Well-dated benthic foraminifer oxygen isotopic records (d18O) from different water depths and locations within the Atlantic Ocean exhibit distinct patterns and significant differences in timing over the last deglaciation. This has two implications: on the one hand, it confirms that benthic d18O cannot be used as a global correlation tool with millennial-scale precision, but on the other hand, the combination of benthic isotopic records with independent dating provides a wealth of information on past circulation changes. Comparing new South Atlantic benthic isotopic data with published benthic isotopic records, we show that (1) circulation changes first affected benthic d18O in the 1000-2200 m range, with marked decreases in benthic d18O taking place at ~17.5 cal. kyr B.P. (ka) due to the southward propagation of brine waters generated in the Nordic Seas during Heinrich Stadial 1 (HS1) cold period; (2) the arrival of d18O-depleted deglacial meltwater took place later at deeper North Atlantic sites; (3) hydrographic changes recorded in North Atlantic cores below 3000 m during HS1 do not correspond to simple alternations between northern- and southern-sourced water but likely reflect instead the incursion of brine-generated deep water of northern as well as southern origin; and (4) South Atlantic waters at ~44°S and ~3800 m depth remained isolated from better-ventilated northern-sourced water masses until after the resumption of North Atlantic Deep Water (NADW) formation at the onset of the Bølling-Allerod, which led to the propagation of NADW into the South Atlantic.

Core ENAM9321 age model is mainly based on correlation to GRIP ice d18O (Rasmussen et al., 1998). Core NA87-22 age model is fully described by Waelbroeck et al. (2001, 2006). Core MD95-2037 is published in the work by Gherardi et al. (2009). Core M35003-4 is described by Rühlemann et al. (1999). Core GeoB1711 is described by Vidal et al. (1999) and Waelbroeck et al. (2006). Core MD99-2334K is described by Skinner et al. (2003). Resulting errors on the calendar age have been computed in the present study as the maximum between the error on the calibrated age and the offset between the calibrated age and GISP2 age (Meese et al., 1997). Core KNR31 GPC-5 14C dates are from (Keigwin and Jones, 1994). Calendar ages have been computed using the Calib6.0 software (Stuiver and Reimer, 1993) and calibration data set marine09.14C (Reimer et al., 2009), assuming no additional surface reservoir age correction. Core MD07-3076 age model is fully described by Skinner et al. (2010).

Supplement to: Waelbroeck, Claire; Skinner, Luke C; Labeyrie, Laurent D; Duplessy, Jean-Claude; Michel, Elisabeth; Vázquez Riveiros, Natalia; Gherardi, Jeanne-Marie; Dewilde, Fabien (2011): The timing of deglacial circulation changes in the Atlantic. Paleoceanography, 26(3), PA3213