Using a cost-efficient climate model, the effect of changes in overturning circulation on neodymium isotopic composition,ϵNd, is systematically examined for the first time. Idealized sequences of abrupt climate changes are induced by the application of periodic freshwater fluxes to the North Atlantic (NA) and the Southern Ocean (SO), thus mainly affecting either the formation of North Atlantic Deep Water (NADW) or Antarctic Bottom Water (AABW). Variations in ϵNd reflect weakening and strengthening of the formation of NADW and AABW, changes in ϵNdof end-members are relatively small. Relationships betweenϵNd and the strength of NADW or AABW are more pronounced for AABW than for NADW. Atlantic patterns of variations in ϵNd systematically differ between NA and SO experiments. Additionally, the signature of changes in ϵNd in the Atlantic and the Pacific is alike in NA but opposite in SO experiments. Discrimination between NA and SO experiments is therefore possible based on the Atlantic pattern of variations in ϵNd and the contrariwise behavior of ϵNd in the Atlantic and the Pacific. In further experiments we examined the effect of variations in magnitudes of particle export fluxes. Within the examined range, and although settling particles represent the only sink of Nd, their effects on ϵNd are relatively small. Our results confirm the large potential of ϵNd as a paleocirculation tracer but also indicate its limitations of quantitative reconstructions of changes in the Atlantic Meridional Ocean Circulation.
 Reconstructions of past changes in ocean overturning circulation have been proposed from records of paleocirculation proxies that reflect the distribution of water masses or the rate of ocean circulation and that can be measured in sediment cores. For this reason, considerable work has been put into the investigation of the cycling of various elements, as well as elemental and isotope ratios (see Lynch-Stieglitz  for an overview). However, an unequivocal reconstruction of changes in past overturning circulation is difficult, and still not even possible for the Last Glacial Maximum (LGM), a period of time for which a wide range of data is available [Lynch-Stieglitz et al., 2007; Marchal and Curry, 2008; Hesse et al., 2011]. A relatively novel and promising quasi-conservative tracer of water mass distribution and mixing is the isotopic composition of neodymium (ϵNd) [von Blanckenburg, 1999; Burton and Vance, 2000; Frank, 2002].
 The purpose of this study is to systematically examine the effect of changes in overturning circulation on ϵNd in idealized sequences of abrupt climate change for the first time. The timing of the sequences is chosen to shed some light on the characteristics of ϵNdexpected during Heinrich and Dansgaard-Oeschger (DO) events. This may contribute to a better knowledge of the effects of changes in the strength of the ocean overturning circulation of different magnitude and duration onϵNd.
 The Nd isotopic ratio (143Nd/144Nd) of a sample relative to the “bulk earth” standard value of 0.512638 [Jacobsen and Wasserburg, 1980] is reported as
Isotopes 143Nd and 144Nd have recently been included into a cost-efficient, dynamical ocean model of intermediate complexity and are simulated in reasonable agreement with available observations [Rempfer et al., 2011].
 All these studies indicate the potential of ϵNd as a complementary tracer of past water mass mixing. However, some uncertainty is associated with Nd, complicating attempts to interpret variations in ϵNdsimply due to changes in overturning circulation. First of all, considerable uncertainties are associated with the nature and magnitude of Nd-sources and sinks [e.g.,Goldstein and Jacobsen, 1987; Jeandel et al., 1995; Tachikawa et al., 1999; Frank, 2002; Tachikawa et al., 2003; van de Flierdt et al., 2004; Lacan and Jeandel, 2005a; van de Flierdt et al., 2007; Arsouze et al., 2007; Siddall et al., 2008; Arsouze et al., 2009; Rempfer et al., 2011]. Though, an adequate representation of Nd sources and sinks is an important prerequisite for the simulation of present-day observations of dissolved Nd concentrations andϵNd [Rempfer et al., 2011]. In addition, Tachikawa et al.  indicated that changes in the magnitude and ϵNd of Nd sources over the years potentially affect ϵNd of deep water masses besides changes in water mass mixing. Naturally, the extent of such changes is difficult to constrain. J. Rempfer et al. (Sensitivity of Nd isotopic composition in seawater to changes in Nd sources and paleoceanographic implications, submitted to Journal of Geophysical Research, 2012) reconsidered these concerns and found that rather extreme changes in magnitude and ϵNd in Nd sources are necessary to cause a magnitude in variations in ϵNdas observed, e.g., on glacial-interglacial time-scales. Furthermore, the interpretation of variations in reconstructedϵNdas changes in overturning circulation relies on the assumption that changes in the end-member composition can be neglected. Indeed,ϵNd in the North Atlantic seems to have been relatively stable during parts of the last 30 kyr [van de Flierdt et al., 2006] and during the last 5 glacial cycles [Foster et al., 2007], respectively. However, these studies were either restricted to short time periods [van de Flierdt et al., 2006] or rely on data with relatively low temporal resolution [Foster et al., 2007]. Moreover, Gutjahr et al.  and Arsouze et al.  indicated that during the LGM ϵNd of North Atlantic Deep Water (NADW, i.e., of Glacial North Atlantic Intermediate Water, GNAIW) may have been different. Finally, the paleoceanographic potential of ϵNd has not been examined in a comprehensive way in modeling studies, as these were restricted either by computational costs or by the chosen approach. That is, to date only one modeling study is available that is based on a relatively simple approach of restoring ϵNd at the continental margins [Arsouze et al., 2008]. This study aimed to examine the effects of changes in Meridional Overturning Circulation (MOC) during the LGM on ϵNd.
 Our comprehensive approach together with the high computational efficiency of the Bern3D model allows us to examine the paleoceanographic potential of ϵNd in a systematic way and it will therefore contribute to the interpretation of variations in reconstructed ϵNd as past changes in overturning circulation.
 In order to examine the sensitivity of ϵNd in seawater to transient changes in overturning circulation, different periods and amplitudes of the forcing are investigated (section 4). Besides, we evaluate potential effects of variations in particle fluxes on variations in ϵNd (section 5). Further information that may be derived from variations in ϵNd in sediment cores, such as changes in the rate and the origin of changes in the overturning circulation, is discussed in section 6. Finally, we summarize our findings and draw some conclusions regarding the paleoceanographic potential of ϵNd (section 7).
Table 1. Comparison of Characteristic Numbers of the CTRL Nd Parametrization of Rempfer et al.  and the Parametrization Used in This Studya
fbs (g Nd yr−1)
I (g Nd)
[Nd]p/[Nd]dis the ratio between particle-associated and dissolved Nd,fbs is the magnitude of the boundary source, I is the global ocean Nd inventory, and τNd is the mean residence time of Nd in the ocean (defined as the ratio of I and total sources ftot, i.e., τNd = I/ftot). and indicate global mean average deviations of simulated from observed [Nd]d and ϵNd, respectively, and are calculated following Rempfer et al. [2011, equation 15].
 In order to examine the effect of changes in MOC, e.g., during idealized sequences of abrupt change on ϵNd at the seafloor, we applied periodic varying freshwater fluxes of different amplitude (Fw) and periodicity (TFw, Table 2) to the North Atlantic (NA experiments) and to the Southern Ocean (SO experiments). In NA experiments perturbations were applied between 45 and 70°N, in SO experiments they were applied to the Ross and Weddell Sea areas of the model (170–180°W, 63–71°S and 40–60°W, 63–71°S respectively, see also Figure 4b). The time pattern of the freshwater-fluxes is triangular in each experiment, i.e. freshwater-fluxes vary linearly between +Fw and –Fw and increase from 0 to 1/4·TFw. Predominantly but not exclusively, positive/negative freshwater-fluxes lead to a weakening/strengthening of the formation of deep-water masses in the corresponding region, i.e., in the formation of NADW and Antarctic Bottom Water (AABW). In general, patterns of changes in experiments whereFw = 0.1 and 0.2 Sv are intermediate between experiments where Fw = 0.05 and 0.3 Sv.In NA experiments we decrease the freshwater flux corrections that are applied to match observed atmospheric moisture transport from the Atlantic to the Indo-Pacific from the standard value of 0.34Sv [Ritz et al., 2011b] to 0.17 Sv. Reducing this moisture transport allows us to keep variations in maximum AMOC (ψAMOC) within a reasonable range [Ganachaud and Wunsch, 2000; Talley et al., 2003]. Leaving the moisture transport unchanged would require a larger amount of freshwater for a shut down of the formation of NADW and would lead to stronger AMOC during on-states.
Freshwater fluxes of different duration (TFw) and magnitude (Fw) are applied in the North Atlantic (basinwide between 45 and 70°N, NA) and in the Ross and Weddell Sea areas of the model (170–180°W 63–71°S, and 40–60°W 63–71°S, SO). These regions are indicated as light grey areas in Figure 4b.
 In addition to a general evaluation of how well ϵNd reflects changes in overturning strength, the wide range in the formation of NADW and AABW, resulting from our freshwater experiments, allows us to examine whether ϵNdcan be used to differentiate between a weakening and a cessation of the formation of NADW and AABW, respectively. Moreover, the application of freshwater-fluxes either to the North Atlantic or the Southern Ocean, largely affects the formation of one single water mass only. This makes it possible to evaluate whether inferences on the origin of changes in ocean circulation can be derived from the corresponding large-scale pattern ofϵNd at the seafloor.
 In our model export fluxes of biogenic particles opal, CaCO3, and POC are calculated from prognostic equations and therefore vary, e.g., with changes in ocean circulation. The magnitude of particle export fluxes affects the magnitude of the sink of Nd and thus its mean residence time (τNd), which in turn affects the water mass property of ϵNd [Rempfer et al., 2011]. To evaluate the effect of changes in the magnitude of export fluxes on ϵNd, we perform additional experiments where we scale particle fields of opal, CaCO3, and POC from the CTRL by factors 0.5, 1 and 2 (see Table 3 for a description of experiments).
Table 3. Acronyms for Experiments Where Particle Export Fluxes of the CTRL Experiment Were Scaled by Factors 1/2, 1, and 2 in the North Atlantic and the Southern Oceana
Freshwater-fluxes (TFw = 2 kyr; Fw = 0.3 Sv) were applied in the North Atlantic and the Ross and Weddell Sea areas of the model, thus corresponding to experiments NA2k030Sv and SO2k030Sv, respectively.
 As ϵNd in sediment records has been reported to reflect ϵNd in bottom water [e.g., Piepgras and Wasserburg, 1980; Frank, 2002] we present results of ϵNdfrom lowermost grid cells. In our approach Nd enters the ocean across the sediment-water interface between the sea surface and 3000 m depth [Rempfer et al., 2011] and we therefore focus on results from depths greater than 3000 m.
4. Freshwater Experiments: Variations in MOC Strength and ϵNd
 Meridional Overturning Circulation below the wind driven mixed layer is represented by a stream function which can take positive and negative values (Figures 1a–1o). A positive (negative) stream function is referred to as a clockwise (counterclockwise) overturning circulation.
 In the modern steady state (CTRL) a clockwise overturning cell in the North Atlantic, with a strength of about 14 Sv, is associated with the formation of NADW. About 1 Sv of AABW is advected northward below the tongue of NADW via counterclockwise overturning in the deep South Atlantic (Figure 1a). A southern and a northern cell can also be distinguished in the Pacific. However, in contrast to the North Atlantic, almost no deep convection takes place in the North Pacific and large parts of the basin are filled with southern source water (i.e., AABW, Figure 1b). AABW is formed in the Southern Ocean and is then advected northward into the different basins (Figure 1c).
 On and off-states for experiments NA2k030Sv and SO2k030Sv are shown inFigures 1d–1o as representative examples.
 During NA2k030Sv on-state clockwise overturning in the Atlantic is stronger than in the CTRL (22 instead of 13Sv) and the Atlantic basin is largely filled by NADW. Accordingly, no AABW is advected northward from the Southern Ocean (Figure 1d).
 Compared to the Atlantic, changes are less pronounced in the Pacific Ocean. Northward advection of southern source water, associated with counterclockwise circulation in the deep Pacific, is stronger by about 1 Sv and some intermediate water is formed in the North Pacific (Figure 1e).
 Overturning in the deep Southern Ocean is hardly affected by the changes in the North Atlantic (Figure 1f).
 During NA2k030Sv off-state overturning is sluggish in the Atlantic basin. No NADW is formed in the North Atlantic and hence no northern source water is advected southward (Figures 1g and 1i). Instead, southern source water fills large parts of the basin (Figure 1g).
 Compared to the Atlantic Ocean and similar in magnitude to the on-state, changes are less pronounced in the Pacific, where northward advection of southern source water is weaker than in the CTRL by about 1Sv (Figure 1h).
 During SO2k030Sv on-state clockwise overturning in the North Atlantic is slightly weaker than in the CTRL (11Sv instead of 13 Sv, Figure 1j). At the same time, overturning is considerably stronger in the Southern Ocean (Figure 1l) from where about 7 Sv of southern source water are advected northward in the Atlantic basin. Due to the increase in the formation of AABW, a larger fraction of the Atlantic basin is filled by this water mass, and NADW is confined to shallower depths (Figure 1j).
 The increased formation of AABW can also be observed in the Pacific Ocean, where compared to the CTRL, advection of southern source water to the north is increased by about 6 Sv (Figure 1k).
 During SO2k030Sv off-state the formation of AABW almost ceases (Figure 1o) and no southern source water is advected northward in the Atlantic (Figure 1m). Clockwise overturning in the North Atlantic is only slightly weaker than in the CTRL (11 Sv instead of 13 Sv) and as a result the basin is largely filled by northern source water (Figure 1j). Advection of southern source water to the north is also reduced in the Pacific Ocean (by about 2 Sv compared to the CTRL, Figure 1k).
 In the following, results of ϵNd are presented in Hovmöller plots along tracks through the Atlantic, the Indian, and the Pacific oceans. The corresponding tracks are indicated in Figure 4a. Moreover, time series of ϵNd are shown from selected sites in the Northwest Atlantic, the Northeast Atlantic, the South Atlantic and the Southern Ocean. Their locations are indicated in Figure 4b. Areas where freshwater-fluxes are applied in the NA and SO experiments are also indicated inFigure 4b.
 Before showing results from a number of experiments, and in order to better illustrate the effects of changes in overturning on ϵNd, we will explain two of the experiments, one of each NA and SO experiments, in greater detail (Figure 5). Applying periodically varying freshwater-pulses to the North Atlantic and the Ross and Weddell Sea areas of the model affects maximum AMOC (ψAMOC), maximum PMOC (ψPMOC) and the minimum global MOC (ψSOMOC) to some extent (Figures 5a–5d). However, while in experiment NA2k030Sv major changes are confined to ψAMOC, changes are largest in ψSOMOC in experiment SO2k030Sv. Overall, ψAMOC and ψSOMOCare roughly in phase with freshwater-perturbations. Both become weaker when freshwater is added and stronger when freshwater is removed from the corresponding region in the North Atlantic and the Southern Ocean (Figures 5a and 5c and 5b and 5d).
 Global maps of maximum variations in ϵNd (ΔϵNd, i.e., the maximum variation in ϵNd; ϵNdmax − ϵNdmin) indicate that ϵNd is affected throughout the global ocean and that variations in ΔϵNd are not globally uniform (Figures 5g and 5f). In experiment NA2k030Sv ΔϵNdis particularly large in the South Atlantic where main water masses alternate during on and off-states as well as along the western boundary in the North Atlantic where NADW is advected southward during on-states. Similarly, in experiment SO2k030Sv ΔϵNd is large in the South Atlantic. However, ΔϵNd is of similar magnitude in the Southern Ocean in the area where formation of AABW takes place and is smaller in the North Atlantic where changes in addition are largely confined to the eastern boundary.
 Spatially averaged time series from North Pacific and South Atlantic regions (10 to 50°N and from the equator to 40°S, respectively) show that ϵNd increases to more positive values due to a shutdown of the formation of NADW and drops to more negative values following a resumption of the formation of NADW (Figure 5g). On the other hand, the temporal evolution of ϵNd in the North Pacific and the South Atlantic is in antiphase in experiment SO2k030Sv (Figure 5h). In this case, ϵNd becomes more negative in the South Atlantic and more positive in the central Pacific due to a reduction in ψSOMOC. Correspondingly, ϵNd increases to more positive values in the South Atlantic and decreases to more negative values in the central Pacific following an increase in ψSOMOC.
 Overall, global maps of ΔϵNd and spatially averaged time series of ϵNd indicate substantial differences between NA and SO experiments.
4.1. Northern Perturbation
 Adding freshwater to the North Atlantic (NA experiments) leads to a reduction in ψAMOC, while removing freshwater leads to an increase (Figures 6e–6h). In NA experiments ψAMOC varies between about 5 and 9 Sv (for TFw = 0.4 kyr and Fw = 0.05 Sv), and between about 0 and 24 Sv (for TFw = 2 kyr and Fw = 0.3 Sv). ψPMOC, varies between about 3 and 5 Sv (for TFw = 0.4 kyr and Fw = 0.05 Sv), and between about 2 and 12 Sv (for TFw = 20 kyr and Fw = 0.3 Sv). Variations in ψPMOC are in phase with variations in ψAMOC for TFw = 0.4, 1, 2 kyr, but are in antiphase for TFw = 20 kyr (Figures 6i–6l). ψSOMOC varies between about –22 and –24 Sv (for TFw = 0.4 kyr and Fw = 0.05 Sv) and between about –12 and –28 Sv (for TFw = 20 kyr and Fw = 0.3 Sv, Figures 6m–6p). While smallest variations in general are associated with TFw = 0.4 kyr and Fw = 0.05 Sv, largest variations are not necessarily associated with experiments where TFw = 20 kyr, but rather with experiments where TFw= 2 kyr. This is probably due to long-term basin-scale adjustments in the former case.
 In general, ϵNd is affected only to a small extent in experiments where both Fw and TFw are small. On the other hand, ϵNd varies considerably if TFw and/or Fw are larger (Figure 7). Apart from TFw and Fw the magnitude of ΔϵNd also depends on the geographic location.
 In the deep Atlantic, ΔϵNd is largest (up to 5 ϵNd-units) between the equator and 40°S, where northern and southern source waters alternate during AMOC-on and AMOC-off-states.
 Regarding the deep Pacific, ΔϵNd is largest (up to 4ϵNd-units) in the central North Pacific. Note that ΔϵNd in the Indian exceeds 2 ϵNd-units only in a few, particularly NA experiments (Figures 7d, 7g, 7h, and 10h), and only slightly exceeds 1 ϵNd-units in the other experiments.
 Overall, in the Atlantic, the Indian and the Pacific relatively negative ϵNdduring AMOC on-states is replaced by more positiveϵNdduring AMOC off-states (Figures 8e–8p). While northern source water is exported from the North Atlantic in periods of strong AMOC, less or no northern source water is advected southward in case of a sluggish or ceased AMOC. Instead, a larger part of the Atlantic is filled by southern source water leading to more positive ϵNd (Figures 8e–8h). Accordingly, less relatively negative ϵNd is advected from the Atlantic to the Indian and the Pacific, in periods of a sluggish or ceased AMOC, and thus ϵNd increases in these basins as well (Figures 8i–8p).
 In experiment NA20k030Sv effects on ϵNd in the Pacific are larger than in other experiments (Figures 7h and 8p). In this experiment PMOC is quite strong during AMOC off-states (up to 12Sv, Figure 6l) and an overturning cell emerges in the North Pacific leading to the formation of intermediate water (not shown). This probably increases the export of relatively positive ϵNd out of the Pacific basin leading to less negative ϵNdin the Indian as well as in the Atlantic during AMOC on-states.
 In general, effects of variations in overturning strength on ϵNd are almost identical during the first period and in following periods of freshwater forcing.
4.2. Southern Perturbation
 The effect of adding and removing freshwater to and from the Ross and Weddell Sea areas of the model (“Southern Perturbation”, SO experiments, Figure 9), differs from NA experiments. For example, while major changes occur in ψSOMOC, ψAMOC is only slightly affected in most of the experiments (Figures 9e–9h). An exception is experiment SO20k030Sv in which ψAMOC varies strongly (between about 2 and 15 Sv). Variations in ψPMOC are relatively small, even in experiment SO20k030Sv (between 1 and 4 Sv, Figures 9i–9l). In contrast, variations in ψSOMOC are considerable even in experiments where TFw and Fw are small (between –15 and –20 Sv for TFw = 0.4 kyr and Fw = 0.05 Sv, Figures 9m–9p). Not surprisingly, largest variations are observed for TFw = 20 kyr and Fw = 0.3 Sv (between –8 and –38 Sv). In general, ψSOMOCdecreases/increases in phase with the application of positive/negative freshwater-fluxes to the Ross and Weddell Sea areas of the model (in contrast toψAMOC and ψPMOC, which both are only roughly in phase with freshwater-fluxes).
 Similar to the NA experiments, ϵNd varies with changes in overturning strength and ΔϵNd also depends on the geographic location. However, the spatial pattern of ΔϵNd in SO experiments is different (Figure 10). Major variations in ϵNd are confined to the Atlantic section of the Southern Ocean (up to 4 ϵNd-units) and the North Pacific (up to 3ϵNd-units). ΔϵNd is generally small in the Indian (up to 1 ϵNd-unit). The effect onϵNd in the Atlantic reaches further north for larger amplitudes of TFw or/and Fw. In addition, ΔϵNd in the North Atlantic and in the North Pacific increases with increasing amplitude of TFw or/and Fw. An important feature is that effects on ϵNd in the North Atlantic are confined to the eastern part of the basin and that smaller effects on ϵNd arise in the western part of the North Atlantic (Figures 10e–10g), except for experiment SO20k030Sv (Figure 9h). Therefore, our results indicate differences between NA and SO experiments.
 As mentioned above, effects on ϵNd are different in experiment SO20k030Sv (Figure 10h). In this experiment AMOC is relatively weak (down to 2 Sv, Figure 9h). Therefore, southward transport of NADW-likeϵNdis reduced leading to more southern source water-likeϵNd in the entire North Atlantic, not only in the eastern part of the basin.
 Another important difference between NA and SO experiments is that changes in ϵNd are not of equal sign in the Atlantic, the Indian and the Pacific in SO experiments: In the Atlantic ϵNd is slightly more negative during periods of weak SOMOC and considerably more positive during periods of strong SOMOC (Figures 11e–11h). A similar pattern, although only weakly pronounced, emerges in the Indian (Figure 11i–11l). However, in the Pacific the pattern is vice versa, i.e., ϵNd is considerably more positive during periods of weak SOMOC and slightly more negative during periods of strong SOMOC (Figures 11m–11p). In most of the SO experiments ψAMOC and ψPMOC are affected only slightly, and only ψSOMOC varies over a large range (Figure 9). Changes in ϵNd are therefore mainly due to changes in ψSOMOC. ϵNd of AABW is intermediate compared to ϵNd in the North Atlantic and the North Pacific. Thus, if ψSOMOCis strong, export of relatively positive AABW-likeϵNd from the Southern Ocean into the Atlantic is more pronounced, leading to a shift to more positive ϵNdin this basin. At the same time, export of relatively negative AABW-likeϵNd from the Southern Ocean into the Pacific is enhanced, leading to a shift to more negative ϵNd. On the other hand, ϵNdis more NADW-like in the Atlantic and more North Pacific-like in the Pacific in case of weakψSOMOC. As ϵNdin the Indian is generally AABW-like, effects are small in this basin.
 In contrast, in NA experiments mainly the formation of NADW is affected which carries the most negative ϵNd. An increase/decrease in strength of ψAMOC thus leads to more negative/positive ϵNd in the Atlantic, the Indian and the Pacific.
4.3. A Note on End-Member Stability
 The stability of the corresponding end-member is of importance for the interpretation of downstream variations inϵNd as changes in overturning circulation [e.g., Piotrowski et al., 2004]. Figure 12illustrates Atlantic and Pacific cross-sections of differences inϵNd at times when most positive and most negative ϵNd occur in experiments NA2k030Sv and SO2k030Sv (see the figure caption for more details). It becomes clear, that changes in ϵNdat sites of deep-water formation in the North Atlantic or the Southern Ocean end-member are smaller than further downstream (by about 3–4ϵNd-units). Besides, in case of experiment NA2k030Sv, the small changes in the North Atlantic end-member are not uniform in sign. Taken together, this indicates that in our modeling experiments where overturning strength varies on millennial-scale, the observed changes inϵNdare largely due to changes in water mass distribution and that effects of changes in end-member composition are relatively small.
5. Effect of Changes in Particle Export Fluxes on ϵNd
 In experiments of Schmittner , in which AMOC was shut down due to the application of a freshwater-flux, effects on marine ecosystems were not confined to but were largest in the North Atlantic, i.e., in the region where shoaling of the mixed layer prevents upwelling of nutrients. Independent of the exact global pattern and magnitude of changes in nutrient supply and thus export fluxes, we examine the overall effect of variations in export production onϵNd by comparing simulated ϵNd from freshwater experiments (NA2k030Sv and SO2k030Sv) where we keep particle fluxes constant at different levels. Therefore, we scale particle export fluxes of CTRL (Figure 2) by factors 1/2, 1, and 2, in the North Atlantic (light grey area in Figure 4b) and in the entire Southern Ocean (south of 34°S), respectively (see Table 3 for a description of experiments). Figure 13 illustrates time series of ϵNd at four different sites (Northwest Atlantic, North East Atlantic, South Atlantic, Southern Ocean, locations are indicated in Figure 4b) resulting from these experiments. As changes in ocean overturning circulation are the same as in experiments NA2k030Sv and SO2k030Sv, differences in ϵNd can be attributed to changes in particle export fluxes.
 Although at each site absolute values of ϵNd are affected to some extent, overall patterns in ϵNd do not differ much between individual experiments. However, note that ΔϵNd is more pronounced in experiments where export fluxes are scaled by a factor of 2 (red), compared to experiments where a factor of 1/2 (blue) is applied. These differences as well as differences in absolute values of ϵNd are due to the effect of an increase/decrease in the efficiency of the sink on τNd. τNddecreases/increases in the corresponding case and results in a more/less pronounced inter-basin gradient. Besides scaling export fluxes of all particle types in tandem, we evaluated the effect of scaling export fluxes of POC, opal and calcite individually. Not surprisingly, effects onϵNd are largest if scaling is applied to POC, opal and calcite simultaneously. Results from these experiments are not shown, but lay within the colored areas (Figure 13).
6. Further Inferences From ΔϵNd
6.1. Slowdown Versus Shutdown of Ocean Overturning
 For the interpretation of ϵNd in terms of past circulation changes it is of interest whether the magnitude of changes in overturning can be inferred from variations in ϵNd. As we have shown in Figures 7 and 10 as well as in Figures 8 and 11, different magnitudes and durations of changes in overturning circulation lead to different magnitudes of ΔϵNd in the global ocean. This suggests that information about changes in the rate of the MOC can be obtained from changes in ϵNd. To further examine this suggestion, scatterplots of ϵNd versus ψAMOC and ψSOMOC at four different sites in the Atlantic are shown in Figure 14.
 In NA experiments ϵNd in the Atlantic is rather positive for weaker AMOC and rather negative for stronger AMOC (Figures 14a, 14e, 14i, and 14m). ψSOMOC varies only little in NA experiments, except for experiment NA20k030Sv, and the relationship between ϵNd and ψSOMOC is therefore generally less pronounced (Figures 14b, 14f, 14j,and 14n). Overall, the relationship between ϵNd and ψAMOC or ψSOMOC is not unequivocal at any of the sites as the range of ϵNd is large for given values of ψAMOC and ψSOMOC. For a quantitative reconstruction of ψAMOC or ψSOMOC based on a given value of ϵNd the points in the scatter plots needed to lay on a straight line. From our experiments it therefore appears that it is not possible to assign a unique strength of ψAMOC or ψSOMOC to a given value of ϵNd.
 Similar to ψSOMOC in NA experiments, ψAMOC varies only little in SO experiments (except for experiment SO20k030Sv). Therefore, no close relationship can be detected between ϵNd and ψAMOC, except for experiment SO20k030Sv (Figures 14c, 14g, 14k, and 14o). On the other hand, distinct almost linear relationships exist between ϵNd and ψSOMOC, particularly at sites in the North East Atlantic, the South Atlantic, and the Southern Ocean (Figures 14d, 14h, 14l, and 14p). In general, ϵNd is more positive for strong ψSOMOC and more negative for weak ψSOMOC. Note however, that the relationship between ψSOMOC and ϵNd does not hold for the entire range of ψSOMOC in the Southern Ocean. I.e., the relationship is not anymore unequivocal for ψSOMOC stronger than about 30 Sv.
 Although variations in ψAMOC, ψSOMOC and in ϵNd are less pronounced in experiments where Fw = 0.05 Sv overall results are similar.
6.2. Discriminating Between Northern and Southern Origin of Changes in Water Mass Distribution
 In Figures 7 and 10 we have shown that the magnitude of ΔϵNd depends on the magnitude and duration of changes in the overturning circulation to a certain extent. However, it also becomes clear that the global pattern of ΔϵNd depends on the origin of changes in water mass distribution. In NA experiments, ϵNd is affected most along the western boundary of the Atlantic, i.e. along the flow path of NADW, and in the South Atlantic (Figure 7). In contrast, in SO experiments, ϵNd is particularly affected in the Atlantic sector of the Southern Ocean, i.e. close to the location where AABW is formed (Figure 10). Besides, changes in ϵNd are more pronounced at the eastern boundary of the North Atlantic than in the western part (Figures 10a–10g, 14d, and 14h). Hence it appears that information on the origin of changes in water mass distribution can be inferred from Atlantic patterns of ΔϵNd. While ΔϵNd is largest along the western boundary as well as in the South Atlantic in NA experiments, in SO experiments ΔϵNd is largest in the Southern Ocean, the South Atlantic as well as in the Northeast Atlantic. Locations in the South Atlantic are sensitive in general. However, from experiments presented in this study it appears that it is not possible to determine whether changes in ϵNd are either due to northern or southern changes or both based on locations in the South Atlantic alone. For this purpose additional information is required from the eastern and the western North Atlantic or the Southern Ocean.
 In addition, Figures 5g and 5h as well as Figures 8 and 11 show that changes in ϵNd differ between NA and SO experiments on even larger scales. On the one hand, weaker/stronger AMOC in NA experiments leads to more positive/negative ϵNd on a global scale. On the other hand, weaker/stronger SOMOC in SO experiments causes a shift to more positive/negative ϵNd in the Pacific and to more negative/positive ϵNd in the Atlantic. I.e., changes in the Atlantic and the Pacific are equal in NA experiments but are of opposite direction in SO experiments.
 According to our results, a discrimination between northern and southern origin of changes in ocean circulation is therefore possible based on reconstructions of ϵNd from different sites (locations of these sites are indicated in Figure 4). First, the extent to which changes in ϵNd can be observed in cores from the eastern and western margin of the North Atlantic or the Atlantic sector of the Southern Ocean identifies which water masses are subject to changes. I.e., if ϵNd at the northeastern margin of the Atlantic shows variations in phase with the South Atlantic but ϵNd at the northwestern margin does not, changes in southern source water dominate over changes in northern source water. On the other hand, if similar variations are observed at the northeastern as well as at the northwestern margins, changes in both northern source water as well as in southern source water are a likely reason. Second, if ϵNd in the Atlantic and the central Pacific vary in phase, changes in ψAMOC are a likely cause. On the other hand, if variations in ϵNd in the North Atlantic and the Pacific are in opposite direction, this is probably due to changes in ψSOMOC. Note however, that according to Figures 5g and 5h dating accuracy needs to be within a few hundred years for a reliable interpretation of the contrariwise behavior of ϵNd in the Atlantic and the Pacific.
7. Discussion and Summary
 By analyzing results from numerous model simulations this study systematically examines the effect of changes in meridional overturning circulation on ϵNd for the first time. Four main results of this study are briefly mentioned here and discussed in more detail thereafter: First, variations in ϵNd of up to 5 ϵNd-units result from periodic weakening and strengthening of the formation of NADW and AABW in millennial-scale freshwater experiments. Changes inϵNdof the North Atlantic or Southern Ocean end-members are relatively small, and variations inϵNd, e.g., in the South Atlantic thus closely reflect provenance of northern and southern source water, respectively. Second, although at first sight the magnitude of variations in ϵNd depends on the magnitude of changes in Atlantic meridional overturning strength, no unequivocal relationship is found between ϵNd and the formation of NADW. The relationship between ϵNd and the formation of AABW is more pronounced. Third, the sign of changes in the Atlantic and Pacific basins as well as Atlantic patterns of ΔϵNd differ between NA and SO experiments. Therefore, inferences on the origin of changes (NADW versus AABW) are possible based on the sign of the response in the Atlantic and the Pacific as well as based on the pattern of ΔϵNd in the Atlantic. Fourth, absolute values of ϵNd are affected by variations in the magnitude of particle export fluxes to a certain extent due to the associated changes in the sink of Nd and thus τNd. However, the pattern of variations in ϵNd caused by variations in overturning strength is hardly affected.
 Overall, our results indicate that in experiments where millennial-scale variations are applied to the overturning circulation and thus to the distribution of water masses, variations inϵNd closely reflect these variations. Furthermore, in our experiments changes in ϵNdof end-members are much smaller than variations inϵNddue to changes in water mass distribution. Note however, that this does not allow to exclude changes in end-member composition on glacial-interglacial time-scales (as reported byGutjahr et al.  and Arsouze et al. ).
 Neither the pattern, nor the magnitude of ΔϵNd, and thus the sensitivity of ϵNd to changes in overturning strength, are globally uniform. Sites where records of ϵNd shall be extracted from the sediments hence need to be chosen with care. In general, largest variations in ϵNd are found at locations where the mix of major water masses varies depending on the circulation regime. In seeking maximum response in ϵNd, paleoceanographic studies should therefore focus on such regions.
 More specifically, ϵNd is sensitive to changes in AMOC, particularly in the South Atlantic, where Atlantic main water masses, NADW and AABW, prevail depending on pattern and strength of the AMOC. ϵNd is more negative during periods of strong formation of NADW and more positive during periods when formation of NADW is weak. At the same time ϵNd is more positive if formation of AABW is strong and more negative if formation of SOMOC is weak. Therefore, variations in ϵNd, e.g., in the South Atlantic do not simply reflect increasing/decreasing strength of NADW, but also an increasing/decreasing strength of AABW or a combination of changes in NADW and AABW, respectively. This is in general agreement with the interpretation of Piotrowski et al. [2004, 2005, 2008] and indicates that it is not possible to infer on the origin of changes based on one single core from the South Atlantic, without further constraints from other proxies such as δ13C (comparable to Piotrowski et al. [2008, 2009]) which in turn introduce additional uncertainties. Though, to infer on the origin of changes in the AMOC is of interest with regard to changes on millennial [Weaver et al., 2003; Piotrowski et al., 2004; Gutjahr et al., 2010] as well as on orbital time-scales [Lynch-Stieglitz et al., 2007; Piotrowski et al., 2008]. Our results indicate that constraining the origin of changes in AMOC is possible based on the pattern of ΔϵNdin the northwest and the northeast Atlantic as well as in the Southern Ocean. However, as these differences occur on relatively small, i.e., sub basin scale and as our model is only coarsely resolved, confidence into these results should be enhanced by models with higher resolution. Further large-scale differences in the temporal evolution ofϵNd in the Atlantic and the Pacific can be observed in our NA and SO experiments, thus providing additional constraints on the origin of changes in AMOC. Note, that in this regard limitations could result from dating accuracy which is required to be within the range of a few hundred years.
 On the one hand, in our freshwater experiments we find the magnitude of variations in ϵNd (ΔϵNd) to depend on the magnitude and the duration of changes in strength of the overturning circulation, thus indicating that ΔϵNd yields information about the magnitude of changes, e.g., in the rate of the AMOC. On the other hand, scatterplots of ψAMOC and ψSOMOC versus ϵNd show that it is difficult to derive information on the rate of the AMOC based on ϵNd, because the relationship between ψAMOC and ϵNd is rarely unequivocal. In contrast, the relationship is more pronounced between SOMOC and ϵNd, between which linear relationships exist at some sites. Note however, that the relationship does not hold for the entire range of SOMOC strength (i.e., for large magnitudes), e.g., in the Southern Ocean.
 Scaling of particle fluxes in regions of deep-water formation affects absolute values ofϵNd as well as the magnitude of ΔϵNd to some extent but hardly affects the pattern of ϵNd. Overall, the effect of changes in export fluxes on ϵNd is small compared to the effect of changes in circulation strength. This indicates that changes in pattern and magnitude of export fluxes do not introduce large uncertainty in the interpretation of ΔϵNdas circulation changes. It is important to note, that these results depend on the assumption of non-preferential scavenging, which is reasonable as long as no indication for preferential scavenging is found [Arsouze et al., 2009; Rempfer et al., 2011]. New insight into the marine Nd cycle will emerge, e.g., from the GEOTRACES program in the near future [SCOR Working Group, 2007].
 In this context we also emphasize general uncertainties associated with the nature and magnitude of sources of Nd. Recent studies indicate that ϵNd is modified at continental margins and young volcanic arcs and that continental margins and the seafloor could act as sources of Nd [van de Flierdt et al., 2004; Lacan and Jeandel, 2005a; Johannesson and Burdige, 2007; Rickli et al., 2010; Horikawa et al., 2010; Carter et al., 2012; Stichel et al., 2012]. This means that ϵNdof any water mass not only depends on the isotopic signature of the two main end-members North Atlantic and North Pacific. Our simplified approach takes this into account by assuming a globally homogeneous boundary source and is a reasonable first order approach considering its success in simulating modern [Nd]d and ϵNd. However, it probably does not cover the complexity associated with the boundary source and will thus prove incomplete as soon as more detailed insight is gained into the nature of this source (e.g., through the GEOTRACES program) [SCOR Working Group, 2007].
 In conclusion, our millennial-scale freshwater experiments indicate the large potential ofϵNd as a paleocirculation tracer but also indicate the limitations of quantitative reconstructions of changes in the Atlantic Meridional Overturning Circulation.
 This work was funded through the Marie Curie Research Training Network NICE (Network for Ice sheet and Climate Evolution). Support by the European Project on Ocean Acidification (EPOCA, FP7/2007–2013; 211384), Past4Future (grant 243908), the EU FP7 project CARBOCHANGE (grant agreement 264879), and the Swiss National Science Foundation are acknowledged. Thanks are due to S. Ritz for the development of the coupled Bern3D code as well as for stimulating discussions. We are also grateful to two anonymous reviewers, whose valuable comments led to significant improvements in the manuscript.