A 0.5-Million-Year Record of Millennial-Scale Climate Variability in the North Atlantic

See allHide authors and affiliations

Science  12 Feb 1999:
Vol. 283, Issue 5404, pp. 971-975
DOI: 10.1126/science.283.5404.971


Long, continuous, marine sediment records from the subpolar North Atlantic document the glacial modulation of regional climate instability throughout the past 0.5 million years. Whenever ice sheet size surpasses a critical threshold indicated by the benthic oxygen isotope (δ18O) value of 3.5 per mil during each of the past five glaciation cycles, indicators of iceberg discharge and sea-surface temperature display dramatically larger amplitudes of millennial-scale variability than when ice sheets are small. Sea-surface temperature oscillations of 1° to 2°C increase in size to approximately 4° to 6°C, and catastrophic iceberg discharges begin alternating repeatedly with brief quiescent intervals. The glacial growth associated with this amplification threshold represents a relatively small departure from the modern ice sheet configuration and sea level. Instability characterizes nearly all observed climate states, with the exception of a limited range of baseline conditions that includes the current Holocene interglacial.

Climatic instability on suborbital time scales (103 to 104 years) characterized the North Atlantic region during the last glacial period. Rapid oscillations of 5° to 10°C in the atmospheric temperature over Greenland (1) and faunal shifts equivalent to more than 4°C in sea-surface temperature (SST) in the subpolar ocean (2,3) occurred repeatedly throughout the 100,000 years following the peak interval of the last interglacial. Millennial-scale pacing of climate variability has continued into the current Holocene interglacial (4), but its amplitude has usually been limited to 2° to 3°C over Greenland (1, 5) and to less than 2°C in the North Atlantic (4), implying that rapid climate change may be amplified during glacial intervals and diminished during interglacial intervals. Records of rapid climate oscillations during previous glacial and interglacial intervals of varying intensity may help to establish whether such a link exists between ice sheet size and regional and global climate instability. We present continuous high-resolution climate records from deep-sea sediments encompassing the last 0.5 million years (My) in the subpolar North Atlantic. These records cover a wide range of global ice volume, from somewhat less than the modern total of 80 m sea-level equivalent to well over 200 m sea-level equivalent and indicate that climatic stability on millennial time scales prevails only when the extent of global glaciation falls within a narrow range, including that of the current Holocene interglacial. Climatic instability is more typical, occurring during fully developed glacials, transitional intervals of ice-sheet growth and decay, and even during stadial portions of interglacials, when limited ice-sheet growth occurs. The threshold separating stable and unstable climatic regimes represents a relatively small departure from the modern ice sheet configuration, whereas sufficiently extreme glaciations may also establish less variable climate states.

The sediments for this study were recovered at Site 980 (Fig. 1) on the Feni abyssal drift during Leg 162 of the Ocean Drilling Program (ODP). Previous studies of the last climate cycle (2, 6) confirmed that regional climate variations are faithfully recorded in the rapidly accumulating sediments on Feni (13 cm/ky in the sediments studied here). A chronostratigraphy for Site 980 was established using the oxygen isotope ratio (δ18O) of calcite shells secreted by the protozoan Cibicidoides wuellerstorfi, an epifaunal benthic (bottom-dwelling) foraminifera (Fig. 2). The conventional glacial (even) and interglacial (odd) marine isotope stages (MIS) (7) were easily identified in the benthic δ18O record, and an orbitally tuned age model was derived by graphical correlation with existing deep-sea chronologies (8). The 65-m section we examined represents 0.5 My of deposition and captures five of the large-amplitude glacial-interglacial (G-IG) cycles of ∼100-ky duration that have characterized the global climate record during this time.

Figure 1

Location of ODP Site 980 (55°29′N, 14°42′W, depth of 2179 m) in the eastern North Atlantic.

Figure 2

Time series of IRD and isotopic measurements from Site 980 (sample resolution varies from 300 to 1200 years). IRD is plotted as the number of detrital sediment grains (lithics), larger than 150 μm, per bulk sample weight. Benthic (C. wuellerstorfi, red) and planktic (N. pachyderma d., green) foraminifera δ18O values are plotted relative to the Pee Dee belemnite (Belemnitella) (PDB) standard (7). Isotope measurements were made at the Woods Hole Oceanographic Institution on a Finnigan MAT252 coupled to an automated carbonate preparation device consisting of 46 single-reaction chambers, with 23 chambers on each of two lines. Acid temperature was ∼70°C. The standard deviation of the isotope values of National Bureau of Standards (NBS) carbonate standard NBS19 is 0.08 per mil and 0.04 per mil for δ18O and δ13C, respectively. Calibration to PDB was via NBS19 (δ18O = –2.2 VPDB, δ13C = 1.95 VPDB). Data points at 0 ky are δ18O measurements on modern N. pachyderma d.from Atlantic sediment traps at 47°N (×) and 65°N (+), demonstrating an isotopic range surpassing that of our sedimentary data. The modern data from 65°N are adjusted by 1.0 per mil, a minimum estimate of past mean ocean changes (28), for comparison to glacial data. Interglacial MIS and the last peak interglacial substage 5e are indicated for reference.

We examined two proxies of sea-surface conditions: the relative abundance of ice-rafted debris (IRD), and the δ18O of the planktic (surface-dwelling) foraminifera Neogloboquadrina pachyderma, dextral-coiling (N. pachyderma d.). The glacial occurrence of IRD throughout much of the region is long-established (9), and episodic pulses of IRD (10,11) were among the first clues for suborbital variability in the glacial North Atlantic. Planktic foraminiferal δ18O is influenced by the temperature and δ18O of ambient seawater (7), and is therefore a useful indicator of sea-surface hydrography. We also examined benthic δ13C, a proxy for changes in deep ocean circulation.

Ice-rafting at Site 980 (Fig. 2) is episodic throughout, with repeated peaks of IRD concentrations surpassing 500 lithics per gram. None of these peaks lasts for more than a few thousand years, and the abundance always returns to very low baseline values before rising again, even during times of large continental ice sheets. The size and frequency of the IRD events increases, however, along with the size of the ice sheets during each 100-ky cycle. Within the youngest portion of the Site 980 record, the largest peaks include the well-documented “Heinrich events,” catastrophic iceberg discharges that punctuated the last glacial (10, 11). More than 50 comparable IRD peaks occur throughout the entire record, indicating that such episodic behavior is a fundamental aspect of Pleistocene glacial cycles. Ice-rafting is generally diminished during interglacials. This is most likely related to the simple requirement of sufficient continental ice near the North Atlantic prior to catastrophic discharge, although even the modest ice growth within stadial substages of MIS 5, 7, and 9, the last three interglacials, has associated IRD events. The concentration of the largest IRD peaks within the glacial half of each 100-ky cycle yields a repeat time of approximately 6 ky (40 events in 250 ky). Although this might suggest that a lower frequency of recurrence prevails during interglacial intervals, it is also possible that the frequency of the events themselves remains the same (4), and that amplitude modulation renders the size of the events more prominent, and thus easily identifiable, during glacial intervals.

The relationship between benthic δ18O and both the initial and final IRD pulses of each 100-ky cycle is noteworthy. During each cycle, an IRD event follows the first increase in benthic δ18O to values exceeding 3.5 per mil. Likewise, a large IRD event accompanies each of the rapid deglaciations, or “terminations” (12), from values greater than 4.0 per mil, through 3.5 per mil, to less than 3.0 per mil. The close association of the onset and cessation of ice-rafting events with a particular value in the benthic δ18O proxy for ice volume suggests that this value represents an important threshold in ice sheet growth.

Neogloboquadrina pachyderma d. is abundant in subpolar regions and constitutes >10% of the coretop fossil assemblage underlying nearly all North Atlantic surface waters with summer SST of 10° to 20°C (13). Despite the dramatic and repeated changes in the planktic foraminiferal assemblage that occurred near Site 980 during the last 0.5 My (14), N. pachyderma d. was present throughout, yielding a continuous δ18O record (Fig. 2) containing several large-amplitude (3.0 per mil) 100-ky G-IG cycles. Rapid 1.0–per mil oscillations occur within the glacial portion of each large cycle, while variability is limited to 0.1 to 0.3 per mil during peak interglacials.

Previous studies indicate that N. pachyderma d.δ18O is useful as a proxy for SST during the Late Pleistocene in the subpolar North Atlantic (15). Before considering the climatic implications of the Site 980 N. pachyderma d. δ18O record, we therefore converted the values to estimates of SST by removing the global signature of ice volume, as captured by the normalized benthic C. wuellerstorfi δ18O (16). The resulting isotopic paleothermometer has several distinct benefits for assessing variability throughout the record. First, the large dynamic range is not subject to signal saturation at warm or cold extremes, which might limit other methods, such as those based on faunal percentages. Second, the δ18O-temperature relationship (7, 17) allows a direct comparison of the amplitude of climate changes at different times in the past. Third, any diminished interglacial variability at Site 980 is unlikely to be a result of faunal saturation through seasonal or depth migration of the foraminifera, because peak SSTs are well within the range tolerated byN. pachyderma d. in the modern ocean (13).

The temperature estimates (Fig. 3) confirm that SST varied throughout the last 0.5 My, reaching 13° to 15°C during each interglacial and falling to 5° to 7°C within each glacial. During each G-IG cycle, ΔSST was 8° to 9°C, comparable to faunally derived estimates (18) from studies of each of the last two G-IG cycles near Site 980. Although the SST minima, maxima, and therefore the overall ΔSST are all in close agreement with previous estimates, the character of the Site 980 record is distinctive. Each of the glacial intervals is interrupted repeatedly by rapid 3° to 6°C oscillations. Only within MIS 12 (and possibly MIS 2) is there any extended glacial interval with SST variability of less than 4°C. Interglacial SSTs are generally less variable, although excursions of 3° to 5°C occur within stadial substages of MIS 5 and MIS 9.

Figure 3

Time series of SST and seawater δ18O from Site 980. Seawater δ18O, the ice volume component of benthic δ18O, was obtained by adjusting benthic δ18O for a 3°C interglacial warming of the deep ocean. SST was calculated by subtracting changes in the ice volume signal from planktic δ18O and scaling the isotopic residual as a departure from modern temperature (15). The scaling factor varies from –0.21 to –0.23 per mil/°C, in accord with a paleotemperature equation, T = 15.18 – 4.79(δc – δw) + 0.08(δc – δw)2, derived from recent measurements (17) of the equilibrium oxygen isotopic fractionation (δ18O) between calcite (δc) and water (δw). Filter output results are derived from interpolation of SST data at 500-year increments and application of a zero phase, 128-order, Hamming window with cutoff frequencies of 1/1 and 1/12 ky. Glacial MIS are indicated for reference.

The Site 980 records confirm the fundamentally dynamic nature of the Pleistocene ice ages. Harsh conditions prevailed repeatedly in the North Atlantic, yet never for more than a few thousand years. Variability on millennial time scales has thus been the rule, rather than the exception. This characteristic may help explain the current failure of coupled ocean-atmosphere global climate models to achieve a stable equilibrium glacial climate. As indicated by benthic δ18O, ice volume was instrumental in setting the degree of climate stability. No large oscillations occur when the benthic δ18O is at minimum values and few occur when it is less than 3.5 per mil (Fig. 2). The longest relatively ice-free interval occurred during interglacial MIS 11, which also has the longest interval (30 to 40 ky) without amplified millennial-scale SST (and IRD) variability. MIS 11 has long been held to be an especially mild interglacial in many respects (19), and such a particularly IRD-free interval may indicate diminished calving or size of the Greenland ice sheet, or both, in addition to the absence of other circum–North Atlantic ice sheets. Briefer intervals of relative climate stability occurred during the portions of MIS 5, 7, 9, and 13 with the least ice volume. During these peak interglacials, SST variations were limited to 1° to 2°C. Following the last crossing of the 3.5–per mil threshold, at the time of the Younger Dryas climatic reversal, diminished amplitudes also characterized the variability in MIS 1 (4, 5). Such times of diminished ice volume appear to be respites from the prevailing instability of the last 0.5 My.

The persistence of climatic variability is unequivocal, but its cause remains obscure. The millennial time scale of the observed changes, the apparently heightened sensitivity of the North Atlantic region (20), and the theoretical existence of multiple quasi-stable modes (21), all suggest that changes in the ocean's thermohaline circulation (22, 23) may play a role. In order to assess this possibility, we have also evaluated the relative vigor of the northern component of the deep circulation by examining the carbon isotopic ratio (δ13C) of C. wuellerstorfi at Site 980. Although multiple factors may affect the δ13C of benthic calcite, to first approximation, the signal at Site 980 can be attributed to the varying influence of competing water masses below 2 km water depth in the subpolar North Atlantic (22, 23). Northern source water (North Atlantic deep water, or NADW, and glacial North Atlantic intermediate water, or GNAIW) is nutrient-depleted, with high δ13C values, and southern source water is nutrient-rich, with low δ13C values. Throughout the last 0.5 My, the benthic δ13C at Site 980 has varied repeatedly (Fig. 4) on similar time scales as the SST and IRD signals, with stronger influence of northern source water associated with warmer temperatures and minimal ice-rafting. It is tempting to link these observations, considering the poleward heat transport associated with vigorous deepwater production; however, variability in the δ13C signal continues throughout the record, without a threshold at the 3.5– per mil benthic δ18O value. Significant oscillations (>0.4 per mil) occur even during the peak interglacials at MIS 1, 5e, and 11, when benthic δ18O dropped to 2.5 per mil, SST variations were minimal, and IRD input was negligible. The persistence of variability during relatively ice-free intervals makes it unlikely that the oscillations in thermohaline circulation are driven by temperature or salinity perturbations associated with catastrophic iceberg discharge events. It is more likely that each diminuition of NADW production and associated poleward oceanic heat transport, whether caused by ocean-atmosphere interactions or internal oscillations, contributes to lowered SST and increased ice-rafting. When ice sheets are of sufficient size to produce abundant icebergs, their meltwater would then serve as a positive feedback, stabilizing the upper water column and further diminishing NADW (6, 10,11). This interpretation is supported by the fact that many of the δ13C minima associated with the largest IRD events are particularly prominent, including those that accompany each glacial termination. The lack of a clear correspondence between subtle SST changes and δ13C during peak interglacials may indicate the limited influence of circulation changes on SST in the absence of such feedbacks.

Figure 4

Time series of IRD and benthic carbon isotopic (δ13C, per mil) measurements from Site 980. IRD is plotted as the percentage of sediment grains larger than 150 μm that are detrital. This particular IRD index is not sensitive to changes in mean sedimentation rate, which may be influenced by variations in abyssal currents associated with the strength of thermohaline circulation. Benthic δ13C values are plotted relative to the PDB standard. Interglacial MIS and the last peak interglacial substage 5e are indicated for reference.

Site 980 is at a sensitive depth to monitor changes in the thermohaline circulation because it is bathed by deep water just below the hinge line of glacial partitioning of the water column (23). The proximity to this important boundary also lends caution to our interpretation of the benthic δ13C. Small shifts in water masses across a steep gradient may be recorded as large amplitude changes in physical and chemical properties at a single location. The interpretation of the δ13C record will need to be refined as comparable high-resolution records at additional depths and locations become available. Nevertheless, Site 980 appears to capture persistent, significant variability in the deep circulation. The largest interglacial (>0.4 per mil) and glacial (>1.0 per mil) amplitudes are comparable to the entire range of values found in the modern (24) and glacial (22, 23) North Atlantic. Many smaller amplitude oscillations occur as well, suggesting that the Site 980 δ13C record does not exhibit a square wave alternation between the characteristic values of two different water masses, but rather approximates the relative influence of those water masses at the site, as set by the basin-wide state of circulation.

The occurrence of circulation changes during interglacial intervals suggests that neither a glacial salt-oscillator (25) nor a binge-purge iceberg discharge mechanism (26) can fully account for the persistent variability. One alternative, an extension of the oscillator concept (22), may be that the thermohaline circulation does not operate at an equilibrium level of salt and freshwater transport, but alternates between modes that export an excess of one or the other. Such an oscillator would operate regardless of glacial state, while remaining sensitive to the same glacial influence that is evident in the surface climate proxies.

That the presence and size of ice sheets should influence climate is not surprising, given their affect on the atmospheric circulation and hydrologic balance (27). It is nonetheless instructive to note the small ice volume change associated with the onset of a dramatic shift in the climate response. A rise in benthic δ18O from the peak interglacial minimum of 2.5 per mil to the 3.5 per mil level constitutes half of the range for each 100-ky G-IG cycle, yet it represents proportionately little growth in ice or fall in sea level, because much of the δ18O rise over that interval reflects cooling of the deep ocean (28, 29). The 3.5–per mil threshold thus corresponds to glacial ice growth that lowers sea level as little as 30 m below the modern (2.5 per mil) level. Although this change is not negligible, it represents no more than 25% of the total 120-m difference between modern sea level and that of the last glacial maximum (28). It is also significantly less than the existing 80-m sea level depression which is due to modern ice volume (30).

The sharp contrast in variability at benthic δ18O values above and below approximately 3.5 per mil suggests that the extent of glaciation corresponding to this value constitutes an important and persistent physical threshold. Indeed, a similar relationship between benthic δ18O and IRD events was documented between 1.2 and 1.4 million years ago (31), and major ice-rafting in the subpolar North Atlantic did not begin prior to the attainment of comparable benthic δ18O values 2.5 to 2.7 million years ago (32).

Although the physical significance of the 3.5–per mil threshold is difficult to establish, there are several possibilities. It represents a total amount of global ice (approximately equivalent to 105 to 110 m of sea level), which may itself be important, as the distribution of water between solid and liquid phase has implications for the hydrologic cycle. Alternatively, the location of ice may be the crucial factor, influenced in turn by the amount. The 3.5–per mil threshold may represent ice growth across a key strait such as North Greenland–Canada, or the establishment of grounding and export of ice from marginal-sea nucleation sites (33). The key factor may be sea ice advance associated with glacial ice growth. Extensive sea ice growth in the Labrador and Nordic seas, and in the North Atlantic, would all increase the regional sensitivity to changes in ocean heat transport. Yet another possibility is the importance of the shape and height of ice sheets. The growing Laurentide ice sheet would eventually interact with the jet stream, deflecting it southward and cooling the North Atlantic (27). The associated steepening of meridional SST gradients would also sensitize the subpolar North Atlantic to variations in poleward ocean heat transport. Although such a steepening of gradients has almost certainly occurred (14, 18), it is less well established whether this was due to a particular ice sheet size or geometry. The 3.5–per mil threshold may also result from a combination of the factors mentioned here.

The possibility of another, less dramatic threshold at larger ice volumes is suggested by the relative diminuition of SST variability during MIS 12, the most severe glaciation of the last 0.5 My (29). As was previously documented in the North Atlantic region (1, 2), variability at Site 980 was also diminished during MIS 2, although not in the comparable MIS 6, so these glaciations of 4.5 to 4.6 per mil in benthic δ18O may just achieve a second, glacial, threshold. The evidence from MIS 12 suggests that sufficiently large ice sheets might be associated with climate regimes of diminished variability. It is also possible that the peak interglacial and glacial intervals of the past 0.5 My are not climatic end members, but lie within zones of relative stability, beyond which variability might increase again (Fig. 5). This possibility is supported by model results predicting enhanced instability due to changes in the hydrologic cycle associated with both warmer (34) and colder (35) departures from modern interglacial conditions. Because the range of climate states examined in this study covers nearly the entire range of those found in the Quaternary, subsequent attempts to evaluate this possibility may need to reach farther back in the geologic record, or rely at least partially on mathematical models.

Figure 5

Schematic representation of possible climatic stability regimes.

  • * To whom correspondence should be addressed. E-mail: jmcmanus{at}


View Abstract

Stay Connected to Science

Editor's Blog

Navigate This Article