Thermohaline instability in the North Atlantic during meltwater events: Stable isotope and ice-rafted detritus records from core SO75-26KL, Portuguese margin

. A benthic isotope record has been measured for core SO75-26KL from the upper Portuguese margin (1099 m water depth) to monitor the response of thermohaline overturn in the North Atlantic during Heinrich events. Evaluating benthic õ(cid:127)80 in TS diagrams in conjunction with equilibrium õc fractionation implies that advection of Mediterranean outflow water (MOW) to the upper Portuguese margin was significantly reduced during the last glacial (< 15% compared to 30% today). The benthic isotope record along core SO75-26KL therefore primarily monitors variability of glacial North Atlantic conveyor circulation. The (cid:127)4C-accelerator mass spectrometry ages of 13.54+.07 and 20.46+. 12 ka for two ice-rafted detritus (IRD) layers in the upper core section and an interpolated age of 36.1 ka for a third IRD layer deeper in the core are in the range of published (cid:127)4C ages for Heinrich events H1, H2, and H4. Marked depletion of benthic õ(cid:127)3C by 0.7-1.1%o during the Heinrich events suggests reduced thermohaline overturn in the North Atlantic during these events. Close similarity between meltwater patterns (inferred from planktonic õ(cid:127)80) at Site 609 and ventilation patterns (inferred from benthic õ(cid:127)3C) in core SO75-26KL implies coupling between thermohaline overturn and surface forcing, as is also suggested by ocean circulation models. Benthic õ(cid:127)3C starts to decrease 1.5-2.5 kyr before Heinrich events H1 and H4, fully increased values are reached 1.5-3 kyr after the events, indicating a successive slowdown of thermohaline circulation well before the events and resumption of the conveyor's full strength well after the events. Benthic õ(cid:127)3C changes in the course of the Heinrich events show subtle maxima and minima suggesting oscillatory behavior of thermohaline circulation, a distinct feature of thermohaline instability in numerical models. Inferrred gradual spin-up of thermohaline circulation after H 1 and H4 is in contrast to abrupt warming in the North Atlantic region that is indicated by sudden increases in Greenland ice core õ(cid:127)80 and in marine faunal records from the northern North Atlantic. From this we infer that thermohaline circulation can explain only in part the rapid climatic oscillations seen in glacial sections of the Greenland ice core record


Materials and Methods
A benthic foraminiferal isotope record was measured for core SO75-26KL (37ø49.3'N, 09ø30.2'W, 1099 m water depth) from the upper Portuguese margin. The last glacialinterglacial transition and three IRD layers which occur along the core were sampled at 2 cm intervals to obtain a higher stratigraphic resolution and check for hydrographic and sedimentologic fine structure during these intervals. The rest of the core was sampled at 5-10 cm intervals. Stable isotope measurements were run on 1 to 21 specimens of Cibicidoides wuellerstorfi or C. pseudoungerianus. Within the IRD layers, isotope measurements were carried out on C. pseudoungerianus only, with a minimum of six specimens per isotope sample. The foraminiferal tests were picked from the size fractions >250 gm. In addition, a planktonic isotope record was measured using 25 specimens of Globigerina bulloides from the 315-400 gm size fraction. The planktonic isotope record was used to enhance stratigraphic control on the core. Prior to isotope analysis the foraminiferal shells were cracked open to release potential sediment fillings. They were then ultrasonically rinsed in methanol and transferred to a CARBO KIEL automated carbonate preparation device that is linked on-line to a FINNIGAN MAT 251 mass spectrometer. Long-term reproducibility was 0.08 %0 for fi•80 and 0.05 %0 for •13C as calculated from replicate analyses of an internal carbonate standard (Solnhofen limestone, 63-80 gm) that was routinely run at a ten-sample interval. The isotope data are referred to the Pee Dee belemnite (PDB scale). The •4C ages were determined via accelerator mass spectrometry (AMS) using the 3MV Tandetron  Additional XRD scans were run on samples immediately below and above the IRD layers to obtain the mineralogy of the sediments which were deposited prior to and after the IRD events.

Stratigraphy
Age control on the last glacial-interglacial transition (100-130 cm in core SO75-26KL)was obtained by correlating the planktonic õ180 record with the planktonic õ180 record from a nearby core SU81-18 that has been dated by •4C-AMS in great detail [Bard et al., 1989]. A series of 17 •4C-AMS ages was measured farther down the core (Table 1). One sample at 174-178 cm yields a •4C age (reservoir corrected; see Table 1) of 14.32 ka that fits well with the early phase of deglaciation documented by the initial decrease of benthic and planktonic õ•80 at this core depth (Figures 2a and 2b) Figure 2b). Beyond about 20 ka the Martinson et al. [1987] timescale is based on tuning to orbital parameters compared with UFFh dates. That is the age of 43.9 ka given b y Martinson et al. [1987] for oxygen isotope event 3.13 must be considered a calendar or calibrated age. Laj et al. [1996] suggest a correction of 2000 years at 40 ka (•4C), rapidly decreasing to zero around 47 ka (•4C), to convert the 14C timescale to calendar years. Thus we use a •4C age of 42 ka for oxygen isotope event 3.13 to ensure compatibility with the conventional •4C timescale that we use for core SO75-26KL.

Detailed 14C-AMS Dating of IRD Layers I and 2
(Equivalent to Heinrich Events 1 and 2) Six •4C ages each were measured across IRD layers 1 (127-143 cm) and 2 (264-278 cm) (Figure 3). Of the six 14C ages across IRD 1, the first two show an increase with depth, while the lower four form a plateau in the age-depth function and even decrease with increasing depth by 140 years. This decrease is, however, similar to the standard deviations and therefore statistically not significant. The closely similar ages at different core depths indicate increased sedimentation rates for this part of the core, which contains the heart of the IRD layer 1, between 130 and 142 cm core depth (Figure 3a). The three AMS dates for this interval give a mean age of 13.54+.07 ka (average depth 136 cm). The apparent age discontinuity between the IRD layer and overlying sediments is expected as a result of bioturbation whereby, after the rapid deposition of the IRD layer, younger tests of G. bulloides are mixed down to and into the upper reaches of the IRD layer, but few older tests are mixed up. This leads to artificially young dates for the upper boundary of the IRD layer [Manighetti et al., 1995;Trauth, 1995]. The same reasoning predicts an artificially high age for the bottom of the IRD layer due to lack of young tests being mixed down once rapid deposition of the IRD layer started. This is not observed as sample KIA 0007 at 147 cm shows the same age as the IRD layer. For our age model we use the average value of 13.54+_.07 ka at 136 cm ( Figure 3b). The three dates from the heart of the IRD layer (265-279 cm, Figure 3b) give an age of 20.46_+0.12 ka at 272 cm depth for this episode of rapid accumulation. Again, the two overlying dates are younger, though the age difference to sample KIA 0009 at 20.16_+.2 ka is statistically insignificant. The average of 20.46_+0.12 ka (KIA 0011-0013, Table 1)for IRD layer 2 together with the other three dates is used in the age model ( Table 2). The age model was calculated using a polynomial fit through the InC-AMS data (Figure 4). According to this model, contention that sediment flux from icebergs was reduced off Portugal as the site is outside the region of maximum iceberg flow.
Using the polynomially fitted age model, total duration of HI and H2 is 300 and 400 years, respectively. These estimates depend on how good the age estimates for the top and bottom of the Heinrich layers are. Rapid sedimentation in conjunction with differential bioturbation likely make the top age too young and the bottom age too old. As we have discussed above, our detailed 14C-AMS dating shows some evidence for the first but not for the latter. Still, spacing of the 14C data across both IRD layers is not sufficient to estimate ages of IRD boundaries unambiguously. Using •4C-AMS data and age errors at face value, this would limit IRD 1 to 200-300 years, and IRD 2 to 300-600 years duration. These are still fairly wide ranges, but they fit to similar estimates derived from 230Thex   [Kudrass, 1973]. Our XRD scans show that primary IRD components are quartz, plagioclase, feldspar, and calcite ( Figure  5). Since the XRD measurements were performed on bulk samples, most of the carbonate signal is likely of biogenic origin, that is, calcareous foraminiferal tests and nannoplank- Depth in Core (cm) Figure 5. Clay mineral composition of IRD layers and sediments immediately above and below in core SO75-26KL Dolomite is present in small quantities in IRD layers 1 and 3 (equivalent to Heinrich events HI and H4), calcite is most likely of biogenic origin (foraminifera and nannoplankton).

Mineralogy of IRD Layers Off Portugal
ton. Dolomite has been detected in minor quantities in samples from IRD layers 1 and 3 (equivalent to H1 and H4; Figure  5). After treatment with acetic acid the carbonate peaks disappeared from the diffractograms, but the intensity peak of dolomite was still observed, confirming the presence of dolomite in both IRD layers. Thus sediment cores from outside the immediate area of maximum IRD deposition will provide the only chance to document thermohaline patterns during these meltwater events, provided that IRD deposition at these sites was small enough to ensure benthic foraminiferal abundances high enough for continuous benthic isotope measurements across the IRD layers. These sediment cores need to be taken from advection pathways that are linked to thermohaline overturn in the North Atlantic. Core SO75-26KL fulfills these requirements because (1) sedimentation rates are high, (2) IRD deposition was sufficiently low to leave enough benthic foraminifera for isotope analysis, and ( scenario, NACW temperature is the same as today but salinity is considerably increased by 1.9. To estimate the maximum possible contribution of MOW to middepth waters at the site of core SO75-26KL, the following boundary conditions apply: equilibrium õc of +4%o PDB has to be maintained to be consistent with the observed benthic õ180 value (corrected to the Uvigerina scale) in core SO75-26KL, and density of the ambient water mass at the core site must not exceed that of underlying UNADWLG M. Maximum density for this middepth water mass is defined in the TS-õc diagram as the intercept between the +4%o õc isoline and the 37.6 (•2) isopycnal of UNADWLGM at a TS value of 7.2ø/37.1 (Figure 6b). This point also defines the maximum contribution of MOW as it is closest along the +4 %o õc isoline to the TS coordinate of glacial MOW (Figure 6b).

Dolomite-bearing carbonate bedrock is widespread in the Laurentide domain, but small outcrops also exist in Ireland and
As is shown in Figure 6b, the cold and warm NACW scenarios both indicate a maximum possible contribution of 10% to the middepth water mass at the Portuguese margin. In the case of the cold NACW, maximum MOW contribution is entirely defined by mixing between MOW and NACW; TS values of the mixing product are 6ø/36.4. In the case of a warm NACW, MOW mixes with a water mass that consists of roughly equal parts of UNADW and NACW. TS values are 7.2ø/37.1, as defined by the intercept between the +4 %o õc isoline and the 37.6 (•2) isopycnal of UNADWLGM (Figure 6b).
Both scenarios imply that the contribution of MOW to middepth waters at the upper Portuguese margin was only 10% compared to 30% today. These numbers change slightly if we use a more negative freshwater õw value of-30%o (SMOW) to account for lower glacial precipitation temperatures in the North Atlantic region and a stronger contribution of glacial meltwater (Figure 6c). Using a õw value of-30%o (SMOW), the slope of the tic lines in the TS field is steeper and the +4%o õc fractionation line does not intersect the density isoline of UNADW. Thus the maximum possible MOW contribution is defined by the intercepts of the +4%o õc fractionation line with the mixing lines between MOW and warm or cold NACW.

Maximum
MOW contributions are 5% if it mixes with cold NACW and 15% if it mixes with warm NACW (Figure 6c). That is, the contribution of glacial MOW to the middepth North Atlantic would have been reduced by 50% to more than 80% of its present contribution. Sea level lowering of as much as 120 m during the last glacial would reduce the geometry of the Strait of Gibraltar thus reducing the through flow of MOW to the North Atlantic [Bryden and Storereel, 1984]. Therefore it appears plausible to assume that MOWLG M contributed less than today to the hydrography at the upper Portuguese margin. This contention is further supported by physical considerations that TS values of 8ø/40.3 estimated for MOWLGM yield a density of 40 (•2) which is considerably higher than that of 37.6 for UNADWLGM. The density contrast between MOW and underlying deep waters is thus increased from 0.7 today (Figure 6a) to 2.4 at the LGM (Figures 6b and 6c). If MOW still contributed to the hydrography of the shallow North Atlantic, significant mixing with less saline North Atlantic waters was required to lower MOW density and to allow it to flow at shallow depths. The combination of enhanced density, which requires intensive mixing with less saline Atlantic waters to increase its buoyancy, and reduced volume of MOW therefore supports the conclusion that MOW played a less significant role in the glacial maximum North Atlantic. Decreases in benthic •13C of 0.7-1.1%o that parallel the occurrence of IRD layers in core SO75-26KL document strongly reduced water mass ventilation at the upper Portuguese margin during Heinrich events H1, H2, and H4 (Figures 4d and 7a) indicate that thermohaline reduction started well before the events and lasted much longer than the events. Inferred gradual spin-up of thermohaline circulation over periods of 1-3 kyr after all three Heinrich events is in direct contrast to sudden warmings seen in the GRIP data.

On the basis of elevated benthic õ13C levels during the last
Benthic •j13C changes during H2 are more rapid (Figure 7). The 1513C decline prior to and during H2 is less severe, that is, ' 0.8%0 compared to 1.0-1.1%o during HI and H4. It thus seems that the reduction in thermohaline overturn was less intense during H2, because glacial meltwater flux was either less or not continuously targeted at the site of convection because of variable surface circulation. Apparently, the North Atlantic's conveyor circulation was less inert during H2, allowing thermohaline overturn to respond rapidly to changes in surface forcing.
Whether the inferred changes in thermohaline overturn were caused by gradual increases and decreases of glacial meltwater flux as the Laurentide ice sheet grew, collapsed, and later sta•bilized, or by changes in surface ocean and/or atmospheric circulation remains speculative. The changes in benthic fi13C that were associated with the Heinrich events were not monotonous but show subtle maxima and minima (Figure 7). This suggests oscillatory behavior of thermohaline circulation which is also indicated by numerical models that link convective instabilities to changes in surface ocean forcing [Weaver and Hughes, 1994;Rahrnstorf, 1994Rahrnstorf, , 1995. These models also predict convective discontinuities during which thermohaline overturn abruptly jumps to minimum rates as freshwater forcing exceeds critical threshold values. The abrupt depletion of benthic fil3C that is