The Atlantic-Mediterranean exchange of water at Gibraltar represents a significant heat and freshwater sink for the North Atlantic and is a major control on the heat, salt and freshwater budgets of the Mediterranean Sea. Consequently, an understanding of the response of the exchange system to external changes is vital to a full comprehension of the hydrographic responses in both ocean basins. Here, we use a synthesis of empirical (oxygen isotope, planktonic foraminiferal assemblage) and modeling (analytical and general circulation) approaches to investigate the response of the Gibraltar Exchange system to Atlantic freshening during Heinrich Stadials (HSs). HSs display relatively flat W–E surface hydrographic gradients more comparable to the Late Holocene than the Last Glacial Maximum. This is significant, as it implies a similar state of surface circulation during these periods and a different state during the Last Glacial Maximum. During HS1, the gradient may have collapsed altogether, implying very strong water column stratification and a single thermal and δ18Owater condition in surface water extending from southern Portugal to the eastern Alboran Sea. Together, these observations imply that inflow of Atlantic water into the Mediterranean was significantly increased during HS periods compared to background glacial conditions. Modeling efforts confirm that this is a predictable consequence of freshening North Atlantic surface water with iceberg meltwater and indicate that the enhanced exchange condition would last until the cessation of anomalous freshwater supply into to the northern North Atlantic. The close coupling of dynamics at Gibraltar Exchange with the Atlantic freshwater system provides an explanation for observations of increased Mediterranean Outflow activity during HS periods and also during the last deglaciation. This coupling is also significant to global ocean dynamics, as it causes density enhancement of the Atlantic water column via the Gibraltar Exchange to be inversely related to North Atlantic surface salinity. Consequently, Mediterranean enhancement of the Atlantic Meridional Overturning Circulation will be greatest when the overturning itself is at its weakest, a potentially critical negative feedback to Atlantic buoyancy change during times of ice sheet collapse.
The Mediterranean-Atlantic exchange through the Strait of Gibraltar dominates the heat, salt, mass, and energy budgets of the Mediterranean Sea [Bethoux, 1979; Bryden and Stommel, 1984] and provides a major influence on eastern North Atlantic circulation at depths <2000 m close to 35°N [Özgökmen et al., 2001] and at intermediate depth (500–1500 m) throughout much of the midlatitude North Atlantic [Iorga and Lozier, 1999]. As the Gibraltar Exchange effectively replaces North Atlantic Central Water with relatively cold and salty Mediterranean Outflow Water [Bryden and Stommel, 1982; Garcia-Lafuente et al., 2009], it is an important heat and freshwater sink, promoting and sustaining deep convection in the Nordic Seas [Bigg et al., 2003]. The export of buoyancy from the Atlantic caused by the Gibraltar Exchange does not seem to be a dominant control on the relatively vigorous meridional convection of today [Rahmstorf, 1998] but is anticipated to be of far greater importance during phases of slow or stagnant convection in the Arctic [Rogerson et al., 2006a]. Consequently, it is vital to constrain the flux and properties of the water masses exchanged at Gibraltar in the past, and given that the exchange is primarily driven by the Mediterranean-Atlantic salinity gradient [Bryden et al., 1988], it is particularly important to constrain the behavior of the exchange during phases of Atlantic freshening.
Although Sierro et al.  inferred from stable oxygen isotope ratios in planktonic foraminifera that surface water of reduced salinity was advected from the Atlantic into the Mediterranean through the Strait of Gibraltar, no IRD has been found on the Mediterranean side of the Strait, so that icebergs themselves do not seem to have entered the Mediterranean. In other words, no Heinrich Event sensu stricto (defined by IRD) has been found within the Mediterranean. To describe the incursions of reduced-salinity, cold (as defined by alkenone paleothermometry [Cacho et al., 1999, 2001]) surface water through the Strait of Gibraltar at the time of Heinrich Events in the North Atlantic (i.e., during Heinrich Stadials), we therefore define the term “Alboran Freshening Event” (AFE). We need to introduce this new term to allow differentiation of the consequences for Mediterranean hydrology of the freshened surface water inflow (the focus of this study) from the climatic impacts of the Heinrich Stadials proper (as described, for example, by Bar-Matthews et al.  and Rohling et al. ).
Arid conditions within the Mediterranean basin result in a net evaporative loss (termed “excess evaporation,” Xmed) from the Mediterranean surface waters of 52–66 cm year−1 [Bryden et al., 1994; Garrett et al., 1993], which causes salinity to be higher in the Mediterranean water mass than in the adjacent Atlantic [Bryden and Kinder, 1991]. This results in a distinct interface between relatively high and low salinity water within the narrow and shallow bottleneck of the Strait of Gibraltar [Bryden et al., 1994] (Figure 1). The interaction of dense Mediterranean subsurface waters with relatively buoyant North Atlantic Central Water (NACW) on either side of the Strait of Gibraltar drives a two-layer exchange [Bryden et al., 1988]. Modified Atlantic Water (MAW) flows into the western Mediterranean at the surface and Levantine Intermediate Water with a minor admixture of western Mediterranean Deep Water [Kinder and Parrilla, 1987] flows as a bottom layer into the Atlantic, forming the Mediterranean Outflow Water (MOW) [Bryden et al., 1994; Kinder et al., 1988; Millot et al., 2006; Stommel et al., 1973].
The two-layer structure of the Gibraltar Exchange dominates circulation within the adjacent Gulf of Cadiz and Alboran Sea (Figure 1) [Lacombe and Richez, 1982]. Superimposed on this structure, significant tidal, annual and meteorologically produced sub-inertial (i.e., short wavelength) variations occur [Bormans et al., 1986; Candela et al., 1989; Garrett et al., 1990; Millot, 2008; Tsimplis and Bryden, 2000]. These influences combine to cause the surface circulation to attain one of three circulation states [Folkard et al., 1997] which occur in (1) the summer under easterly winds, (2) the summer under westerly winds, and (3) the fall-winter-spring. During prevalence of circulation pattern 1, the central Gulf of Cadiz displays colder SST than the Alboran Sea, whereas the Gulf of Cadiz is warmer than the Alboran Sea during the other two circulation patterns [Folkard et al., 1997]. At all times, minimum SST is found within the Strait of Gibraltar, due to local upwelling of MOW on the eastern side of the shallowest section, the Camarinal Sill [Folkard et al., 1997].
Seasonally, sea surface salinity varies by less than 0.5 within the Gulf of Cadiz [MEDATLAS, 2002], and it is thus justified to assume a constant annual base-line value of 36.5 for the present-day configuration. Similarly, we assume single annual base-line values for surface salinity in the Alboran Sea (37.3) and in the MOW (38.5) [MEDATLAS, 2002]. Simple two-end-member mixing can be invoked to explain the ∼0.8 offset between Gulf of Cadiz and Alboran Sea surface salinity values, indicating a 42 ± 8% admixture of upwelled MOW water, varying slightly seasonally. The upwelling within the Strait of Gibraltar also affects the offset in SST between the Gulf of Cadiz and Alboran Sea described in the previous paragraph. This influence varies seasonally due to smaller difference between winter SST in the Gulf of Cadiz (∼16.6°C) and MOW temperature (∼12.9°C, which varies only slightly seasonally [Tsimplis and Bryden, 2000]), compared to that between Gulf of Cadiz summer SST (∼22.6°C) and MOW temperature [MEDATLAS, 2002].
As the Gibraltar Exchange is known to have been active during Heinrich Stadials and the Last Glacial Maximum (LGM) [Rogerson et al., 2006a; Voelker et al., 2006], the basic pattern of surface exchange outlined above would have been similar, leading to rapid west to east transit of Atlantic sourced hydrographic signals. Consequently, any change on the Atlantic side of the Strait of Gibraltar will be transmitted to the Alboran Sea within years or decades, well below the resolution of most marine paleoceanographic records. Consequently, major hydrographic transitions (such as incursion of IRD, or strong cooling) can be correlated across the region with a high degree of confidence, and the Heinrich Events, though necessarily slightly pre-dating AFE's, can practically be assumed to be coincident with them. Here, we use empirical records of sea surface conditions combined with modeling to test what impact AFE's had on the Gibraltar Exchange, and thus on the balance of freshwwater, salt and heat transport between these basins.
2.1. Synchronization of Records
We use ten records from eight locations, representing most of the well-dated, high resolution records from the transect area, which together provide good coverage of the Gibraltar region. All of the records used in this study have independent age models, with extensive AMS 14C chronostratigraphic control (for original sources see Table 1). However, to facilitate comparison between the various records, it is necessary to confirm that the key periods under investigation are synchronous in the records and that events such as the Heinrich Stadials were not time-transgressive. We thus define a single standard chronology, for which we selected MD95-2043 from the Alboran Sea [Cacho et al., 1999; Sierro et al., 2005] on the basis of its exceptional coherence with the GISP2 Greenland ice core major ion series [Rohling et al., 2003] and the fact that this core lies at the down-stream limit of North Atlantic-sourced freshening pulses. Only Heinrich Stadial 1 and Heinrich Stadial 2 are investigated here, as older events were reached by only a few cores. Full chronological “tuning” of the records is not necessary for our assessment, as our analysis requires only that Heinrich Stadials 1 and 2 are regionally synchronous. To determine the position of these events within the core records, we use significant changes in surface temperature, ecology or the occurrence of IRD as proxy markers; the bases of the assignment of the position of HS1 and 2 in individual cores are summarized in Table 1. We use a broad definition for the LGM, defining it as the period between the Heinrich Stadial 1 and Heinrich Stadial 2 peaks of Arctic planktonic foraminiferal species and/or IRD. This makes age determination of this period solely dependent on accurate recognition of regionally synchronous Heinrich Stadials, also allowing analysis of mean conditions during this period without further chronostratigraphic analysis being necessary.
Figure 2 shows the age assignment of HS1 and HS2 within the records as identified within the original published chronostratigraphic models of the records. To assess the degree of error in age assignments (shown as bars above and below the interval of the HS events within the core records in Figure 2), the original AMS 14C datings were recalibrated to calendar age using the OxCal software, the IntCal04 atmospheric isotope curve and a 400 year reservoir age. Linear interpolation of the 2σ confidence intervals of 14C radiocarbon dates then provides a means of assessing the approximate magnitude of the dating error for any point within the chronostratigraphy of the record (for further information on this approach, see the chronostratigraphic discussion by Rogerson et al. ).
Figure 2 indicates that datings for HS1 have an excellent degree of agreement across the range of the locations used within this work, confirming that this is an isochronous event and it is justified to assume that conditions during this period can be reconstructed as a discrete time-slice. HS2 shows a higher degree of scatter, as would be expected for an older event, and MD99–2339 shows an age for HS2 significantly different from MD95-2043. However, this may be a consequence of deviation of sedimentation rate in this core from the linear interpolation of the 14C-based chronostratigraphic model, as HS2 is constrained only by 14C determinations at 19.039 and 28 cal. ka BP. It should be noted that the original publication of the MD99–2339 record [Voelker et al., 2006] incorporated an additional age constraint at 23.8 ka BP using the N. pachyderma (s) maximum associated with HS2, implicitly assuming that there has been some drift from linearity in the accumulation rate through this interval.
The high degree of synchronicity between these records for HS1 gives high confidence that a genuine time-slice can be created for paleo-hydrographic analysis, and that we will not be collapsing a diachronous event into a single scenario. Notwithstanding the deviation in MD99–2339, we will also assume synchronicity for HS2 and reconstruct this as a time-slice as well. Differences in sampling resolution, difficulties of precise sub-millennial-scale correlation and uncertainty regarding the precise timing and occurrence of single or multiple meltwater pulses within each Heinrich Stadial prevent compilation of data for “peak” conditions. Consequently, we determine (and use) the mean value of each record for Heinrich Stadial 1, Heinrich Stadial 2, and the LGM interval. We further use core top data as a means of representing the system during the late Holocene for comparison of proxy and modern hydrographic data.
2.2. Compilation of Microfossil and Geochemical Data Sets
The majority of data used has been previously published, except for the planktonic foraminiferal fauna from cores M39029–7 and MD99–2339 and the new SST estimates based on planktonic foraminiferal assemblages. SSTs were reconstructed with the Artificial Neural Network (ANN) method [Malmgren et al., 2001], using the MARGO-project database for the Atlantic [Kucera et al., 2005]. This database was selected due to the location of the Mediterranean cores, as western Alboran Sea surface water is essentially Atlantic surface water giving plankton assemblages more affinity to modern Atlantic locations than modern eastern Mediterranean ones. Moreover, initial investigations revealed that estimates of SST were similar regardless of the database used, especially in the case of samples from Heinrich Stadial 1. Values represent the mean of 10 simulations, with samples where the standard deviation exceeded 1 discarded. Data sources are summarized in Table 1, and original δ18OG. bulloides and relative abundance records of Neogloboquadrina pachyderma, Turborotalita quinqueloba and Globigerinoides ruber are presented in Figures 3 and 4, respectively. Three δ18OG. bulloides records are available for M39029 (cores −4, −7 and −8): all three are presented in Figure 3, but a single set of average values is used to represent this site for later analyses.
2.3. One-Dimensional Modeling of the Gibraltar Exchange
To minimize the number of assumptions needed, we use a straightforward representation of the Gibraltar exchange, which employs a hydraulic control model in combination with mass and salt conservation statements. The model assumes that the transports are maximal, i.e., that the two-layer Froude number ≥1 [Bryden and Kinder, 1991]. This model is represented by
where Qtotal determines the two-way exchange across the sill (Qatl-Qmed, where Qmed is a negative vector term), ΔSgib is the vertical salinity difference between layers within the Strait, Ws and Ds are the width and depth at the sill, C is a constant depending on sill geometry ( = 0.28 [after Bryden and Kinder, 1991]), g = 9.81 m s−2, β is a coefficient used to convert ΔSgib to Δρgib (β = 7.7 × 10−4 g cm−3ppt−1 [after Bryden and Kinder, 1991]), and ρmed is density of the outflowing Mediterranean water.
One aspect of the system that is not addressed via this simple representation is the way excess Mediterranean salinity will be purged under enhanced exchange conditions. ΔSgib may be expected to decline over the period of freshened water supply, forcing a gradual decrease in Qtotal. Consequently, the behavior of the 1D modeling described above must be assumed to reflect conditions during the first decades to centuries of the AFEs only, and more sophisticated modeling is required for a fully functional understanding of the Mediterranean response to Atlantic freshening to be developed.
2.4. General Circulation Modeling
General Circulation Model (GCM) results are obtained using an LGM setup [Levine and Bigg, 2008] of a global ocean GCM that is coupled to a simple energy balance atmosphere. The ocean component of the model is described by Wadley and Bigg , and uses a curvilinear coordinate system [Madec and Imbard, 1996]. The horizontal model grid emphasizes the North Atlantic deep-water formation areas around the Nordic Seas, where the resolution is 1°–2°, while the Southern Hemisphere resolution is 6°–8°. The ocean has 19 levels in the vertical, varying in thickness from 30 m near the surface to 500 m at depth, and has a free surface formulation for the barotropic model [Webb, 1996]. The ocean model is coupled to a simple radiative, advective atmosphere, which consists of the energy moisture balance model used in the UViC Earth Systems model [Fanning and Weaver, 1996], with the addition of advection of water vapor. The GCM is adapted for the glacial simulations with global ice sheet cover from Peltier , and by adjusting sea level, orbital parameters, and CO2 levels for the LGM (21 ka BP) period. The surface fluxes of heat and freshwater are determined from ocean-atmosphere interaction, while the atmospheric wind stress has an annual cycle [cf. Dong and Valdes, 1998]. A dynamic and thermodynamic iceberg trajectory model is also coupled to the ocean model, for both the control run and for simulations of Heinrich events [Levine and Bigg, 2008]. The performance of the model is discussed more fully by Levine and Bigg , but for the present-day North Atlantic the sea surface properties compare reasonably well with the climatological values, although the gradients associated with the Gulf Stream are not fully resolved. The net Atlantic overturning strength is too high at ∼28 Sv (observed values are ∼18.7 ± 5.6 Sv [Cunningham et al., 2007]) but much of this excess is contained within an unrealistically strong, but spatially limited, North Atlantic cell, and the export fluxes out of the North Atlantic of ∼15 Sv are similar to observations [Gordon, 1986; Schmitz, 1995]. The exchange through the Strait of Gibraltar is similar to the observational value of ∼1 Sv [Bryden et al., 1994].
For the glacial state, the coupled model was run until it reached a steady state that is reasonably consistent with paleo-observations, with North Atlantic Deep Water being found at intermediate depths [Weinelt et al., 1996]. The glacial overturning rate is reduced to ∼10 Sv, while the flux through the Strait of Gibraltar is ∼2 Sv. This is different to the analytical model, which predicts glacial Qtotal to be reduced relative to today, because of low spatial resolution in the Strait of Gibraltar. The GCM tends to be too cold in the northern Atlantic; there is an ice-covered Nordic Sea but paleo-observations suggest this was seasonally ice-free [Kucera et al., 2005].
Heinrich Events are simulated by releasing a large flux of freshwater locally from the Hudson Strait for a period of 500 years. A flux of 0.4 Sv is chosen to ensure a complete collapse of the Atlantic overturning circulation, which occurs within 20 years. This is similar to, but a little higher than, existing flux estimates [Roche et al., 2004]. We also carried out a simulation where the same flux of freshwater was released in the form of icebergs, which were allowed to move about the Atlantic and slowly melt, producing a more realistic spreading of the direct release of freshwater over a much wider region, but at a slower rate [Levine and Bigg, 2008].
3.1. The δ18OG. bulloides Gradients
The core top gradient in stable oxygen isotope values of the planktonic foraminiferal species Globigerina bulloides (δ18OG. bulloides) between 10°W and 3°E is similar to that predicted for calcium carbonate precipitated in equilibrium from surface water in early spring (Figure 5), the main season of G. bulloides growth in the region today [Pujol, 1980]. This suggests that the paleodata compilations presented in Figure 3 offer a useful representation of surface hydrographic conditions, revealing consistently heavier δ18OG. bulloides values in the Mediterranean than in the Atlantic throughout the last 30 Cal. ka BP (Calibrated kilo-years Before Present; hereafter referred to as ka). This W–E gradient in δ18OG. bulloides is exemplified by comparison of cores D13898 [Rogerson et al., 2004] and MD95-2043 [Cacho et al., 1999; Sierro et al., 2005] (Figure 6). These cores respectively represent the Gulf of Cadiz and Alboran Sea, have similar and relatively high sampling resolution and also have a high degree of synchronicity in the age assignment of HS1 and HS2 making them the most easily comparable pair of records. Figure 6 shows a stable offset of about 0.7‰ during the early Holocene, 1.0–1.1‰ during the Younger Dryas and Bølling-Allerød, and about 1.8‰ during the LGM. During Heinrich Stadial 1, however, the isotope contrast between D13898 and MD95-2043 appears to have reduced to <0.5‰. This reduction is supported by the compilation of records shown in Figure 5. A smaller reduction in gradient is found during Heinrich Stadial 2 (Figure 5), although this is not so evident from comparison of D13898 and MD95-2043 alone (Figure 6). Linear regressions to the data from all cores (shown in Figure 5) indicate a core top west to east δ18OG. bulloides gradient of 0.129‰ degree−1, which was increased to 0.258‰ degree−1 during the LGM. Heinrich Stadials 1 and 2 show west to east gradients that are much reduced compared to the LGM of 0.098‰ degree−1 and 0.145‰ degree−1, respectively (Figure 5).
3.2. SST Gradients
Compilation of mean summer and winter gradients across the Strait of Gibraltar, including values from hydrographic data, is shown in Figure 7. It is encouraging to note that ANN values for core tops suggest a similar seasonal change in gradient as that seen in hydrographic data, even though the absolute values are somewhat different (Figure 5). However, before ∼9 ka BP summer SST appears to have been higher in the Gulf of Cadiz (Figure 7), a reversal of the core top pattern that indicates a significant change in mean surface circulation conditions comparable to a change resulting from a reversal of the prevailing wind.
During the LGM, comparison of the Atlantic and Mediterranean ANN-derived summer and winter SST records shows significantly higher SST in the Gulf of Cadiz than in the Alboran Sea, with differences of up to 7°C in summer and up to 5°C in winter (Figure 7). More detailed comparison based on D13898 and MD95-2043 (Figure 8) shows that this SST gradient between the Gulf of Cadiz and the Alboran Sea generally decreases at the end of the deglaciation, followed by a minor further reduction during the Holocene. Between 17.5 and 16 ka (Heinrich Stadial 1) the Atlantic-Mediterranean temperature difference is reduced by more than 3°C, with a complete collapse centered on 16.5 ka. No analogous collapse is seen at the time of Heinrich Stadial 2, when the maximum summer SST drop in the Gulf of Cadiz is only ∼3°C (in M39029–7) leaving a significant offset from Mediterranean SST, which is largely unchanged during this time. However, the results for Heinrich Stadial 2 must be treated with some caution due to the low sampling resolutions of the records through that interval.
3.3.1. One-Dimensional Modeling
The primary impact of surface freshening related to AFEs is an increase in the net exchange (Qtotal) across the Camarinal Sill. The magnitude of this impact is illustrated in Figure 9a. The relationship between Qtotal and freshening is slightly nonlinear across the range of freshenings investigated (ΔQtotal ∝ 0.13ΔSatl − 4.7 × 10−3ΔSatl2). The response of Qtotal to sea level change is linear (Figure 9b), where ΔQtotal ∝ 8.8 × 10−3ΔRSL. This behavior is consistent with the response reported for ΔQatl previously reported by [Rohling and Bryden, 1994], which is one half the size of the change in ΔQtotal as a result of the relationship between these two flux parameters (ΔQtotal = 2ΔQatl − Xmed). Xmed is the least constrained of the boundary conditions, but planktonic foraminiferal δ18O evidence from the Mediterranean suggests that it varied only slightly during Heinrich Stadial 1 [Bigg, 1995; Rohling, 1999]. Nevertheless, exchange at Gibraltar is highly sensitive to the freshwater export flux, with higher fluxes driving enhanced exchange and vice versa. Mediterranean freshwater export flux (Xmed) shows a nonlinear control on Qtotal (Figure 9c), where ΔQtotal ∝ 7.15ΔXmed2/3.
3.3.2. GCM Modeling
Figure 10 shows the response of the Gibraltar exchange to two North Atlantic iceberg release experiments, designed to simulate a Heinrich Event. With the onset of the Heinrich freshwater forcing (time 0), both simulations register an increase in Qatl, with the iceberg simulation having a smaller effect because the freshwater is released over a wider region. Subsequent to the initial increase, both models show a decline in the scale of the anomaly back toward values typical of the control run, reflecting recirculation of low-density surface water within the Mediterranean. Significantly, an anomaly remains in both models until termination of the freshwater forcing after 500 years (Figure 10), with a non-smooth return over time toward the original exchange flux.
Figure 11 shows the zonal velocity anomaly for the iceberg release experiment (hosing-control) just to the east of the Strait of Gibraltar, in the western Alboran Sea. Virtually no response is present below 500 m, apart from some slight increases in westward flux around 750 m. Virtually no response is present below 500 m, apart from some slight increases in westward flux around 750 m. The increase in Qtotal therefore concerns predominantly the intermediate waters above that level (primarily Levantine Intermediate Water), compensated by an increase in Atlantic Inflow. This is consistent with previous suggestions based on multiproxy data reflecting bottom water renewal processes [Rogerson et al., 2008].
4.1. Thermal Gradient
Two significant changes occur in the Strait of Gibraltar SST offset (ΔSSTSoG), namely, (1) an expansion of ΔSSTSoG in all seasons during the LGM and (2) a significant reduction of ΔSSTSoG during Heinrich Stadial 1.
4.1.1. Increased Surface Temperature Gradient During the LGM
The strengthening (relative to the present) of the LGM ΔSSTSoG implies either increased vertical mixing in the Strait of Gibraltar or that cooling of some origin is being applied to surface water during its flow through the Gibraltar transect. The difference between the mean winter temperature offset between the Gulf of Cadiz and Alboran Sea (ΔwSST) is 4.7 ± 0.8°C (error for ΔwSST estimated by propagation of 1σ values for all Gulf of Cadiz and Alboran Sea LGM values via σΔwSST = ), which is significantly greater than the late Holocene separation between wSST for the Gulf of Cadiz and MOW (3.6°C). Peak values in the separation between D13898 and MD95-2043 are ∼7°C (Figure 8), double the present-day difference between Gulf of Cadiz winter SST and the temperature of MOW. It therefore seems unlikely that the observed glacial to interglacial changes in ΔSSTSoG can be explained by admixture of MOW to surface waters alone.
Confirmation of the LGM strengthening in ΔSSTSoG is provided by the δ18OG. bulloides data, which shows an approximate doubling of the gradient (Figure 5). The δ18OG. bulloides data also offer some potential to further constrain the origin of the cold water, as admixture of MOW would cause a strong increase in the offset of δ18OG. bulloides from the Gulf of Cadiz and Alboran Sea, both due to the change in temperature of calcification and due to the admixture of isotopically heavy intermediate and deep waters to surface water in the Alboran Sea [Schmidt et al., 1999]. Modern MOW is 0.5‰ heavier than regional surface waters, and the difference for the LGM likely was significantly higher due to the residence time effect [Rohling, 1999]. If the δ18Owater offset increased approximately in proportion with ΔSGib, this would suggest a glacial offset of ∼1‰. For the LGM, Δδ18OG. bulloides is 1.52 ± 0.42 ‰ and, assuming that δ18Ocalcite decreases with increasing temperature by 0.23‰ °C−1 [O'Neil et al., 1969], this suggests that the δ18Owater difference between Gulf of Cadiz and Alboran Seawater was in the region of 0.44‰. This indicates that the proportion of upwelled MOW in the surface waters of the Alboran Sea was similar to today (40–50%). However, propagating the standard deviations from the raw data gives a very high 1σ range for this estimate (±0.47‰) so it is not possible to make this case unambiguously.
Changes in the hydrographic structure on both sides of Gibraltar have been previously reported for the LGM: to the west the Azores Front is argued to have penetrated into the Gulf of Cadiz [Rogerson et al., 2004], and to the east the Alboran Gyres are believed to have been inactive before about 8 ka [Rohling et al., 1995]. Incorporation into the Gibraltar inflow of cold surface water transported along the northern shelf of the Gulf of Cadiz would effectively place the Gulf of Cadiz and Alboran Sea locations on opposing sides of the Azores Front, which represents a ∼4°C transition [Gould, 1985]. Incursion of cold surface water from the Gulf of Lions transported down the southern margin of Iberia would similarly provide cooling without significant increase of δ18Owater, with strong LGM cooling of the northwest Mediterranean magnifying this influence [Hayes et al., 2005; Kuhlemann et al., 2008]. Indeed, the concept of flow of cold water down the east Iberian margin during the LGM is given implicit support by the reconstructions of Hayes et al. [2005, Figure 9b] and also in those of Kuhlemann et al. , which both show a strong temperature anomaly in this area. Notably, the strongest temperature anomaly in the Hayes et al.  reconstruction is during the summer, when we find ΔSSTGib to be reversed relative to today. Despite the uncertainty of the large propagated error in δ18Owater, we therefore argue that the glacial to interglacial changes in gradient most likely represent a significant reorganization of regional surface water circulation to a state not reported in modern data sets [Folkard et al., 1997], with the collapse of the Alboran gyres allowing incursion of surface water from the Gulf of Lions transported southward due to enhanced atmospheric flow and reduced momentum in the Alboran surface layer caused by the reduced inflow of Atlantic water [Rogerson et al., 2005; Rohling and Bryden, 1994].
4.1.2. Reduction in Gradients During Heinrich Stadials
During Heinrich Stadial 1, both the summer and winter SST and δ18OG. bulloides gradients were considerably reduced (Figure 7). The strong and consistent cooling across the Strait of Gibraltar found during the LGM therefore does not appear to have occurred during Heinrich Stadial 1, indicating a reduction in mixing during transport through the Strait. However, mean Δδ18OG. bulloides for Heinrich Stadial 1 is 0.59 ± 0.61 ‰, which in the absence of strong SST gradients would imply slightly heavier δ18Owater to the east of Gibraltar, and consequently the maintenance of some upwelling within the strait. In combination, these observations imply that Gulf of Cadiz surface water was at a similar temperature to MOW. The timing of this period is consistent with an observed collapse of the gradient between one of the cores in this study (D13898) and MD95-2042 on the Portuguese margin [Rogerson et al., 2004], and it seems likely that surface water gradients were generally small from offshore Lisbon to the eastern Alboran Sea. The inferred simultaneous collapse/withdrawal of the Azores Front and reduction in mixing within the Strait of Gibraltar indicate a significant enhancement of surface water stratification, as would be expected during admixture of a large flux of freshwater from iceberg melting.
During Heinrich Stadial 2, the response in the δ18OG. bulloides records was smaller than during Heinrich Stadial 1 (Figure 5) and very little SST change is observed (Figure 7). We infer that the response of the Gibraltar exchange to this event was small relative to Heinrich Stadial 1. In this context, it should be noted that Heinrich Stadial 2, unlike Heinrich Stadial 1, also failed to cause a collapse in the surface hydrographic gradients between the Gulf of Cadiz and the Portuguese margin and so does not appear to have strongly affected the Azores Front [Rogerson et al., 2004]. It should be noted, however, that the amount of data available for Heinrich Stadial 2 is far less than for Stadial 1, the proof of synchronicity is less secure and it therefore remains possible that this apparently lower magnitude impact is an artifact of sampling.
4.2. Synthesis of Modeling and Empirical Evidence
The surface gradients at Gibraltar appear to be approximately comparable for Heinrich Stadial 1 and the early Holocene, in spite of significant climatic differences between these two periods. The reason for this similarity must therefore reflect a similarity in the level of control over the region exercised by the Gibraltar Exchange at these times. Both 1-Dimensional and GCM modeling indicate that the response of the Gibraltar Exchange to a Heinrich Event in the Atlantic would be for the Gibraltar exchange flux (Qtotal) to increase (Figures 9 and 10). Consequently, during both the Holocene and Heinrich Stadial 1, Qtotal was significantly higher than during the LGM, making the underlying regional hydrographic structure similar during these two times. The empirical and modeling evidence therefore provide a coherent picture in which enhanced Gibraltar exchange causes simplification of regional surface circulation patterns, effectively flattening regional surface property gradients.
4.3. Temporal and Spatial Structure of Enhanced Gibraltar Exchange During Heinrich Stadial 1
The methods presented in this study show a high level of consistency in their reconstruction of hydrographic changes during Heinrich Event 1, and add confidence to inferences of freshened surface water incursion into the westernmost Mediterranean reported in previous literature. A subtle but potentially important difference lies in the fact that, whereas in this study AFE's are apparently sustained throughout Heinrich Events, the low δ18Owater pulses reported by Sierro et al.  are often of extremely short duration (a few centuries) and positioned toward the end of each Heinrich Stadial period [see Sierro et al., 2005, Figure 3]. Given the control of the system detailed in Figure 9, negligible rates of sea level rise during Heinrich Stadial 1 [Hanebuth et al., 2000; Peltier and Fairbanks, 2006] and the lack of evidence for enhanced freshwater export flux in marine δ18O compilations [Rohling, 1999], the inference must be that surface water in the Gulf of Cadiz was freshened throughout the Heinrich Stadial 1 period. Freshening in the Gulf of Cadiz therefore did not only occur as discrete, short duration peaks of iceberg melting, but was sustained throughout Heinrich Stadial periods. This continuous meltwater input requires a considerable flux of icebergs/meltwater into the southwest Iberian margin region, as has been suggested from GCM experiments [Levine and Bigg, 2008].
The GCM results provide insight into the vertical structure of the anomalous flow within the Strait of Gibraltar induced by Atlantic freshening, showing virtually no change in flow at depth within the westernmost Mediterranean but strong velocity anomalies in the surface and intermediate layers. Consequently, enhanced circulation within the western Mediterranean seems to be restricted to the upper 750 m. This observation is important, as it may partially decouple the Gibraltar exchange changes described in this study from changes in bottom ventilation in the Alboran Sea described elsewhere [Cacho et al., 2000, 2006; Frigola et al., 2007; Jiménez-Espejo et al., 2007; Sánchez-Goñi et al., 2002; Sierro et al., 2005]. Transmission of ventilation changes to the deep western Mediterranean basin therefore must have occurred via the Bernoulli suction mechanism outlined by Rogerson et al. .
4.4. Consequences of an Enhanced Gibraltar Exchange
For the Mediterranean, the consequences of enhanced exchange are reduced residence time and slightly reduced mean salinity. Neither of these impacts is easily resolvable in the proxy record beyond the immediate vicinity of Gibraltar [Rogerson et al., 2008]. However, from an Atlantic perspective, enhanced Gibraltar exchange would result in an increased flux of surface buoyancy out of the Atlantic and an increased flux and decreased salinity (and therefore increased buoyancy) of MOW injection into intermediate depths. Increased flux is reflected in peaks in MOW activity throughout the Gulf of Cadiz during Heinrich Stadials [Llave et al., 2006; Toucanne et al., 2007; Voelker et al., 2006]. Enhanced flow at relatively shallow depths [Llave et al., 2006; Toucanne et al., 2007] also provides evidence for reduced-density related shoaling of the MOW plume [Rogerson et al., 2005]. A larger, shallower MOW plume within the North Atlantic would have consequences for recovery of Atlantic meridional overturning subsequent to freshening related to Heinrich Event iceberg meltwater [McManus et al., 2004; Sarnthein et al., 1994; Schönfeld and Zahn, 2000], as the northward transport of intermediate depth salt would be enhanced and the probability of mixing this salt to the surface would also be enhanced.
4.5. Further Significance of the Exchange Enhancement Process
Increased trans-Gibraltar SST and δ18OG. bulloides gradients during the LGM indicates a circulation not analogous to the present. The δ18OG. bulloides and SST data combined indicate increased mixing with cold water during eastward transport of surface water through the Strait of Gibraltar during the LGM compared to today. The size of the LGM SST offset between the Gulf of Cadiz and Alboran Sea makes it unlikely that all of this additional cooling is due to increased upwelling of MOW. Admixture of northern surface waters transported from the Portuguese margin and the Gulf of Lions are proposed as alternatives, with the latter preferred due to published evidence of southward extension of cold, surface water in this direction [Hayes et al., 2005; Kuhlemann et al., 2008]. Removal of temperature effects from δ18OG. bulloides is ambiguous due to large propagated errors, but provides some support for this concept. Transport of northern surface waters down the eastern margin of Iberia reflects decreased momentum within the surface layer of the Alboran Sea, related to the decreased flow of inflowing Atlantic water.
Significant reductions in the sea surface temperature gradient and planktic foraminiferal δ18OG. bulloides gradient across the Strait of Gibraltar during Heinrich Stadial 1 indicates a major change in surface circulation patterns. Despite significant climatic differences, the Heinrich Stadial 1 and early Holocene gradients are similar, suggesting similar circulation patterns and thus high fluxes of inflowing Atlantic water. 1-D and GCM modeling provides an explanation for this, namely that during both periods the Gibraltar exchange was significantly higher than during the LGM. The changes are not so apparent during Heinrich Stadial 2, suggesting that the impact of this event on the Gibraltar exchange was smaller. However, this event is not as well resolved as Heinrich Stadial 1 in our data set, so some doubt over this observation remains.
Enhanced Gibraltar exchange is consistent with changes in ocean bottom conditions in the Alboran Sea and changes in the activity of the MOW. Furthermore, changes in North Atlantic water masses during the last deglacial provide an explanation for the repeated intensifications of the MOW reported by previous authors [Rogerson et al., 2006a]. Consequently, when surface water in the North Atlantic freshens, reducing the rate of Meridional Overturning, the exchange at Gibraltar is actually enhanced, increasing the loss of buoyancy via cooling and evaporation within the Mediterranean basin. This observation could be critical in understanding how Atlantic Meridional Circulation restarts after periods of stagnation.
The GCM modeling work was supported by the NERC Joint Rapid Climate Change program through grant NE/C509523/1. E. J. Rohling acknowledges support from NERC projects Quantifying the Earth System (QUEST-deglaciation) and Response of Humans to Abrupt Environmental Transitions (RESET, NE/E01531X/1). We thank Michal Kucera (Universität Tübingen) and Montserrat García (University of Salamanca) for their help with the MARGO database and the ANN-SSTs calculations. A. Voelker acknowledges support from FCT through the MOWFADRI project. We also thank M. Kucera, L. Skinner, an anonymous reviewer, and the editors for their comments, which significantly improved the quality of this paper.