to show the depth of a particle. (convection sites are depicted by different shades of gray: the deeper the convection, the darker the shade). We use the sunlight spectrum colors, from dark red in the uppermost layer (< 100 m) to violet and black in the two deepest layers, to visualize vertical migration of the water parcels. In contrast, Figure 4 shows the particles' pathways with both depth and elapsed time shown in small rectangles attached to the trajectories (only two pairs of the trajectories are repeated in the black and white Figure 4). Although the model time in the trajectory-tracing calculations was over 500 years, only the tracks for the first 100-200 years of the elapsed time are shown in the maps to avoid confusion.

A striking feature of the trajectory map is the change in the glacial deep ocean circulation regime which is not so obvious from the velocity maps. This change is far more complex than a simple increase in the zonality of the surface current, a well-known feature of the LGM surface circulation (CLIMAP, 1981; Ruddiman and McIntyre 1981; Kellogg 1980).

The deep water production at the LGM is found in the model only in the central part of the north central North Atlantic (Figure 3), which is in agreement with the convection pattern in Figure 4b. We note that water descends in the subpolar gyre in spite of upward motion induced by Ekman divergence. Hence thermohaline currents would drive water along isopycnals in the subsurface layers regardless of ventilating convection; i.e., ventilation of subsurface and intermediate water would occur regardless of deep convection. The pronounced convective chimney forms an intensive cyclonic circulation around a homogenized column of high-density water (Figure 3b). As particles deployed at the surface during the LGM became trapped inside this column, they were propelled downward, forming most of the 8 Sv of glacial NADW (see above).

Figure 3b indicates that there was some cold intermediate water produced in the NGS at the LGM. This water, flowing farther into the North Atlantic over the Greenland-Iceland sill, was not dense enough to subduct under the intermediate to deep water formed in the central part of the subpolar gyre. It mixed with the subpolar water and stayed at an intermediate depth flowing along the S-shaped route at that depth in the north central part of the basin (see in Plates 1d and 2b in Seidov and Haupt 1997). Most of the deep return flow occurred along the eastern slope of the Mid-Atlantic Ridge, though some water still contoured the American east coast. In contrast, water sinking in the NGS today is dense enough to descend even deeper after spilling over the sills into the North Atlantic. Together with the portion of NADW formed east of Greenland and in the Labrador Sea, this water travels southward in the western boundary current comprising most of the 13 Sv of simulated modern NADW outflow. Most of the present-day deep flow contours the American east coast, and a smaller portion is routed along the west slope of the Mid-Atlantic Ridge. We point out that the curtailment of the forward conveyor in the western part of the ocean was not complete. This implies that the conclusions based on analysis of the glacial proxy data assembled along the eastern meridional sections probably are not valid for the western part of the basin.

Figure 3a and 3b compares present and the glacial trajectories' "spaghetti" in the subtropical anticyclonic gyre. The modern and LGM trajectory maps conform to the Luyten-Pedlosky-Stommel (LPS) ventilated thermocline theory (Luyten et al. 1983) and the computer experiment of Cox and Bryan (1984). The thermocline in the subtropics is maintained by Ekman pumping. Therefore as water circulates in the gyre, it descends and ventilates the thermocline. The water is brought up to subsurface layers in the western boundary along upward sloping isopycnals (orange-colored segments of the trajectories in the vicinity of the western boundary in Figure 3). Flowing eastward, in the segment of western boundary current outflow, the water convects because of heat loss to the atmosphere (because the specified SST is colder than the outflowing subsurface water in this outflow zone). This shallow convection is shown in light gray in Figure 3. The subsurface water parcels in this ventilation zone come in contact with the atmosphere. They then start to descend again to repeat the whole ventilation cycle. Those water parcels which flow northeastward below the convection depth do not contact the atmosphere again and therefore do not ventilate. Yet, at any given time, new water parcels descend from the surface in the ventilation zone. However, ventilation is restricted to the central and western parts of the gyre. In complete agreement with the LPS theory, there is a shadow zone attached to the eastern boundary, practically unreachable for ventilation. Modern elapsed times (Figure 3a) for a single ventilation cycle in the gyre agree well with the estimates of Cox and Bryan (1984); short loops have an advection age of no more than 3 to 5 years before the water returns to the ventilation zone, whereas the longer paths in the subtropical gyre take more than 10 to 15 years to return to ventilation zone. It is obvious from the colors of the trajectories in Figure 3 that the glacial thermocline is more deepely ventilated than its modern analogue. There is evidence (Slowey and Curry 1992 1995) that the subtropical thermocline was indeed better ventilated during the last glacial period. As reveals, our Lagrangian calculations agree well with these findings.

Figures 3 and 4 indicate that the speed of the particles in the upper and intermediate layers was not lower at the LGM than today. The deep and abyssal glacial flows, though routed differently from today, were even stronger in the eastern part of the basin. This confirms the concept that the LGM conveyor was at least as intense as the present one (Yu et al. 1996; Boyle 1996). However, we stress that, in contrast to a similar or even higher-intensity meridional conveyor at intermediate depths, the forward conveyor was definitely weaker at the upper to intermediate depths.

The MWE trajectories indicate that at that time slice there was no ventilation of the deep ocean in the NA (Figure 3c), although there was still quite intensive ventilation of the subtropical thermocline. The spaghetti form of the MWE trajectories indicate also that the Nordic Seas were rather isolated at that time. The selected trajectories illustrate the collapse of the forward conveyor and delineate the shallow and slow motion within the upper ocean layers.

6 Modeling of the Global Conveyor

6.1 Convection Regime

Figure 6 shows the present-day velocity vectors at 150 m (a) and 2500 m (b) depths. Figures 7 and 8 depict the same for LGM and MWE, respectively. A comparison of the MWE and HM and LGM deep ocean currents (Figures 6b, 7b, and 8b) indicates that the major change of the conveyor indeed took place at MWE, as the consideration of the convection regime implies. Here we concentrate on tracing the conveyor using the Lagrangian particles. As was mentioned above, these trajectories visualize the true three-dimensional water motion because they take into account both vertical motion and mixing in the convection chimneys. Since the vector maps suggest that the dramatic change of the conveyor took place at MWE rather than at LGM, we present the trajectories for the present day and MWE time slices only. Figure 9a shows two pairs of trajectories calculated using the modern velocity field and Figure 9b depicts the trajectories calculated using the MWE velocities. The particles to trace modern and MWE conveyor were deployed in different places. Two different sites were chosen because no deep convection has been found in the northern NA, whereas a site of deep convection exists southwest of Australia at both HM and MWE. However, only at MWE did particles deployed in the latter area progress westward and enter the central Atlantic. First, we briefly overview the OGCM results represented by velocity vectors and then emphasize how the semi-Lagrangian calculation enhances water transport analysis and understanding of the circulation change.

6.2 OGCM Results

Because the structure of the glacial conveyor in the NA is not principally different from the modern one, the LGM conveyor is not illustrated here. Yet the glacial conveyor is characterized by a noticeable (here almost 60%) decrease in its intensity and the deep flows, especially in the NA, take different routes. Also, the NA conveyor branch became shallower than at present. However, as shown in Seidov et al. (1996), the NA conveyor still existed at the LGM. Since there was a weaker modern conveyor in the cited study, the relative changes are larger here (see above). The present-day overturning rate (here we ended up with 23 Sv) is perhaps an overestimate. We tried to obtain the convection in the northern NA that would penetrate deep enough (deeper than 2 km) by cooling the sea-surface by 2 °C (see above). The course resolution employed in this study led to an overestimate of the convective mixing and to overturning somewhat stronger than in other coarse resolution studies. The mediating estimate of the glacial to present-day overturning ratio is perhaps within 50 to 70%.

The major changes took place at the MWE. Although the impact is tied to the NGS and the northern NA only, the whole deep conveyor to the eastern Australian coast is affected. The NADW production was completely switched off, and there is no deep southward flow in the western Atlantic. Moreover, the reversed deep ocean flow is found in this simulation to the north of a substantially curtailed deep ACC in the Indian and Atlantic sectors of the Southern Ocean. Although our calculations do not unambiguously indicate the reversal of the whole deep conveyor branch from the Atlantic to the northern Pacific, one may easily recognize that the deep flow has indeed an opposite direction over rather a long leg from the eastern Indian ocean to the northern NA. Additionally, a clearly seen deep southward-flowing western boundary current developed in the northern Pacific (Figure 8b). This flow, which is absent from the present day deep current system, is a signature of the possibly reversed Pacific branch of the global conveyor.

6.3 Trajectory Tracing

Some of the particles deployed in the NGS and northern NA penetrated into the deep ocean and traveled southward in the western boundary current (Figure 9a). Some of these particles pass further into the South Atlantic, although many remained trapped in the NA subtropcal gyre. Some of those which manage to travel to the South Atlantic turn backward. About 10 to 20% of all particles deployed in the shaded area in Figure 5a reached the ACC and were transported further eastward in this current. Only 2 to 5% of these particles ever emerged in the northern Pacific. Hence, this study indicates that only a tiny fraction of the NADW can physically travel along the whole leg of the deep conveyor. However, it does not mean that the NADW water cannot reach the Pacific in significant amounts. One must consider the duration of the process, which can be of several thousand years of a stable modern-like conveyor. Some additional calculations that combine the Lagrangian technique with incorporation of geochemical tracers are required to quantify the total amount of NADW reaching the northern Pacific at the slow rate indicated by our experiments (no more than 1-2 Sv in the model).

In spite of uncertainties in our present-day calculations due to a rather coarse horizontal resolution and a somewhat incomplete glacial and meltwater sea-surface climatology, the MWE curtailment of the conveyor and emergence of the reversed deep conveyor is a very robust feature. In many additional experiments with additional disturbances of MWE sea-surface conditions, a switched-off or even reversed NA conveyor was a permanent feature. All attempts to find a trajectory originating somewhere in the NA and continuing into at least the Indian Ocean have failed. On the contrary, the particles deployed at the surface to the southwest of Australia travel far into the South Atlantic. A tiny fraction of the deployed ensemble was found north of the equator. However, no particles managed to pass into the northern NA. Hence, although there was indeed a strong incursion of the AABW into the NA, it upwelled mostly to the south of 50°N, which agrees well with the results of Seidov et al. (1996), who show isopycnal outcrop to the south of 50°N. Such a southward-shifted density outcrop isolated the northern NA from the rest of the World Ocean. Hence, the main driving mechanism of the deep ocean circulation during MWE was restricted to the Southern Ocean.

7 Discussion and Conclusions

The Lagrangian calculations help to elaborate the true three-dimensional water motion and therefore are the only means by which conveyor modes can be genuinely visualized. Moreover, the trajectory-tracing technique may indicate whether specific parts of the global ocean are interconnected via the deep conveyor branches, or essentially isolated. For example, the northern NA during the MWE is characterized by very old nonventilated intermediate-to-deep water (Sarnthein et al. 1995). Our calculations may shed some additional light on this problem. If the MWE water in the NA was a mixture of the AABW, originating in the Weddell Sea and some water sinking in the eastern Indian ocean, it may explain extreme aging of the Atlantic water, stronger than it would be if only the AABW ventilated the NA areas.

A combined circulation/sedimentation/particle tracing modeling approach was employed to understand particular aspects of the glacial-interglacial change of the North Atlantic and World Ocean circulation which are difficult to address using single-component models. Primary among those are the ventilation regimes, including the characterization of vertical mixing in convective chimneys and the advective ages of ventilated water. Circulation studies might focus on the NADW outflow or deep ocean circulation routing. Most of these questions are traditionally addressed using geochemical tracers such as d ¹⁸O, d ¹³C, and D ¹⁴C, which are extremely useful in ocean circulation studies and hence widely employed. However, the tracers alone cannot provide sufficient constraints over a simulated past circulation, as has been shown recently by LeGrand and Wunsch (1995). In their study they showed that there exist an infinite number of states that would satisfy a tracer distribution aimed at constraining the circulation, at least for the currently available proxy data sets.

Another problem arising in paleoceanographic investigations is the parallel analysis of both surface and benthic habitats and/or sediment transport features. Commonly, assumed features of a water flow thought to be suitable for explaining a particular distribution of proxy data are largely based on speculation. There is no guarantee that this hypothetical flow would satisfy hydrodynamic restrictions posed over the ocean by the wind stress, ocean geometry, bottom morphology, and sea-surface heat and fresh waterfluxes.

Based on only partly known SSS, recently corrected SST, and simulated glacial wind stress, our results largely conform to current ideas about ventilation and overturn in the North Atlantic at the height of the last glaciation and during the subsequent meltwater event near 13 500 ¹⁴C years B.P. The sediment transport model output contains features that agree well with interpretation of sediment data in the northern North Atlantic. In Seidov and Haupt (1997) we discussed the LGM sedimentation pattern resulting from our simulation in comparison with the observed or inferred data of Cremer et al. (1993) and McCave and Tucholke (1986). It was pointed out that our simulations are in good qualitative agreement with their findings. Moreover, in our experiments, the sediment load in the vicinity of the North American coast is lower during the LGM than today, in agreement with the interpretation given by Boyle and Keigwin (1987) on the basis of nutrient concentration analysis. The Caribbean water at depths just above 2 km was nutrient-depleted during the glacial period. One interpretation is that there was a lower proportion of northern component water, which, in turn, means a weaker deep western boundary current and a stronger southern source water incursion.

In the eastern part of the NA, our sediment model agrees with data that indicate increased flow along the eastern flank of the Mid-Atlantic Ridge. Dowling and McCave (1993) and Robinson and McCave (1994) provide evidence that the Feni Drift was substantially enhanced at the last glacial maximum. Generally, the glacial sediment record indicates an enhanced Holocene bottom current driven by the Iceland-Greenland overflow characteristic of modern NADW production in the Norwegian-Greenland Seas.

Our trajectory-tracing model reveals very different fates of the water volumes in the HM and LGM cases, once exposed to the air-sea interaction and then mixed downward in convective chimneys. NADW production in the center of the glacial North Atlantic is strongly supported by the proxy data [Duplessy at al. 1988 1991; Sarnthein et al. 1994 1995). In general, the most recent studies indicate enhanced upper NADW production and decreased lower NADW production (e.g., Oppo et al. 1995). However, there exists evidence that, despite decreased lower NADW production, the glacial conveyor was not, in total, weaker than today (Yu et al. 1996). This may appear contradictory to the fact that the glacial northward heat transport was substantially weaker at the LGM. Our trajectory analysis explains this seemingly controversial result. It shows that although the forward conveyor responsible for meridional heat transport became weaker, the abyssal reversed branch of the conveyor strength increased proportionally to allow the same rate of the southward transport of the tracers in the deep ocean found by Yu et al. (1996). Mediating the results of Seidov et al. (1996), Seidov and Haupt (1997) and the global conveyor simulations presented above, one may give the ratio of LGM to HM conveyor intensity as within 50-70% of modern magnitude.

Duplessy et al. (1988) argue that during glacial times most of the deep eastern North Atlantic was filled with southern source water which probably penetrated up to 45°N. Our experiments, deploying particles in the surface layers in the subpolar gyre and in the deep and abyssal areas in the subtropical gyre, are in complete agreement with this conclusion. Moreover, Michel et al. (1995) point to a steep gradient in d¹³C distribution northward of 30°N in the North Atlantic delineating the border between southern and northern source water at the LGM. We add that this border probably shifted farther to the north in the eastern part of the basin during that time period.

Moreover, we point out that the trajectory analysis is in agreement with the notion of a far better ventilated and deeper glacial thermocline in the subtropical North Atlantic (Slowey and Curry 1995). These authors provide indications that the glacial thermocline was shallower than today with its base raised by about 100 m. However, the water inside the glacial thermocline was up to 4 °C cooler than today. Our Figure 3 confirms that the glacial thermocline was better ventilated and that there was an enhanced production of subtropical mode water. Deeper glacial ventilation, evident in Figure 3b, implies that the base of thermocline had to rise upward because water pumped into the thermocline was more than 4°C colder at the glacial ventilation points (westward boundary outflow region; compare Figure 3a, b). To quantify this effect, the difference between today's and the LGM simulated temperature is shown in a meridional section of the temperature anomaly field in Figure 8. In essence, Figure 8 displays a penetration of cold anomalies deep into the thermocline. It is clear, however, that this penetration is limited to the upper 1 km and that the lower 500 m of this layer is more strongly cooled than the upper 500 m. This implies that the thermocline base has been raised. A comparison of the temperature profiles in the subtropical gyres indicates that on average the thermocline depth was about 200 m shallower than today, which differs from the estimates of Slowey and Curry (1995). There may be several reasons for this discrepancy. Firstly, the subtropical gyre is dangerously close to the sponge layer at the southern wall, which might have distorted the behavior of the thermocline base. Second, we use the annual mean surface forcing and, therefore, we are unable to simulate an extreme winter southward migration of the density outcrop intersecting the Ekman pumping within the subtropical gyre. Nevertheless, the general tendencies sketched in Figure 1 of Slowey and Curry (1995, p. 717) are evident in our trajectory maps  (Figure 3b as compared to Figure 3a). The particles dive deeper into the thermocline and stay there longer. The northern limit of the subtropical gyre, marked by the shallow convection along the North Atlantic Drift, is shifted noticeably southward.

As some findings have indicated (e.g., Sarnthein et al. 1995), the Nordic Seas were more isolated from the northern North Atlantic at the LGM than today. This is strongly supported by our Lagrangian calculations. As Figure 7 displays, the water from the subtropics probably had a very limited chance to enter the Nordic Seas, largely because the outflow over the Iceland-Greenland Ridge was reduced during the LGM.

Based on a threefold numerical simulation of the North Atlantic circulation and sedimentation and twofold simulation of the World Ocean water transport, we draw the following conclusions:

1. Present-day and LGM meridional thermohaline circulation is characterized by a forward global deep ocean conveyor. At MWE the freshwater fluxes that maintain the "observed" high latitudinal sea-surface salinity appear to be sufficient to suppress this forward conveyor. A reversed deep ocean conveyor replaced the forward one on the whole distance from NA to the southeast Indian Ocean.
2. The convection regime is the most crucial process for the conveyor dynamics. At MWE the bimodal convection regime, with both NADW and AADW sources, was replaced by the regime with only southern deep water source driving the conveyor. As a result, a reversed conveyor emerged in the Indian-Atlantic sector of the Southern Ocean. The meltwater North Atlantic was essentially isolated from the other parts of the World Ocean.
3. Sedimentation in the North Atlantic is non-linearly coupled to the circulation modes associated with different surface climatology. The LGM and MWE sedimentation patterns differ fundamentally from the Holocene/Modern pattern in a way that cannot be expected a priori. The major differences indicated by sediment transport and trajectory-tracing models are summarized in following conclusions.
4. The results of sediment transport model suggest that there was almost no sediment deposition in the Gulf Stream area, and a lower rate than at present in the Caribbean during MWE.
5. The trajectory-tracing model indicates far stronger ventilation of the thermocline in the glacial low latitudes than is observed today. The trajectory-tracing model indicates complete absence of the deep ocean ventilation at the MWE. The model reveals disappearance of the deep western boundary current and westward contraction of the subtropical warm pool. This contraction leads to collapse of the eastern continuation of the North Atlantic drift and contributes to formation of the upper ocean reverse conveyor.
6. Concerning the use of numerical models as a principal source of sediment transport and Lagrangian calculations, it should be stressed that at high latitudes, the sediment transport and water-volume tracing models can only operate with the simulated rather than observed circulation patterns. The water transport calculations depend critically on knowledge about where the convection occurs and how deep it ventilates.

Acknowledgements

References