North Atlantic ocean circulation and abrupt climate change during the last glaciation

See allHide authors and affiliations

Science  29 Jul 2016:
Vol. 353, Issue 6298, pp. 470-474
DOI: 10.1126/science.aaf5529

An ocean of climate impacts

Large decreases in Atlantic meridional overturning circulation accompanied every one of the cold Northern Hemispheric stadial events that occurred during the heart of the last glacial period. These events, lasting on average around 1000 years each, have long been thought to result from changes in deep ocean circulation. Henry et al. used a suite of geochemical proxies from marine sediments to show that reductions in the export of northern deep waters occurred before and during stadial periods (see the Perspective by Schmittner). This observation firmly establishes the role of ocean circulation as a cause of abrupt glacial climate change during that interval.

Science, this issue p. 470; see also p. 445


The most recent ice age was characterized by rapid and hemispherically asynchronous climate oscillations, whose origin remains unresolved. Variations in oceanic meridional heat transport may contribute to these repeated climate changes, which were most pronounced during marine isotope stage 3, the glacial interval 25 thousand to 60 thousand years ago. We examined climate and ocean circulation proxies throughout this interval at high resolution in a deep North Atlantic sediment core, combining the kinematic tracer protactinium/thorium (Pa/Th) with the deep water-mass tracer, epibenthic δ13C. These indicators suggest reduced Atlantic overturning circulation during every cool northern stadial, with the greatest reductions during episodic Hudson Strait iceberg discharges, while sharp northern warming followed reinvigorated overturning. These results provide direct evidence for the ocean’s persistent, central role in abrupt glacial climate change.

Unlike the relatively stable preindustrial climate of the past 10 thousand years, glacial climate was characterized by repeated millennial oscillations (1). These alternating cold stadial and warm interstadial events were most abrupt and pronounced on Greenland and across much of the northern hemisphere, with the most extreme regional conditions during several Heinrich (H) events (2), catastrophic iceberg discharges into the subpolar North Atlantic Ocean. These abrupt events not only had an impact on global climate but also are associated with widespread reorganizations of the planet’s ecosystems (3). Geochemical fingerprinting of the ice-rafted detritus (IRD) associated with the most pronounced of these events consistently indicates a source in the Hudson Strait (HS) (4), so we abbreviate this subset of H events as HS events and their following cool periods as HS stadials. During northern stadials, ice cores show that Antarctica warmed, and each subsequent rapid northern hemisphere warming was followed shortly by cooling at high southern latitudes (5). Explanations for the rapidity and asynchrony of these climate changes require a mechanism for partitioning heat on a planetary scale, initiated either through reorganization of atmospheric structure (6) or the ocean’s thermohaline circulation, particularly the Atlantic meridional overturning circulation (AMOC) (710). Coupled climate models have successfully used each of these mechanisms to generate time series that replicate climate variability observed in paleoclimate archives (9, 11). We investigated the relationship between Northern Hemispheric climate as recorded in Greenland ice cores and marine sediments, along with isotopic deep-sea paleoproxies sensitive to changes in North Atlantic Deep Water (NADW) production and AMOC transport during marine isotope stage three (MIS3). Throughout that time, when global climate was neither as warm as today nor as cold as the last glacial maximum (LGM), ice sheets of intermediate size blanketed much of the northern hemisphere, and large millennial stadial-interstadial climate swings (6, 8) provide a wide dynamic range that allows examination of the ocean’s role in abrupt change.

Sediment samples were taken from the long (35 m) core KNR191-CDH19—recovered from the Bermuda Rise (33° 41.443’ N; 57° 34.559’ W, 4541 m water depth) in the northwestern Atlantic Ocean (Fig. 1), near previous seafloor sampling at Integrated Ocean Drilling Program (IODP) site 1063—and coring sites KNR31 GPC-5, EN120 GGC-1, MD95-2036, OCE326-GGC5, and others. Because this region of the deep North Atlantic is characterized by steep lateral gradients in tracers of NADW and Antarctic Bottom Water (AABW), the Bermuda Rise has been intensively used to explore the connection between changes in ocean circulation and climate (7, 12). In this study, we measured the radioisotopes 231Pa and 230Th in bulk sediment, age-corrected to the time of deposition, along with stable carbon (δ13C) and oxygen (δ18O) isotope ratios in the microfossil shells of both epibenthic foraminifera (Cibicidoides wuellerstorfi and Nuttallides umbonifera) and planktonic foraminifera (Globigerinoides ruber), respectively, yielding inferences on relative residence times and the origin of deep water masses on centennial time scales.

Fig. 1 Study core location and coretop distribution of Pa/Th.

Location sediment core CDH19 indicated with a star (33° 41.443’ N; 57° 34.559’ W, 4541-m water depth), with Pa/Th ratios (black dots) in core top sediments used with Ocean Data View Data-Interpolating Variational Analysis gridding to produce the color contours. White areas contain no data.

Isotopes of protactinium and thorium, 231Pa and 230Th, are produced from the decay of 235U and 234U, respectively, dissolved in seawater. This activity of 231Pa and 230Th in excess of the amount supported by the decay of uranium within the crystal lattice of the sediment’s mineral grains is denoted by 231Pxs and 230Thxs. Because the parent U isotopes have long residence times, U is well mixed throughout the ocean. This yields a 231Paxs/230Thxs (hereafter Pa/Th) production ratio (Pa/Th = 0.093) that is constant and uniformly distributed (13, 14). Both daughter isotopes are removed by adsorption onto settling particles, with Th more efficiently scavenged than Pa. The residence time of 231Paxsres = ~200 years) in seawater is thus greater than that of 230Thxsres = ~30 years), allowing 231Paxs to be redistributed laterally by changes in basin-scale circulation before deposition (7, 1416), with the additional potential influence of removal because of changes in particle rain associated with biological productivity (17). Settling particles (18) and surface sediments throughout the basin reveal a deficit in 231Paxs burial that is consistent with large-scale export by the deep circulation (Fig. 1) (19).

The downcore Pa/Th in core CDH-19 ranges from ~0.05 to slightly above the production ratio of 0.093, with a series of well-defined variations throughout MIS3 (Fig. 2). In sediments deposited during Greenland interstadial intervals (1), Pa/Th ratios average 0.0609 ± 0.0074 (2σ), which is substantially below the production ratio (Fig. 2) and only 10% higher than the mean value (Pa/Th = 0.055) of the Holocene, a time of relatively vigorous AMOC (7). Because 230 Thxs is buried in near balance with its production (20), the relatively low Pa/Th indicates a substantial lateral export of 231Paxs, which is consistent with relatively vigorous AMOC during interstadials, although the vertical integration through the water column of this deficit does not distinguish whether this export occurred at deep or intermediate levels. Epibenthic δ13C (δ13CBF) data allow discrimination between these two possibilities and display increased values during each interstadial, implying a greater contribution of the isotopically more positive North Atlantic end member (Fig. 2). During these intervals, this positive isotopic signal suggests that a deeper overturning cell was established, rather than a shallower, yet more vigorous one. This confirms a previous suggestion of intervals of relatively strong AMOC within the most recent ice age (21, 22), although Pa/Th and δ13CBF adjusted for whole-ocean inventory changes (23) rarely reach early Holocene values.

Fig. 2 Climate and circulation indices through MIS3.

Stadials are numbered with vertical bars. (A) NGRIP ice core δ18Oice 75.1°N, 42.32°W (35). (B) SST (°C) from MD95-2036, 33° 41.444’N, 57° 34.548’W, 4462 m (31). (C) Calcium x-ray fluorescence (orange) from core CDH19 (this study) mapped to %CaCO3, with calibration r2 = 0.87 (S.1), with spectral reflectance (blue) from core MD95-2036 (36). (D) Pa/Th from bulk sediment (green) taken from core CDH19. (E) δ13CBF from core CDH19 (purple) alternates between values consistent with southern and northern sourced δ13CBF end members.

Pa/Th increases within each Greenland stadial interval, for a mean duration of 0.531 ± 0.303 thousand years to a Pa/Th value of 0.0797 ± 0.0154, which indicates decreased lateral export of 231Paxs and is consistent with a shallower or reduced overturning cell in the North Atlantic. During these stadials, δ13CBF decreases substantially to negative values [–0.2 per mill (‰) to –0.5‰], suggesting greater influence of the glacial equivalent of modern Antarctic Bottom Water (AABW), an isotopic result that is consistent with reduced AMOC from a coupled climate model (10). Although the northern and southern water mass end members are not well known throughout the last glaciation, deep waters in the Atlantic during the LGM ranged from less than –0.5‰ in the south to greater than 1.5‰ in the north (23). If these values prevailed throughout MIS3, then the low δ13CBF indicates a dominant stadial influence of southern waters and substantial northward retreat or shoaling of the AABW/NADW mixing zone, which is consistent with the deep water mass configuration that has previously been reconstructed for the LGM (23, 24), although not for millennial-scale stadial intervals within the glaciation.

The mean Pa/Th of both stadials and interstadials is consistent with export of 231Paxs from the subtropical North Atlantic during most of MIS3. During peak interstadials, when low Pa/Th indicates the local burial of approximately half of 231Paxs production, the remaining half would have been exported. In contrast, the substantial decrease in the lateral export of 231Paxs evident in higher Pa/Th, along with lower δ13CBF during each stadial interval, points to repeated reductions in AMOC and its attendant northward heat transport throughout MIS3. The contrast between apparent deep, vigorous overturning during interstadials and shallower (25), weaker overturning during stadials is most pronounced in conjunction with all HS stadials (Fig. 2), when catastrophic discharge of melting icebergs from Canada flooded the subpolar North Atlantic (4).

Sediments deposited during HS stadials are characterized by a mean duration of 1.65 ± 0.545 thousand years and an average Pa/Th of 0.095 ± 0.016, which is indistinguishable from the production ratio. These results therefore indicate no net export of 231Paxs from the subtropical North Atlantic during these events sourced from the Hudson Strait. This balance between seawater radiometric production and underlying sedimentary burial would be expected under conditions with a substantial reduction in AMOC or other lateral transport and might imply a near cessation of 231Paxs export through deep circulation. Although variable scavenging may also contribute to sedimentary Pa/Th, values throughout MIS3 bear only a weak relationship with bulk and opal fluxes [coefficient of determination (r2) = 0.19] (19), which therefore constitute secondary influences.

These new results reveal that AMOC variations were associated with every MIS3 stadial-interstadial oscillation, with the largest reductions during HS stadials. The well-resolved interval 35 thousand to 50 thousand years ago provides a good example (Fig. 3). This iconic interval contains H4, H5, and the intervening series of oscillations that have served as a basis for conceptual and computer models seeking to explain such variability (811, 26, 27). A previous Pa/Th record (21) covering this interval captured much of the overall amplitude, and the new data resolve each stadial increase in Pa/Th, indicating that only HS4 and HS5 reach the production ratio of 0.093. Because the interstadial values are similar to each other, the subsequent abrupt increases in AMOC and regional warming are also the greatest and occur within the century-scale response time of Pa/Th. Throughout the records, the Pa/Th and δ13CBF bear a striking similarity to model output forced by freshwater anomalies (11).

Fig. 3 Detail of millennial cyclicity in glacial climate and the deep ocean.

(A) through (E) are as in Fig. 2, A to E. (F) Simulated NADW (Sv) in a coupled ocean/atmosphere model (11), with (D) published Pa/Th (gray squares) (21) and δ13CBF data (blue crosses) (12).

Combined with previous investigations (7, 28), these new results confirm that all HS events of the past 60 thousand years were associated with a dramatic increase in Pa/Th and are evidence for major reduction in AMOC in association with the largest IRD events (29). In contrast, H3, the sole Heinrich event stadial that fails to reach the production ratio (peak Pa/Th = 0.079), displays smaller IRD fluxes across the subpolar Atlantic (29), with provenance inconsistent with a Hudson Strait source (4). This muted result for H3 is consistent with evidence from the Florida Straits (30) showing a smaller reduction at that time in the northward flow of near-surface waters that feed the overturning circulation. As with all stadials, the HS events are characterized by lower δ13CBF, suggesting diminished influence of NADW and proportionately greater AABW on Bermuda Rise. Combined Pa/Th and δ13CBF results therefore indicate a persistent pattern of stadial weakening and interstadial strengthening, with a repeatedly largest reduction in AMOC associated with all HS events. Although these observations are consistent with a number of numerical model simulations (11, 27) as well as conceptual models for the mechanisms of abrupt change, they have previously been difficult to document and fully resolve.

Recent data from the Western Antarctic ice sheet provide compelling evidence for a robust lead of Greenland climate over Antarctica (5). That analysis revealed a Northern Hemisphere lead of 208 ± 96 years, indicating that the interhemispheric teleconnection propagates from north to south on time scales consistent with basin-scale ocean circulation. To ascertain whether Northern Hemisphere climate is forced or reinforced by changes in AMOC, we investigated the phase relationship between surface and deep-sea properties. Cross-correlations were performed on each of δ13CBF, Pa/Th, sea surface temperature (SST), and CaCO3 with North Greenland Ice Core Project (NGRIP) δ18O from both sediment cores CDH19 and MD95-2036 from the Bermuda Rise. The optimal correlation of δ13CBF leads NGRIP δ18O by approximately 2 centuries (Fig. 4). This lead is corroborated by Pa/Th phasing, which when considering the century-scale response time of the proxy (13, 14) is consistent with AMOC changes indicated by δ13CBF. The SST reconstruction from MD95-2036 was aligned with Greenland δ18O, yielding a correlation of r2 = 0.83 (31). SST and Pa/Th are synchronous with NGRIP to within the estimated bioturbation error of 8 cm within the core, displaying correlations with Greenland of r2 = 0.47 for Pa/Th and r2 = 0.65 for SST. The optimal correlation of %CaCO3, r2 = 0.64, lags NGRIP δ18O by nearly 200 years.

Fig. 4 Phasing lag correlations.

Correlation of NGRIP ice core δ18O with CDH19 %CaCO3 (orange), Pa/Th of bulk sediment from CDH19 (green), δ13CBF from CDH19 (purple), and SST °C from MD95-2036 (31) (red).

The consistent lead of variations in δ13CBF before SST and Greenland temperatures, repeated over multiple millennial cycles, indicates the potential influence of AMOC on NH climate and confirms that the Bermuda Rise is exposed to shifts in deep-water mass mixing. Initially, deep circulation changes, which is evidenced overall by the timing of δ13CBF. Pa/Th shifts are essentially in tandem with regional temperature when circulation accelerates, and soon thereafter as it responds to weakening AMOC (19). Given the response time of Pa/Th to instantaneous shifts in North Atlantic overturning (13, 14), this also suggests that changes in AMOC precede regional temperature change, although the exact timing may have differed during cooling and warming phases. Both SST and Greenland temperature proxies lag the ocean circulation in a consistent fashion, and in turn, these northern changes have been demonstrated to lead Antarctic temperatures (5). Calcium-carbonate concentration is the last of the proxies to respond to AMOC change, which is consistent with the longer time scale of preservation, dissolution, and dilution in the deep ocean.

The relative timing of the observed AMOC changes has important implications for regional and global climate. Whereas numerous computer simulations suggest that melting icebergs and other freshwater input associated with H events may have shut down NADW production (9, 11, 27), recent results examining the phasing of North Atlantic SST and IRD suggest that stadial conditions began to develop before ice-rafting (32). The evidence here nevertheless indicates that the greatest AMOC reduction and the coldest stadial intervals accompanied the largest iceberg discharges. This suggests that the iceberg discharges may have provided a positive feedback mechanism to accelerate the initial cooling within each multimillennial climate cycle. In addition, the extended H-stadial reductions in AMOC observed in this study coincide with intervals of rising atmospheric CO2 (33), whereas CO2 declined when AMOC increased during the subsequent sharp transitions to northern interstadials, supporting a potential influence on the atmosphere by the deep circulation on millennial time scales (34).

The robust relationship of reductions in export of northern deep waters evident in reduced 231Paxs export and decreased δ13CBF before and during stadial periods, and the dramatic increases in both during interstadials, provide direct evidence for the role of AMOC in abrupt glacial climate change. The sequence of marked circulation changes and northern hemisphere climate detailed here, combined with the demonstrated lag of Antarctic temperature variations (5), strongly implicates changes in meridional heat transport by the ocean as a trigger for abrupt northern hemisphere warming and the tipping of the “bipolar seesaw” (26).

Supplementary Materials

Materials and Methods


Figs. S1 to S7

References (3755)

Database S1

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
Acknowledgments: Data are available at This research was supported in part by a NSF Graduate Research Fellowship to L.G.H., by awards from the Comer Science and Education Foundation and NSF grant ATM-0936496 to J.F.M., and an award from the LDEO Climate Center to L.G.H. and J.F.M. L.D.K. and W.B.C. were supported by NSF grant ATM-0836472, and L.D.K. was supported by NSF grant AGS-1548160. We thank M. Jeglinski and K. Rose for technical support. The authors thank R. Anderson, S. Hemming, and C. Hayes for constructive discussion leading to improvement of the manuscript; R. Anderson for unpublished data included in the Fig. 1 map; and M. Fleisher for analytical support.

Stay Connected to Science


Navigate This Article