Glacial  13 C decreases in the western South Atlantic forced by millennial changes in Southern Ocean

10 Abstract. Abrupt millennial – scale climate change events of the last deglaciation (i.e., Heinrich Stadial 1 and the Younger Dryas) were accompanied by marked increases in atmospheric CO 2 presumably originated by outgassing from the Southern Ocean. However, information on the preceding Heinrich Stadials during the last glacial period is scarce. Here we present stable carbon isotopic data (  13 C) from two species of planktonic foraminifera from the western South Atlantic that reveal major decreases (up to 1‰) during Heinrich Stadials 3 and 2. These  13 C decreases are most likely related to millennial – 15 scale periods of intensification in Southern Ocean deep water ventilation presumably associated with a weak Atlantic meridional overturning circulation. After reaching the upper water column of the Southern Ocean, the  13 C depletion would be transferred equatorward via central and thermocline waters. Together with other lines of evidence, our data are consistent with the hypothesis that the CO 2 added to the atmosphere during abrupt millennial – scale climate change events during the last glacial period also originated in the ocean and reached the atmosphere by outgassing from the Southern Ocean. The 20 temporal evolution of  13 C during Heinrich Stadials in our records is characterized by two relative minima separated by a relative maximum. This “w – structure” is also found in North Atlantic and South American records, giving us confidence that such structure is a pervasive feature of Heinrich Stadial 2 and,

the Brazil Basin (Stramma et al., 1990;Peterson and Stramma, 1991). Around 38° S the BC encounters the northward flowing Malvinas Current (MC) (i.e., Brazil/Malvinas Confluence), where the opposing flows turn south-east and flow offshore. The offshore region is characterized by intense mesoscale variability. After collision and considerable mixing the warm-salty BC fractions flow eastward as the South Atlantic Current (Olson et al., 1988;Peterson and Stramma, 1991), while the majority of the cold fresh MC waters veer southeastward to rejoin the Antarctic Circumpolar Current. 5 The BC transports Tropical Water (TW) and South Atlantic Central Water (SACW). TW occupies the mixed layer, i.e., the upper ca. 100 m of the water column, with a mean temperature of 20 °C and mean salinity of 36 psu (Tsuchiya et al., 1994).
TW originates in the tropics-subtropics transition region by subduction, creating a subsurface salinity maximum capping the central waters (Memery et al., 2000;Tomczak and Godfrey, 2003) (Fig. 1). SACW occupies the permanent thermocline from ca. 100 to 500 m water depth. Its temperature ranges from 6 to 20 °C and its salinity spans from 34.6 to 36 psu (Memery et 10 al., 2000). Two types of SACW have been identified (Stramma et al., 2003). The low-density type of SACW which is mainly found in the South Atlantic subtropical gyre is formed by subduction of a low-density type of Subantarctic Mode Water (SAMW) along the southern edge of the gyre (Gordon, 1981;Stramma and England, 1999). A denser variety of SACW originates in the South Indian Ocean and is brought into the South Atlantic by the Agulhas Current (Sprintall and Tomczak, 1993) (Fig. 1). 15 Just below the permanent thermocline, Antarctic Intermediate Water (AAIW) occupies the water column from ca. 500 to 1200 m water depth (Stramma and England, 1999). AAIW is characterized as a cold and low salinity water mass (Piola and Georgi, 1982;Tomczak and Godfrey, 2003). Around the southern tip of South America, AAIW originates by subduction of cold and fresh Antarctic Surface Water across the Antarctic Polar Front, and by contribution of a dense type of SAMW that originates from deep winter convection in the Subantarctic Zone (Molinelli, 1981;Naveira Garabato et al., 2009). AAIW is 20 advected eastward through the Drake Passage by the Antarctic Circumpolar Current and turns northwards with the MC into the South Atlantic (Piola and Gordon, 1989). Since AAIW circulation follows the anticyclonic flow of the subtropical gyre the majority of the northward flow occurs in the eastern basin (McCartney, 1977;Stramma and England, 1999;Tomczak and Godfrey, 2003). However, intense mixing in the Brazil/Malvinas Confluence also leads to direct northward influence in the western South Atlantic that can, to some extent , influences the formation region of SACW (e.g., Piola and Georgi, 1982) 25 ( Fig. 1 and 2).
In the modern South Atlantic, the distribution of dissolved inorganic carbon  13 C ( 13 C DIC ) allows the identification of its major water masses. TW and SACW show high  13 C DIC values of ca. 2‰. AAIW presents  13 C DIC values of ca. 0.7‰. NADW derives from the North Atlantic and shows  13 C DIC values of ca. 1‰. The NADW layer is surrounded by Upper and Lower CDW which present  13 C DIC values of ca. 0.4‰ (Kroopnick, 1985). Since planktonic foraminiferal  13 C reflects the 30  13 C DIC of the ambient seawater, we use it as a proxy for the past oceanic carbon system Spero, 1992).
Changes in upper ocean properties and circulation patterns are also closely associated with changes in the atmospheric circulation. Positive sea surface temperature (SST) anomalies in the western South Atlantic, likely associated to changes in Clim. Past Discuss., doi:10.5194/cp-2016-59, 2016 Manuscript under review for journal Clim. Past Published: 20 June 2016 c Author(s) 2016. CC-BY 3.0 License. the strength of the AMOC (Knight et al., 2005), have been correlated with positive anomalies in the strength of the SAMS and, consequently, with the increase of precipitation over SESA (Chaves and Nobre, 2004). The SAMS and its main componentsthe ITCZ, the South Atlantic Convergence Zone (SACZ), and the South American Low Level Jet (SALLJ)are the main atmospheric drivers of the hydroclimate of tropical and subtropical SESA to the east of the Andes (Garreaud et al., 2009). The ITCZ is a global convective belt in the equatorial region, and the SACZ is an elongate NW-SE convective 5 belt that originates in the Amazon Basin and extends southeastward above the northern portion of SESA and the adjacent subtropical South Atlantic. The SALLJ is a NW-SE humidity flux from the west Amazon Basin to the subtropical region of SESA (Zhou and Lau, 1998;Carvalho et al., 2004;Schneider et al., 2014). This southward water vapour flux is a crucial source of precipitation to the Plata River drainage basin (Berbery and Barros, 2002), which is a source of continental borne sediments to our core site. 10

Marine sediment core
We investigated sediment core GeoB6212-1 (32.41° S, 50.06° W, 1010 m water depth, 790 cm core length) (Schulz et al., 2001) collected from the continental slope off SEAS where the upper water column is under the influence of the BC, and thus the TW and SACW (Fig. 1). This gravity core was raised at the Rio Grande Cone, a major sedimentary feature in the 15 western Argentine Basin. As our focuses here are HS3 and HS2, we analysed a section from the bottom of the core (768 cm core depth; ca. 33 cal ka BP) up to 290 cm core depth (ca. 20 cal ka BP). Visual core inspection provided evidence for the presence of sand lenses at 330 and 368 cm core depth (Schulz et al., 2001;Wefer et al., 2001). Therefore we did not sample these depths. The section of interest of GeoB6212-1 was sampled every 2.5 cm with syringes of 10 cm 3 . All samples were wet sieved, oven-dried at 50 °C and the fraction larger than 150 m was stored in glass vials for subsequent analyses. 20

Age model
The age model of core GeoB6212-1 is based on 14 AMS radiocarbon ages from planktonic foraminifera (Table 1, Fig. 3).
For each sample, we hand-picked under a binocular microscope around 10 mg of planktonic foraminifera shells from the sediment fraction larger than 150 m. Samples were analysed at the Poznan Radiocarbon Laboratory, Poland, and at the Beta Analytic Radiocarbon Dating Laboratory, USA (Table 1). All radiocarbon ages were calibrated with the calibration curve 25 IntCal13 (Reimer et al., 2013) with the software Bacon 2.2 (Blaauw and Christen, 2011). A marine reservoir correction of 400 years was applied (Bard, 1988). All ages are reported as calibrated years before present (cal a BP; present is 1950 AD).
To construct the age model we used Bayesian statistics in the software Bacon 2.2 (Blaauw and Christen, 2011). Default parameter settings were used, except for mem.mean (set to 0.4) and acc.shape (set to 0.5). Ages are modelled as drawn from Clim. Past Discuss., doi: 10.5194/cp-2016-59, 2016 Manuscript under review for journal Clim. Past Published: 20 June 2016 c Author(s) 2016. CC-BY 3.0 License. a t-distribution, with 9 degrees of freedom (t.a=9, t.b=10). 1,000 age-depth realizations were used to estimate mean age and 95 % confidence intervals at 0.5 cm resolution (Fig. 3).

Stable carbon isotope analyses
Around 10 tests of G. ruber white sensu stricto (Wang, 2000) within the size range 250-350 m and 8 tests of G. inflata non-encrusted with 3 chamber in the final whorl (Groeneveld and Chiessi, 2011) within the size range 315-400 m were 5 hand-picked under a binocular microscope every 2.5 cm from 290 to 768 cm core depth. While the first species records the conditions at the top of the mixed layer (down to ca. 30 m) (Chiessi et al., 2007;Wang, 2000), the second species records the conditions at the permanent thermocline (ca. 350-400 m) (Groeneveld and Chiessi, 2011), allowing the reconstruction of the  13 C signal of the TW and the SACW, respectively. The  13 C analyses were performed on a Finnigan MAT 252 mass spectrometer equipped with an automatic carbonate preparation device at the MARUM -Centre for Marine Environmental 10 Sciences, University of Bremen, Germany. Isotopic results are reported in the usual delta-notation relative to the Vienna Peedee belemnite (VPDB). Data were calibrated against the house standard (Solnhofen limestone), itself calibrated against the NBS19 standard. The standard deviation of the laboratory standard was lower than 0.05‰ for the measuring period.

Age model 15
Our age model covers the period between 32.6 and 5.7 cal ka BP (Table 1, Fig. 3). Sedimentation rates change markedly during this time interval with values ranging from 3.8 to 111 cm ka -1 . Three main peaks in sedimentation rate were identified at ca. 26, 23 and 15 and one minor peak at 11 cal ka BP. The two oldest sedimentation peaks occur within our period of interest (i.e., from ca. 33 until 20 cal ka BP), and received special attention due to the higher sedimentation rate which provides increased temporal resolution (Fig. 3). The mean temporal resolution of our  13 C records is ca. 90 yr with values 20 ranging from 28 and 195 yrs.

Stable carbon isotopes analyses
The G. ruber  13 C record shows two long-term decreases, from ca. 32.6 to 28.5 cal ka BP with amplitude of ca. 1‰ and from ca. 26.5 to 24.8 cal ka BP also with amplitude of ca. 1‰ (Fig. 4a). These two negative long-term trends are separated from each other by an abrupt increase of ca. 1.3‰ ending at ca. 27 cal ka BP. Both long-term decreasing slopes were 25 interrupted by brief positive excursions, one from 30.6 to 30.4 cal ka BP with amplitude of ca. 0.7‰ and other from ca. 26.2 to 25.8 cal ka BP with amplitude of ca. 1‰. After the second long-term decrease, the  13 C values of G. ruber varied around 0.7‰. Both long-term negative excursions determine a pattern we refer to as "w-structure". The G. inflata  13 C record shows four negative excursions departing from a baseline of ca. 0.8‰ (Fig. 4b). The first occurs from ca. 32.5 to 30.6 cal ka BP with an amplitude of ca. 0.5‰, the second from ca. 29.8 to 28.3 cal ka BP with the same amplitude, the third from ca. 26.5 to 26.4 cal ka BP with an amplitude of ca. 0.8‰, and the forth from ca. 25.8 to 24.4 cal ka BP with an amplitude of ca. 0.9‰. Also in the  13 C record from G. inflata two w-structures are present and are defined by the previously described negative excursions. 5 The w-structures for both species as well as the  13 C minima are synchronous (Fig. 4).

Discussion
The synchronous w-structures present in the  13 C records of both planktonic foraminiferal species analysed here occur in consonance with the millennial-scale events HS3 and HS2 (Sarnthein et al., 2001;Goni and Harrison, 2010) (Fig. 4). Based on modern conditions, we expect our core site not to be influenced by significant changes in the nutrient content of the upper 10 water column since the region is dominated by the oligotrophic BC, characteristic of western boundary currents and is far from upwelling cells (Brandini et al., 2000). Thus, it is unlikely that changes in our  13 C records are associated with local productive events driven by nutrient-cycle processes .
A pervasive feature of planktonic foraminiferal  13 C records in the Indo-Pacific Ocean (Spero and Lea, 2002), Southern Ocean (Ninnemann and Charles, 1997), and South Atlantic Ocean (Oppo and Fairbanks, 1989) is a negative excursion 15 during HS1. Ninnemann and Charles (1997) suggested that the source for this signal is in the Southern Ocean. They further proposed that the anomaly is related to the transfer of a preformed  13 C signal from the Southern Ocean via SAMW and/or AAIW. A low-density type of SAMW actually contributes to SACW that spreads into the South Atlantic (Stramma and England, 1999). Additionally, AAIW also influences SACW through vigorous eddy mixing at the Brazil/Malvinas Confluence (Piola and Georgi, 1982). Thus, SACW represents a potential conduit for the  13 C signal from the sub-Antarctic 20 region to the subtropical South Atlantic ( Fig. 1 and 2). Therefore, we propose that the negative excursions in our  13 C records are related to the transfer of a preformed  13 C signal from the subantarctic zone to the western South Atlantic via central and thermocline waters.
The reduced AMOC would decrease the sub-tropical heat transport towards the north, leading to rising temperatures in the circum-Antarctic region (EDML, 75° S, 0° E, EPICA) (EPICA Community Members, 2006) (Fig. 5i). Furthermore, during phases of weak AMOC the Southern Hemisphere westerlies are stronger and shift southward strengthening CDW upwelling 5 (Anderson et al., 2009;Denton et al., 2010). Increased upwelling would supply the surface of the Southern Ocean to the south of the Antarctic Polar Front with more low- 13 C waters and with a higher concentration of Si(OH) 4 (Anderson et al., 2009;Hendry et al., 2012). Since upwelled CDW is hypothesized to be the dominant source of the upper and intermediate waters that leave the Southern Ocean (i.e., SAMW and AAIW) (Fig. 2), increased upwelling would transfer the low  13 C signal as well as the positive Si(OH) 4 anomaly northward into the adjacent subtropical gyres (Oppo and Fairbanks, 1989;10 Ninnemann and Charles, 1997;Spero and Lea, 2002;Anderson et al., 2009;Hendry et al., 2012). These signals would then propagate through the thermocline (i.e., SACW) of the South Atlantic, and be transferred to the mixed layer by vertical exchange process (i.e., TW) (Tomczak and Godfrey, 2003). However, we cannot exclude the possibility that the upwelled low- 13 C respired CO 2 could have been first outgassed from the Southern Ocean, and then re-dissolved into the ocean via air-sea exchanges at the formation regions of SACW and TW, eventually reaching the upper water column at our core site. 15 Higher concentrations of Si(OH) 4 were described in benthic organisms at intermediate water depths (i.e., 1048 m water depths) of the western South Atlantic (ca. 27° S) close to our core site during abrupt millennial-scale climate change events (Hendry et al., 2012), suggesting that the preformed signal from the Southern Ocean indeed reached subtropical latitudes in the South Atlantic.
Millennial-scale changes of the Southern Ocean temperature and deep water ventilation also led to the increase in CO 2atm 20 (Spero and Lea, 2002;Ahn and Brook, 2008;Ahn and Brook, 2014;Gottschalk et al., 2015). During HS3 and HS2, positive  (Fig. 5j). However, the CO 2atm peaks occur ca. 1 ka later than the initiation of the  13 C decrease in our records. Spero and Lea (2002) also observed a similar offset between the increase in CO 2atm and the decrease in Pacific Ocean planktonic foraminifera  13 C during HS1, and attributed this apparent offset to uncertainties in the age models of their records. An alternative explanation relates to a possible time lag between the weakening of the AMOC and 30 the increase in CO 2atm (Ahn and Brook, 2014).
Thus, our records are consistent with the hypothesis that the increase in CO 2atm during abrupt millennial-scale climate change events of the last glacial period is originated by ocean processes (Smith et al., 1999;Ahn and Brook, 2008;Bereiter Clim. Past Discuss., doi:10.5194/cp-2016-59, 2016 Manuscript under review for journal Clim. Past  Ahn and Brook, 2014) and is most likely related to a weak AMOC and associated strengthened Southern Ocean upwelling.

Continental responses
Paleoclimate records from South America have shown marked hydrological changes during abrupt millennial-scale climate events (Arz et al., 1998;Peterson et al., 2000;Baker et al., 2001;Cruz et al., 2006;Stríkis et al., 2015). Reconstructions of 5 the SAMS activity suggest its strengthening during HSs (Cruz et al., 2006;Kanner et al., 2012;. Changes in speleothem oxygen isotopic composition from the western Amazon Basin (NAR-C, Cueva del Diamante cave, northern Peru, 5.4° S, 77.3° W)  as well as changes on gamma radiation records from the Bolivian Altiplano (Salar de Uyuni, 20.3° S, 67.5° W) (Baker et al., 2001) (Fig. 5f, g) suggest increased precipitation during HS3 and HS2. To the north of the equator, a reflectance record from the Cariaco Basin (off northern Venezuela, MD03-2621, 10.7° N, 65° W) 10 (Deplazes et al., 2013) suggests decreased precipitation during the same millennial-scale events (Fig. 5e). The opposite behaviour of these sites reflects the interhemispheric anti-phase response of tropical precipitation during HSs (Wang et al., 2007;. Importantly, during HS3 and particularly HS2 the three above mentioned records (Fig. 5e, f, g) show a w-structure similar to the one observed in our  13 C records. Stríkis et al. (2015) reported a similar w-structure during HS1 related to two distinct hydrologic phases within HS1. 15 Periods of intensified SAMS would have strengthened the discharge from the Plata River drainage basin (Chiessi et al., 2009), increasing the delivery of terrigenous sediments to the Rio Grande Cone (Lantzsch et al., 2014), our coring site. We show for the first time increased sedimentation rates during a HS off SESA. Thus, the increased sedimentation rates during HS2 in our records corroborate the suggestion of Chiessi et al. (2009). Furthermore, GeoB6212-1 sedimentation rates also show a w-structure during HS2 (Fig. 3), hinting for a sensitive response of the Plata River drainage basin to the increase in 20 activity of the SAMS.
The increased continental runoff that led to increased delivery of terrigenous sediments to our core site could have also enhanced the nutrient availability and the local primary productivity, affecting our planktonic foraminiferal  13 C records.
However, we discard this possibility because the expected signal of stronger local primary productivity on planktonic foraminiferal  13 C would be opposite to the one observed at GeoB6212-1 . 25 The occurrence of a similar w-structure in North Atlantic records, in South American records and in our  13 C and sedimentation rate records gives us confidence that such w-structure is indeed a feature of HS2, and possibly also HS3.

Conclusions
Our mixed layer and permanent thermocline  13 C records from the western South Atlantic show in-phase millennial-scale decreases of up to 1‰ during the HS3 and HS2. We hypothesize that the source of the low  13 C signal can be explained by 30 millennial-scale changes in the Southern Ocean deep water ventilation. A weak AMOC during HS3 and HS2 would produce stronger Southern Ocean upwelling that in turn, would supply the surface of the Southern Ocean with more low- 13 C waters as well as promote increased outgassing of this old low- 13 C respired CO 2 . The low- 13 C waters at the surface of the Southern Ocean would be subducted into the central and thermocline waters and transferred equatorward via the South Atlantic subtropical gyre circulation towards our core site. Together with other lines of evidence, our data are consistent with the hypothesis that the CO 2 added to the atmosphere during abrupt millennial-scale climate change events of the last glacial 5 period originated in the ocean and reached the atmosphere by outgassing of the Southern Ocean. Moreover, the occurrence of a similar w-structure during HS2 (and possibly HS3) in North Atlantic and South American records as well as in our planktonic foraminiferal  13 C and sedimentation rate records gives us confidence that such w-structure is a pervasive feature that characterizes HS2 (and possibly HS3).

Data availability 10
The data reported here will be archived in Pangaea (www.pangaea.de).