340,000-Year Centennial-Scale Marine Record of Southern Hemisphere Climatic Oscillation

See allHide authors and affiliations

Science  15 Aug 2003:
Vol. 301, Issue 5635, pp. 948-952
DOI: 10.1126/science.1084451


In order to investigate rapid climatic changes at mid-southern latitudes, we have developed centennial-scale paleoceanographic records from the southwest Pacific that enable detailed comparison with Antarctic ice core records. These records suggest close coupling of mid-southern latitudes with Antarctic climate during deglacial and interglacial periods. Glacial sections display higher variability than is seen in Antarctic ice cores, which implies climatic decoupling between mid- and high southern latitudes due to enhanced circum-Antarctic circulation. Structural and temporal similarity with the Greenland ice core record is evident in glacial sections and suggests a degree of interhemispheric synchroneity not predicted from bipolar ice core correlations.

Climatic instability and abrupt changes are recognized features of the last glacial period and have been found in climate archives worldwide (16). To date, the best-resolved records of changes in the Southern Hemisphere are from Antarctic ice cores (5, 6). The dearth of continuous high-resolution paleoclimatic records from mid- and high southern latitudes precludes the tracing of climatic signals from Antarctic records to the extra-polar Southern Hemisphere. This highlights the importance of establishing fine-scale marine records that enable the mapping of millennial to submillennial climatic changes across southern latitudes, which is necessary for constructing a coherent picture of interhemispheric climate patterns. Here we present sea surface temperature (SST) and stable oxygen isotope (δ18O) records from a marine sediment core that span the past three climatic cycles and resolve mid-southern latitude climate variability at a temporal resolution close to, and in some sections higher than, the Antarctic Vostok climate record.

Thirty-six-meter-long IMAGES (International Marine Past Global Changes Studies) (7) core MD97-2120 was retrieved from Chatham Rise, east of New Zealand (45°32.06′S, 174°55.85′E; water depth, 1210 m). The hydrography at the core site is controlled by the Subtropical Convergence (STC), which lies immediately to the north, separating warm, saline, subtropical surface waters from colder, fresher subantarctic waters (8). We established centennial-scale records of planktonic foraminiferal (Globigerina bulloides) δ18O (δ18Oplk) and Mg/Ca ratios (9) that reflect surface water hydrographic changes over the past 340,000 years.

An age scale has been established for the records through the application of a series of iterative steps. The age of the upper core section (0 to 10.6 m) is constrained by 13 accelerator mass spectrometry (AMS) 14C dates and by the Kawakawa ash [radiocarbon age, 22.59 thousand years (ky); equivalent calendar-year age, 26.17 ky (10, 11)]. The 14C ages between 0 and 20 ky were converted to calendar ages using the marine calibration data set (12), together with a local reservoir age anomaly of 240 ± 40 years (13). AMS ages between 26.6 and 32.3 ky were reservoir-corrected by 640 years and converted into calendar years using a marine 14C–versus–calendar year correlation that traces geomagnetic paleointensity variations (9, 14, 15). To further constrain our age model, we measured a fine-scale benthic δ18O section that was correlated with the glacial δ18O section of core MD95-2042 in the northeast Atlantic (9, 16) in the interval from 40 to 72 ky. The chronology of core MD95-2042, through its δ18Oplk, is tied to the Greenland Ice Sheet Project 2 (GISP2) time scale (16). The glacial section of the benthic δ18O record of core MD95-2042 mimics the structure of the Antarctic Vostok ice core deuterium (δD) record (16), which is also seen in our benthic δ18O record, thus allowing the transfer of the Greenland ice core chronology from core MD95-2042 to our records. The age scale for the lower part of the core, below 10.6 m, is derived from the correlation of our Mg/Ca-derived SST (SSTMg/Ca) record with the Antarctic Vostok δD atmospheric temperature record on its orbitally tuned age scale (17). We used the lower-resolution SSTMg/Ca record for this correlation, because δ18Oplk combines local hydrological and global ice volume effects that impede a firm correlation with Antarctic temperature. According to the age model, sedimentation rates along core MD97-2120 are between 15.5 cm/ky during glacials and 6 cm/ky during interglacials. Mean time steps along the δ18Oplk and Mg/Ca records are 192 ± 83 years (at a 2-cm sampling interval) and 931 ± 405 years (at a 10-cm sampling interval). The resolution of the δ18Oplk record is close to that of the Vostok δD record (time step, 126 ± 100 years). In the lower sections, beyond ∼230 ky, the resolution of δ18Oplk is the same as and higher than in the Vostok record.

Mg/Ca signals in the carbonate of foraminiferal shells are temperature-dependent and can be converted into a record of SST through the application of empirical equations. We derived SST from Mg/Ca using the calibration of Mashiotta et al. (9, 18). The core-top SSTMg/Ca estimate of 11.8°C is within the modern SST range of 8.5° to 14.5°C around the core location (19) and agrees well with a calcification of G. bulloides in austral spring (20). SSTMg/Ca on glacial-interglacial time scales ranges from 6.5° to 16°C, with mean glacial-interglacial changes in the range from 4.9° to 6.8°C (Fig. 1). On millennial time scales, SSTMg/Ca displays amplitudes of up to 3°C; that is, 40 to 60% of the total glacial-interglacial amplitude (Fig. 2). The last glacial-interglacial SSTMg/Ca change of 4.9°C is within the range of amplitudes of 4° to 6°C that is derived from temperature-sensitive alkenones and foraminiferal census counts at nearby core sites (21, 22).

Fig. 1.

340-ky-long δ18Oplk and SSTMg/Ca records from core MD97-2120 compared to the Vostok δD record (5). (A) G. bulloides δ18O was measured at a sampling interval of 2 cm, equivalent to a time step of 192 ± 83 years. (B) SSTMg/Ca record (time step, 931 ± 405 years). The Mg/Ca scale is also given. Mg/Ca ratios were converted to SST using the equation of Mashiotta et al. (18) [Mg/Ca = 0.4740.107T (T, temperature)]. (C) δD record (time step, 126 ± 100 years) of the Antarctic Vostok ice core on the orbitally tuned age scale of (17). Beyond ∼230 ky, the resolution of δ18Oplk is equal to or exceeds that of the Vostok record. δD documents air temperature changes that are indicated along the δD axis as deviations from the modern value (ΔT) (5). Vostok ΔT and our SSTMg/Ca are scaled to the same temperature range to facilitate comparison of both records. (D) Residual δ18Oplk [variability ≥5 ky was removed from δ18Oplk in (A)] (39). Amplitudes of the residual δ18Oplk are systematically increased during glacials, indicating increased climate variability during periods of enhanced global ice volumes. Steep δ18Oplk gradients labeled TI through TIV are glacial Terminations I through IV. Blue vertical bars mark the Antarctic Cold Reversal (ACR) [14 to 12.4 ky (5, 6)], and a prominent δ18Oplk reversal in mid–Termination III. Gray vertical bars denote δ18Oplk and SSTMg/Ca excursions not seen in the Vostok δD record. In contrast to the ACR, the δ18Oplk reversal in TIII does not coincide with a drop in SSTMg/Ca, suggesting a local seawater δ18Ow anomaly at this time. The mid-TIII reversal is not as clearly developed in the Vostok δD record but exists as a distinct anomaly in the Vostok Ar record (29). Age control points used to construct the age model are shown along the bottom axis as follows: 14C AMS dates (gray), Kawakawa tephra (red), tie points for correlating the benthic δ18O section to that of northeast Atlantic core MD95-2042 (9, 16) (blue), and tie points for correlating SSTMg/Ca to the Vostok δD record (black).

Fig. 2.

δ18Oplk, SSTMg/Ca, and local δ18Ow records of core MD97-2120 compared to the Greenland GISP2 δ18Oice record and the Antarctic Vostok and Byrd ice core records for the past 90 ky. (A) δ18Oice record from GISP2 (2). Numbers denote D/O interstadials during MIS 3. (B) δ18Oplk record of core MD97-2120. (C) SSTMg/Ca record. A Mg/Ca scale is given along the SST axis. (D) Local δ18Ow anomaly as derived from δ18Oplk and SSTMg/Ca after correcting δ18Oplk for mean ocean δ18Ow changes (9, 32). For δ18Ow calculation, δ18Oplk was sampled at the time step of the lower-resolution SSTMg/Ca record. The δ18Ow anomaly was calculated using the mean Holocene (0 to 6 ky) value as a reference. The equivalent salinity anomaly (the inside scale along the δ18Ow axis) is based on a δ18Ow-salinity relation for high southern latitudes (9). The onset and trend of increasing Australian aridity at 40 ky are from (31). Control points along bottom axis were used to construct the age model as in Fig. 1, now with error bars of the calibrated 14C ages. (E and F) Antarctic Vostok δD (5) and Byrd δ18Oice ice core records (6) are shown for comparison. Single-headed arrows along the δ18Oplk record in (B) denote rapid transitions and short events not seen in Antarctic ice core records. The double-headed arrow in (B) marks the large δ18Oplk and SST Mg/Ca excursion during MIS 4, which is not present in the Vostok δD record. The event immediately follows the warm anomaly in Byrd at ∼70 ky and partly overlaps with the Greenland D/O 19 event. Bars along the δ18Ow anomaly in (D) indicate increased IRD delivery on Campbell Plateau (CP0 to CP8), south of Chatham Rise (48° to 53°S) (34), and in the South Atlantic at 41° and 53°S (SA0 to SA6) (33). SA0 and CP0 are coeval with the ACR and coincide in our SSS record with a discrete minimum. SA1 to SA4 are associated directly with, or are on the shoulders of, SSS minima. The off-SSS-peak positions of some SA events are due to age model uncertainty. Linking with times of increased IRD delivery on Campbell Plateau (CP1-6) is more direct because of proximity to our core site; these phases correspond to SSS minima. SA5 and SA6 and CP7 and CP8 are exceptional in that they correlate with a broad SSTMg/Ca maximum in our record but with only moderately increased SSS.

Carbonate dissolution during cold periods at our 1210-m core site (23) has the potential to lower Mg/Ca and thus the inferred SSTMg/Ca (24, 25). The effect of this would be to increase the temperature amplitudes between glacial and interglacial periods. The Ca content of the sediment varies from 5 to 25%, which by itself might imply dissolution, but there are substantial dissimilarities as well as similarities between the SSTMg/Ca record and bulk sediment carbonate concentrations along the core (26). Additionally, the comparison with independent SST reconstructions from nearby cores (see above) shows good agreement: There is, at most, a 0.9°C overestimation of the last glacial-interglacial SSTMg/Ca. This is much smaller than the millennial-scale SSTMg/Ca changes of up to 3°C, suggesting that the effect of any dissolution on the SSTMg/Ca variability in our core is minor.

SSTMg/Ca and δ18Oplk co-vary, with SSTMg/Ca leading on average by 1.7 ± 0.3 ky, 2.2 ± 0.1 ky, and 0.8 ± 0.1 ky (90% confidence intervals) in the orbital eccentricity, obliquity, and precession bands, respectively. SST leads over planktonic δ18O values of similar magnitude have been found at low-latitude (27, 28) and Southern Ocean sites (18) and are also seen in Antarctic temperature over global ice volume (17, 29). However, the leads in our records are concentrated primarily within the cooling trends of interglacial marine isotope stage (MIS) 5e-d, MIS 7c-a, and the MIS 9a-8 transition; whereas in glacial sections and during glacial terminations, SSTMg/Ca is synchronous with δ18Oplk (9). The variable phasing we observe between our records indicates that the contribution of local temperature and salinity effects to δ18Oplk varied as a function of glacial-interglacial climate.

δ18Oplk, like Mg/Ca, is SST-sensitive, with additional contributions coming from global ice volume and local surface water δ18O (δ18Ow) signals in δ18Oplk. The combined use of SSTMg/Ca and δ18Oplk in paleotemperature equations enables us to infer δ18Ow that can then be translated into sea surface salinity (SSS) equivalents (9). The SSS record shows that salinity was systematically decreased during glacials by an average of 1.3 to 1.5 units (Fig. 3). On faster millennial time scales, salinity likewise varies at high amplitudes, around 1.2 units, with salinity decreases concentrated around stadial (cold) phases (Fig. 2). This is in apparent contrast to evidence from pollen assemblages that implies increased aridity in the region during glacials (30), from which one would expect increased evaporation-driven surface salinity. Moreover, paleoenvironmental and archaeological evidence from Australia indicates a transition from humid conditions between 50 to 40 ky to higher aridity between 40 to 30 ky (31). Our SSS record shows several maxima during this time span, but they all occur before a background of decreased SSS (Fig. 2). A viable modulator of regional surface salinity that operates independently from evaporation and precipitation is the hydrographic gradient associated with the circum-Antarctic ocean fronts that separate warmer, more saline surface waters to the north from colder, fresher waters to the south. Northward migration of the mobile fronts during glacial and stadial periods is a plausible means by which the core site was brought under the influence of fresher sub-antarctic surface waters (23). Sporadic meltwater surges from South Island, New Zealand, and melting Antarctic icebergs have contributed to the observed salinity anomalies on suborbital time scales. By applying a modern δ18Ow-salinity relation, we may overestimate the amplitude of salinity changes to some extent, because increased oxygen isotope fractionation during precipitation at low glacial temperatures and an influence of low-δ18Ow meltwaters from Antarctica and disintegrating New Zealand glaciers have likely altered the δ18Ow-salinity relation. We also note that the mean ocean δ18Ow record (32) that we used to correct δ18Oplk for global ice volume effects does not offer true millennial-scale resolution, which introduces uncertainty in the range of 0.2 to 0.3 salinity units (9). But these factors do not alter the overall picture of low SSS during cold periods. In fact, some of the minima along our SSS record are coeval with ice-rafted debris (IRD) events in the South Atlantic and on Campbell Plateau, south of Chatham Rise (Fig. 2) (33, 34), which in turn demonstrates the importance of these events for the wider circumAntarctic region.

Fig. 3.

δ18Oplk of core MD97-2120 and mean ocean δ18Ow (32) used to estimate local δ18Ow and SSS changes. (A) δ18Oplk record. (B) Mean ocean δ18Ow record (32) used to correct δ18Oplk for global ice volume changes. (C) Local δ18Ow and salinity anomalies as in Fig. 2, here for the past 340 ky. Blue and gray vertical bars mark events as described in Fig. 1. Bars along the bottom age scale refer to periods of enhanced IRD delivery on Campbell Plateau (CP), south of Chatham Rise (34), and in the South Atlantic (SA) (33); CP9 and CP10 coincide with low-SSS anomalies that are interrupted by short excursions to higher SSS values. The broad salinity minima at 96 to 84 ky and 125 to 110 ky coincide with meltwater surges between 80 and 130 ky as described in (34). Provenance studies on Campbell Plateau suggest an Antarctic origin for the IRD (34). A glacially intensified Southland Current flowing northeastward along South Island would have isolated Campbell Plateau from local iceberg drifts and meltwater surges from South Island. These may have reached Chatham Rise, thus causing recurrent low-SSS anomalies during glacial periods.

The high temporal resolution and fine structure of the δ18Oplk and SSTMg/Ca profiles of core MD97-2120 allow for a detailed comparison with Antarctic ice core climate records. On orbital time scales, our records directly mirror the structure of the Vostok δD record, which reflects air temperature changes over Antarctica (5) (Fig. 1). The similarities, for instance, hold for the interglacial peak heights (MIS 5e, 7e, and 9c), which in both records exceed Holocene levels. Temperature amplitudes across Terminations I to III are higher in the Vostok record (9.5°, 12.3°, and 9.3°C) than in SSTMg/Ca (8.8°, 8.9°, and 7.2°C), which indicates strong Antarctic cooling during glacials in connection with increased thermal isolation due to enhanced circum-Antarctic circulation (35). During Termination I, δ18Oplk and SSTMg/Ca display a brief reversal of 0.6 per mil (‰) and 1.1°C, respectively, that is similar in structure and timing to the 1.9°C cooling during the Antarctic Cold Reversal in the Vostok and Byrd ice core records (5, 6) (Fig. 2). A similar short-lived δ18Oplk reversal is developed during Termination III (250 to 242 ky) that is likewise documented in the Vostok δD and argon records (5, 29) (Fig. 1). The mid–Termination III reversal is covered by five data points in our SSTMg/Ca record that show a transient halt in the warming trend but no coeval SSTMg/Ca decrease. This suggests that the δ18Oplk reversal reflects the incursion of a short-lived local salinity maximum (Fig. 3).

Dissimilarity exists with the Antarctic ice core record during glacial periods in that δ18Oplk shows enhanced variability with abrupt transitions that are not seen in the ice cores (Figs. 1 and 2). Some short-lived δ18Oplk and SSTMg/Ca changes are reminiscent of the glacial climatic oscillations recorded in Northern Hemisphere records (1, 2, 16) (Fig. 2). For instance, the sequence of δ18Oplk anomalies between 46 and 41 ky mimics the structure of Dansgaard/Oeschger (D/O) events 9 to 12 in the Greenland ice core record (Fig. 2). The pronounced SSTMg/Ca warm anomaly during MIS 6 (154 to 138 ky) reaches a maximum amplitude of 5.7°C and is mirrored by a coeval negative δ18Oplk anomaly of 1.8‰. A similar anomaly is documented in North Atlantic (36) and tropical Pacific (28) marine records, but is absent from the Vostok record. It coincides with an orbital precession minimum and hence a maximum in Northern Hemisphere insolation, suggesting northern forcing as the cause for the warm anomaly. The likewise pronounced SSTMg/Ca excursion of 4.4°C during MIS 4, between 68.3 and 65.6 ky, is mirrored by a 1.6‰ δ18Oplk decrease that is not developed in the benthic δ18O section of our core (9). This event on our age scale immediately follows the 70-ky warm anomaly in the Byrd ice core record but overlaps with D/O event 19 (69 to 66 ky) in Greenland (Fig. 2).

On glacial-interglacial time scales, our records show a varying degree of similarity with the paleoclimate record from Antarctic ice cores. Evidently, the degree of climatic coupling of mid-southern latitudes to Antarctica changed as a function of global climate, requiring a mechanism that perturbs this climatic link during cold intervals. Varying Antarctic Circumpolar Current (ACC) intensity, as an indirect response to variations in westerly wind strength (37), provides for such a mechanism in that the ACC modulates the transmission of climate signals across high southern latitudes (38). In response to stronger decoupling from Antarctic climate, the mid-latitude Southern Hemisphere was more likely to be influenced by climatic oscillations and abrupt transitions not felt in Antarctica. A significant increase in the amplitude of δ18Oplk during glacial periods of the past 340 ky becomes apparent in the δ18Oplk record, when variability ≥5 ky is removed from the data (39) (Fig. 1). This provides clear evidence that the magnitude of millennial-scale climatic oscillations is dependent on the state of the global climate, notably the presence of large ice sheets in the Northern Hemisphere (40).

Paleoclimatic records from various archives in the Southern Hemisphere have been used to infer both interhemispheric synchrony and asynchrony (6, 27, 4143). However, radiocarbon dating and the uncertainty related to variations of local reservoir ages (9, 13), as well as graphical tuning of the records to other paleorecords, do not offer the control needed for robust assessment of interhemispheric temporal patterns of climate change. For instance, we can fit parts of the glacial sections of our δ18Oplk record to the Greenland δ18O record without violating the graphical correlation between the benthic δ18O sections of our core and core MD95-2042 from the northeast Atlantic (16). Thus, our age model is consistent with, but cannot prove, a direct mechanistic connection and synchrony between the short-lived excursions in our δ18Oplk record and rapid Northern Hemisphere D/O events.

A notable implication from our study is that extrapolar climates in the Southern Hemisphere were more variable than is inferred on the basis of Antarctic ice cores alone. Intensified circumpolar ocean and atmospheric circulation during glacials increasingly isolated Antarctica from extrapolar climates, which therefore are not adequately represented in Antarctic paleoclimate records. The similarity of our δ18Oplk record with D/O cycles in Greenland that is evident in some glacial sections is suggestive of direct climatic linking and synchrony between the hemispheres.

Supporting Online Material

Materials and Methods

Figs. S1 to S3

Tables S1 and S2


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article