Subsurface North Atlantic warming as a trigger of rapid cooling events: evidence from the early Pleistocene (MIS 31–19)

. Subsurface water column dynamics in the subpo-lar North Atlantic were reconstructed in order to improve the understanding of the cause of abrupt ice-rafted detritus (IRD) events during cold periods of the early Pleistocene. We used paired Mg / Ca and δ 18 O measurements of Neogloboquad-rina pachyderma (sinistral – sin.), deep-dwelling planktonic foraminifera, to estimate the subsurface temperatures and seawater δ 18 O from a sediment core from Gardar Drift, in the subpolar North Atlantic. Carbon isotopes of benthic and planktonic foraminifera from the same site provide information about the ventilation and water column nutrient gradient. Mg /


Introduction
Rapid climate events in marine and continental sediments, as well as ice-core records, are a pervasive feature during the last glacial period (Dansgaard et al., 1993;Heinrich, 1988).Millennial-scale oscillations (Dansgaard-Oeschger (D-O) and Heinrich events) are characterized by abrupt shifts between warm and cold conditions, associated with icesheet oscillations, as evidenced by major ice-rafting events recorded in North Atlantic sediments (Grousset et al., 2001;Heinrich, 1988).The mechanism responsible for these fluctuations is not fully understood.Most accepted hypotheses relate rapid oscillations in the Atlantic Meridional Overturning Circulation (AMOC) to insulating effect of extensive iceshelves and sea-ice on the air-sea fluxes and/or to freshwater perturbations causing changes in the heat and salinity transport to the high-latitude northern North Atlantic (Ganopolski and Rahmstorf, 2001;Clark et al., 2001;Hátún et al., 2005;Li et al., 2005).
More recently, a number of studies have proposed that increased iceberg discharge during cold stadial events may have resulted from the destabilization of marine ice-shelves by a basal melting caused, in turn, by enhanced subsurface oceanic warming (Alvarez-Solas et al., 2010;Rasmussen and Thomsen, 2004;Marcott et al., 2011;Moros et al., 2002;Peck et al., 2008;Jonkers et al., 2010b;Ezat et al., 2014;Naafs et al., 2013).Model simulations indicate that weakening of deep convection at high latitudes in the North Atlantic results in a slow warming of intermediate depths (above Published by Copernicus Publications on behalf of the European Geosciences Union.2500 m) by downward diffusion of heat at low latitudes (Rühlemann et al., 2004).This heat is accumulated at the subsurface, and wind-induced circulation enables northward transport of warm and salty waters in the northern North Atlantic (Shaffer et al., 2004;Mignot et al., 2007;Liu et al., 2009).This mechanism involves the coupling of the AMOC with ice-sheet dynamics through an increase of the heat and salt export from low latitudes, and the warming of subsurface waters that would have acted as a positive feedback in ice-shelf collapse.General agreement between model and proxy evidence supports this explanation for abrupt climate shifts such as those associated with Heinrich and D-O events.
Application of Mg / Ca palaeothermometry to deepdwelling planktonic foraminiferal species constitutes a potential recorder of subsurface conditions (Kozdon et al., 2009;Simstich et al., 2003;Volkmann and Mensch, 2001) for testing the feasibility of this hypothesis.Moreover, as foraminiferal δ 18 O is controlled by temperature and seawater δ 18 O (δ 18 O sw ), combining foraminiferal Mg / Ca temperature reconstructions with δ 18 O from the same species and samples allow one to reconstruct δ 18 O sw as a proxy for salinity (Schmidt et al., 2004).Several palaeoceanographic studies using paired δ 18 O and Mg / Ca measurements have been produced for the Marine Isotope Stage (MIS) 3 (Jonkers et al., 2010b;Peck et al., 2006Peck et al., 2008)).However, studies with a similar approach are still required to understand subsurface temperature and circulation changes linked to AMOC reorganization during older time intervals, such as the early Pleistocene.Although the palaeoceanographic community has extensively studied climate disruptions during the most recent time periods, relatively little attention has been devoted to high-frequency climate variability in earlier periods when large Northern Hemisphere (NH) ice-sheets were the same size as in the ate Pleistocene.Part of the gap of the study of these rapid climate oscillations in older time periods was due to the absence of high-resolution palaeoclimate records.However, during recent years, several studies carried out on International Ocean Discovery Program (formerly the Integrated Ocean Drilling Program) cores have found robust evidence of abrupt climate events (Bolton et al., 2010;Ferretti et al., 2010;Kleiven et al., 2011;Hernández-Almeida et al., 2012;Bartoli et al., 2006), with similar structure during transitions between cold (stadial) and warm (interstadial) phases of the D-O cycles as those found during the last glacial period.
To further constrain the relationship between subsurface ocean temperature and ice-sheet instabilities during the early Pleistocene, we present here a new millennial-scale reconstruction of the temperature and δ 18 O sw of the subsurface Atlantic inflow using paired Mg / Ca and δ 18 O measurements of the planktonic foraminifera Neogloboquadrina pachyderma (sinistral -sin.) from IODP Site U1314.This is the first Mg / Ca temperature record produced in the subpolar North Atlantic for the early Pleistocene.Previous palaeosea surface temperature (SST) records in the region are de-rived from planktonic foraminifera-based transfer functions (Hernández-Almeida et al., 2012) or alkenones (McClymont et al., 2008), but none of them give information about the thermocline conditions.The location of this core is at an ideal latitude for monitoring changes in ice-sheet mass balance, and Mg / Ca values derived from N. pachyderma (sin.)allow one to record changes in the subsurface temperatures (upper thermocline, ∼ 200 m depth; Nürnberg, 1995) associated with oscillations in the AMOC.Our data suggest subsurface warming and salinity increases prior to and during the iceberg events, providing evidence of coupling between basal melting and ice-sheet collapse as a mechanism controlling the millennial-scale events in the early Pleistocene.

Study site and materials
Records were made using sediments from IODP Site U1314 (56.36 • N, 27.88 • W, 2820 m depth) from the southern Gardar Drift in the subpolar North Atlantic (Fig. 1a).Sedimentation rates average 9.3 cm ka −1 from 1069 to 779 ka, dated by tuning our benthic δ 18 O curve to the benthic isotope stack of Lisiecki and Raymo (2005, hereinafter referred to as LR04) by using AnalySeries software (Paillard and Yiou, 1996;see Hernández-Almeida et al., 2012 for further details).
Site U1314 lies in the path of an extension of the North Atlantic Current (NAC), the Irminger Current (IC), which splits from the NAC and turns toward the Greenland coast.The core of this relatively warm and salty water mass is distinguishable by its properties vertically down to 700 m depth.As the IC travels westwards, it mixes with the colder and fresher waters of the East Greenland Current (EGC), becoming less saline and colder (Malmberg, 1985).
Although today the limit of winter sea-ice (Arctic Front) lies north of Site U1314, it is known to have migrated southward during glacials of the Pleistocene, bringing much cooler waters and potentially also sea-ice south of 60 • N (Ruddiman, 1977).Today, modern hydrographic conditions at Site U1314 are characterized by seasonal water temperatures ranging between 7.7 and 11.7 • C at 10 m depth and 7.4 and 8 • C at 200 m (Locarnini et al., 2013) with nearly constant salinity of 35.1-35.2practical salinity units (PSU, Antonov et al., 2006;Fig. 1b).
Winter convection of the cooled Atlantic surface waters in the Nordic seas results in the formation of North Atlantic Deep Water (NADW), which flows southward as the Iceland-Scotland Overflow Water (ISOW; Fig. 1a).This water mass flows at Site U1314 depth (Bianchi and McCave, 2000).
Subsurface water column conditions were determined through Mg / Ca ratios and stable isotopes measured on deep-dwelling planktonic foraminifera N. pachyderma (sin.).This species inhabits and calcifies its shell in the subpolar North Atlantic at the upper thermocline, at ∼ 200 m depth (Kohfeld et al., 1996;Simstich et al., 2003;Nürnberg, 1995; (Locarnini et al., 2013;Zweng et al., 2013).Map generated with Ocean Data View v.3.4.3.software (Schlitzer, 2008).Volkmann and Mensch, 2001).Therefore we assume that δ 13 C on deep-dwelling foraminifera N. pachyderma (sin.)provides information on the ventilation rates of the subsurface water mass (Hillaire-Marcel et al., 2011), while Mg / Ca measurements of the same species reflect water temperature changes and combined with δ 18 O provides a record of seawater δ 18 O (sw) of the subsurface ocean (Peck et al., 2006).
Around 50-60 well-preserved tests of planktonic foraminifera N. pachyderma (sin.)(> 150 µm size fraction, non-encrusted tests) were analysed in 542 samples for Mg / Ca ratio following the procedure of Pena et al. (2005) which includes a reductive cleaning step.Dissolved samples were analysed using a Perkin Elmer Elan 6000 inductively coupled plasma mass spectrometer (ICP-MS) at the Scientific and Technological Centers of the University of Barcelona (CCiT-UB).External reproducibility for the Mg / Ca ratio is estimated at 1.8 % (2σ ) based on the analysis of high-purity gravimetrically prepared standard solution (1.629 mmol mol −1 ) measured routinely every four samples.Elemental ratios of Mn / Ca and Al / Ca ratios were analysed in parallel as quality controls for clay and Mn-rich mineral content.The recorded low values (Mn / Ca < 0.5 mmol mol −1 ; Al / Ca < 0.15 mmol mol −1 ) and their low correlation with the Mg / Ca ratios (R 2 = 0.2 and 0.004 respectively) indicate that the cleaning protocol satisfactorily removed most of the contaminant phases.Final Mg / Ca values were converted into temperatures values according to the equation of Elderfield and Ganssen (2000) (Mg / Ca Temp; Table S1 in the Supplement).
Stable isotopes (carbon and oxygen; Table S1 in the Supplement) records from benthic and planktonic foraminifera correspond to Hernández-Almeida et al. (2013a, b, 2012).Analyses were carried out on planktonic foraminifera N. pachyderma (sin.) and on benthic foraminifera Cibicidoides spp.(mainly Cibicidoides wuellerstorfi) and Melonis pompilioides when the former was absent.An adjustment factor (−0.11 ‰ for δ 18 O and +0.6 ‰ for δ 13 C) calculated from replicates along the core was then applied to the M. pompilioides isotope values to produce a uniform isotope data set.Oxygen isotope values were then ice-volume corrected by scaling to the sea-level curve of LR04 using an last glacial maximum to late Holocene sea-level change of 120 m (Bintanja and van de Wal, 2008).Seawater δ 18 O was calculated introducing paired Mg / Ca-based temperatures and calcite δ 18 O from N. pachyderma (sin.) in the palaeotemperature equation of Shackleton (1974; δ 18 O sw ; Table S1 in the Supplement).It has been widely demonstrated that planktonic species do not always precipitate calcite in equilibrium.Based on the δ 18 O measurements of seawater and N. pachyderma (sin.)tests from the Icelandic continental shelf, Smith et al. (2005) observed a δ 18 O disequilibrium offset of 0.25 ‰.Other authors have also observed a disequilibrium offset in the oxygen isotope composition of N. pachyderma (sin.) of ∼ 0.6 ‰ associated with post-gametogenic processes and thermal stratification of the water column in the Nordic seas (Nyland et al., 2006).However, Jonkers et al. (2010a) did not find any offset in sediment trap samples from the Irminger Sea.Taking into account that samples used in this study are very close to Site U1314, we did not apply

Results
Mg / Ca ranges between 0.7 and 1.25 mmol mol −1 , and Mg / Ca-derived palaeotemperatures range between 1.9 and 12.3 • C (Fig. 2).The Mg / Ca and δ 18 O sw records show different patterns after and before MIS 25.From MIS 31 to MIS 25, the amplitude of the glacial-to-interglacial (G-IG) changes is low; temperatures and δ 18 O sw are stable, only punctuated by frequent millennial-scale oscillations, with temperature decreases of ∼ 3 • C and δ 18 O sw increases up to 1 ‰.After MIS 25, the amplitude of hydrographic changes was larger, with δ 18 O sw increased by ∼ 0.5-1 ‰, and temperature reaching maxima up to 12 • C between MIS 25 and 21.During this interval, there is also a pervasive suborbital variability, especially during glacial onset and during MIS 21.Ice-rafting episodes are characterized by relatively warm and saltier subsurface waters at the Gardar Drift.Rapid temperature and δ 18 O sw increases are observed before the ice-rafted detritus (IRD) deposition, e.g. at 1060, 995, 924, 880 ka, or shortly after the iceberg discharge started (Fig. 3).
There are exceptions, and some events do not show this pattern, like at ∼ 832 and 828 ka, subsurface warming is not observed, but there is increase in δ 18 O sw (Table 1).
The most important feature of the difference between benthic and planktonic δ 13 C ( δ 13 C) are the abrupt decreases of ∼ 1 ‰ during IRD events, when values are around 0 ‰.During warmer periods, δ 13 C ranges between +1 and 1.4 ‰ (Fig. 4).

Discussion
Palaeotemperature estimates based on Mg / Ca of N. pachyderma (sin.) at Site U1314 indicate that many of the IRD events were characterized by an abrupt subsurface warming (Fig. 2).The magnitude of this warming is not always the same across the studied interval, ranging between 2.5 and 8 • C. The δ 18 O sw shows repeatedly higher values, indicating saltier waters during IRD deposition.Although these changes in temperature and salinity were simultaneous with the IRD events, in some cases (e.g. at 995 ka), subsurface waters started to warm up and become saltier even before the icerafting.The positive excursions of the δ 13 C signal from N. pachyderma (sin.) during these events were interpreted to indicate increasing subsurface ventilation in the North Atlantic (Hernández-Almeida et al., 2013b; Fig. 2).Similar conditions of better ventilation at intermediate depths during IRD deposition are also evident from benthic δ 13 C in Site 982 on the Rockall Plateau (Venz et al., 1999), which was suggested to be related to changes in the production of Glacial North Atlantic Intermediate Water (GNAIW) (Fig. 2).Strong coupling between the Mg / Ca temperatures and δ 18 O sw fluctuations and subsurface circulation may reflect a change in the AMOC.
The accumulation of subsurface warming during icerafting events would have corresponded with a rapid devel- opment of the thermocline that stabilizes the water column and via intense basal melting and thinning of marine iceshelves provokes a large-scale instability of the ice-sheets and retreat of the grounding line.With the destruction of iceshelves, ice streams can surge, leading to increased iceberg production.The ice-sheets located in regions with relatively mild conditions and high precipitation rates, such as Scandinavia and Iceland, are indeed very sensitive to millennial climate variability, and then respond quickly to warmer conditions producing iceberg discharges (Marshall and Koutnik, 2006).The difference between benthic and planktonic δ 13 C ( δ 13 C), used to indicate the nutrient gradient between subsurface and bottom water (Charles et al., 2010), gives additional information about the ventilation of subsurface and deep waters.The short-term periods of low δ 13 C values (∼ 0 ‰) during IRD discharges suggest water column vertical mixing and formation of the GNAIW south of the Arctic Front.
After iceberg calving decreased, the sudden release of heat accumulated at the subsurface and broke the upper stratification (Mignot et al., 2007).Inflowing warm and salty Atlantic waters were again in contact with the surface ocean, and there was an efficient release of heat to the atmosphere, resulting in an intensified AMOC characterized by deeper and stronger deep-water circulation (Schmidt et al., 2006;Liu et al., 2009).Onset of deep convection in the Nordic seas and NADW production led to a shutdown of GNAIW production (Venz et al., 1999).The nutrient gradient profile shows rapid increases up to 1.4 ‰ reflecting the establishment of a strong nutricline between deep and subsurface waters (Fig. 4).The switch to deep convection and a strong AMOC overshooting caused a decrease in subsurface temperatures and δ 18 O sw , suggesting the return toward a "normal water column" state.
Although the mechanism that characterizes the subsurface climate instabilities involves higher Mg / Ca temperatures and planktonic δ 13 C and δ 18 O sw , some of the events are miss- ing some of these features.At ∼ 832 and 828 ka, IRD events are not clearly accompanied by subsurface warming, while changes in δ 13 C and δ 18 O sw are evident (Table 1).This could imply that the more active subsurface depth ventilation was due to brine rejection during the wintertime sea-ice production, as occurs in high-latitude seas (Aagaard and Carmack, 1989;Horikawa et al., 2010).However, this alternate mechanism to explain the eventual higher density of subsurface waters in absence of warmer waters is speculative, and more robust evidence of brine rejection during sea-ice formation are needed.We are still uncertain about the driving mechanism that drives northward transport of warm and salty subsurface waters during episodes of weak AMOC.We suggest that analogous mechanisms involving ice-shelf and sea-ice expansion in the NH that are invoked to explain D-O cycles during the last glacial period (Petersen et al., 2013), also operated during the early Pleistocene.Growing ice-shelves in the subpolar North Atlantic during the onset of glaciations would have changed land surface albedos producing a reduction of air sea temperature (Broccoli and Manabe, 1987).This cooling would have increased the extent and thickness of sea-ice, resulting in a higher insulation of the surface ocean (Li et al., 2005;Kaspi et al., 2004), causing convection shutdown in the high-latitude North Atlantic and reduced NADW formation.A weakened subpolar gyre circulation would have supplied less cold and less fresh water to the Atlantic inflow to the Nordic seas, making it saltier (Thornalley et al., 2009;Hátún et al., 2005).Warm and salty waters accumulating at the subsurface would have been eventually transported poleward as there was still convection, although at intermediate depths, and finally this would have caused a temperature inversion and higher salinity at subsurface depths in the subpolar North Atlantic (Shaffer et al., 2004).Alternatively, abrupt slowdown of the AMOC may respond to different mechanisms including internal oscillation regulated via atmospheric CO 2 concentration and Southern Ocean wind intensifications (Banderas et al., 2012;Alvarez-Solas et al., 2011).
Several modelling and palaeoclimate studies also show intermediate or subsurface warming in the North Atlantic during IRD events as a response to AMOC reorganization (Liu et al., 2009;Mignot et al., 2007;Brady and Otto-Bliesner, 2011), accompanied by a southward shift in the convection cell from the Nordic seas to the subpolar North Atlantic (Brady and Otto-Bliesner, 2011;Venz et al., 1999;Voelker et al., 2010;Oppo and Lehman, 1993).This scenario characterized by a temperature inversion represents an analogous situation to modern conditions in Arctic Ocean.In this region, Atlantic waters flowing via the West Spitsbergen Current cause an Atlantic-derived temperature and salinity maximum at 200-500 m water depth, under the permanent sea-ice cover (Bauch et al., 1997).
Temperature sensitive proxies from other North Atlantic sites display similar features that are interpreted as subsurface warming conditions prior to ice-rafting events and deglaciations during the last glacial period and the Holocene.Risebrobakken et al. (2011) documented intensified subsurface warming in the Nordic seas using planktonic foraminifera fauna as a response to the reduced strength of the AMOC through the deglaciation and the early Holocene.Mg / Ca-derived temperatures from N. pachyderma (sin.) in two cores from the north-east Atlantic also support the inferred warming during Heinrich events.These records show upper ocean stratification and high subsurface temperatures initiated during ice-rafting events (Jonkers et al., 2010b;Peck et al., 2008).Jonkers et al. (2010b) explained the low planktonic δ 13 C values of N. pachyderma during these events as a result of reduced ventilation of subsurface waters due to the insulating effect of a meltwater lens and/or a sea-ice layer.Our high planktonic δ 13 C values during these rapid cooling events, however, indicate that more intense subsurface ventilation and/or nutrient depleted subtropical waters were exported to the subpolar North Atlantic, which is supported by the similarity with the intermediate water δ 13 C signal from Site 982 (Venz et al., 1999;Fig. 2).We argue that such disagreement between planktonic δ 13 C profiles could be explained by the southward shift of the polar front as far as 42 • N during cold periods of the Upper Pleistocene (Rud-  diman and McIntyre, 1981;Eynaud et al., 2009), limiting the fraction of nutrient depleted subtropical waters exported northward (Mix and Fairbanks, 1985) compared to the early Pleistocene.
Similar warm conditions during Heinrich events and stadials are also evident from benthic fauna and Mg / Ca ratios in benthic foraminifera from the Nordic seas, indicating that warming was probably extended to intermediate depths (below 1000 m) by downward diffusion of subtropical ocean heat during times of slow North Atlantic overturning (Rasmussen and Thomsen, 2004;Marcott et al., 2011;Ezat et al., 2014).These results are in agreement with subsurface warming events at the subtropics during Heinrich 1 (Schmidt et al., 2012).All of these observations suggest that subsurface warming was a basin-wide phenomenon during periods of reduced AMOC in MIS 3. To better constrain this scenario for the early Pleistocene, more subsurface marine records situated in key regions from the North Atlantic are required.The proposed scenario is in agreement with modelling studies that reveal basal melting of the ice-shelf and periodic pulses of iceberg discharge as a response to strong reduction of the AMOC (Mignot et al., 2007;Shaffer et al., 2004;Alvarez-Solas et al., 2010;Manabe and Stouffer, 1997).
Finally, from the similarity of the palaeoclimatic records with the model simulations and modern observations, we argue that observed increased subsurface ocean warming could play a leading role in the massive break-up of ice-shelves in the Antarctic Ocean (Vaughan and Doake, 1996;Rignot and Jacobs, 2002;MacAyeal et al., 2003).

Conclusions
The Mg / Ca-derived palaeotemperature and δ 18 O sw oscillations prior to and during IRD discharges at Site U1314 across the early Pleistocene  are related to changes in subsurface circulation.The mechanism operating during episodes of rapid climate cooling consists of a reduction in the AMOC during periods of extensive ice-shelves and sea-ice in the subpolar North Atlantic.Deep water convection sites shifted south of the polar front and production of GNAIW increased at the expense of NADW.Poleward transport of warm and salty subsurface subtropical waters during these episodes thinned and destabilized ice-shelves creating pulses of iceberg discharge.Heat accumulated at the subsurface was suddenly released to the atmosphere when the ice-sheet collapsed, resulting in an intensified AMOC.Analogous mechanisms based on subsurface warming as a trigger for millennial-scale climate variability were proposed for Heinrich events or D-O cycles recorded during the late glacial period (Alvarez-Solas et al., 2010;Shaffer et al., 2004), reflecting that rapid switches of the AMOC also occurred during the early Pleistocene.
The Supplement related to this article is available online at doi:10.5194/cp-11-687-2015-supplement.

Table 1 .
Summary of the changes in Mg / Ca, δ 18 O sw and planktonic δ 13 C during the IRD events.The amplitude of the change is calculated from the difference between the point where δ 18 O sw starts to increase prior to the IRD event and the δ 18 O sw maxima during the IRD event.The events are colour-coded, deep red showing the strongest change, and white the weakest.
sw .Contradicting studies indicate that this issue is not well constrained, with a need for further studies.Due to the uncertainties in N. pachyderma (sin.)"vital effect" and low SST during the mid-Pleistocene transition that may overestimate the δ 18 O sw values, we suggest caution when interpreting in absolute terms.