Penultimate Deglacial Sea-Level Timing from Uranium/Thorium Dating of Tahitian Corals

See allHide authors and affiliations

Science  29 May 2009:
Vol. 324, Issue 5931, pp. 1186-1189
DOI: 10.1126/science.1168754


The timing of sea-level change provides important constraints on the mechanisms driving Earth’s climate between glacial and interglacial states. Fossil corals constrain the timing of past sea level by their suitability for dating and their growth position close to sea level. The coral-derived age for the last deglaciation is consistent with climate change forced by Northern Hemisphere summer insolation (NHI), but the timing of the penultimate deglaciation is more controversial. We found, by means of uranium/thorium dating of fossil corals, that sea level during the penultimate deglaciation had risen to ~85 meters below the present sea level by 137,000 years ago, and that it fluctuated on a millennial time scale during deglaciation. This indicates that the penultimate deglaciation occurred earlier with respect to NHI than the last deglacial, beginning when NHI was at a minimum.

Fossil corals are a valuable archive of past sea level, but the density of coral data is biased toward sea-level highstands because of the inaccessibility of fossil corals that grew during lower sea level and are now further submerged. Reconstruction of lower sea levels has relied on dredging and submersible sampling, occasional fortuitous finds in uplifted terraces (1, 2), and the challenging approach of coral-reef drilling. Such drilling, while technically demanding and expensive, has yielded valuable records of sea-level change for the last deglacial (3, 4) and more limited constraints on the onset of the last interglacial (5).

To target deeper and earlier portions of the sea-level curve, Integrated Ocean Drilling Program (IODP) Expedition 310 (known as the “Tahiti Sea Level” expedition) drilled submerged reefs in seawaters ranging from 41.7 to 117.5 m (6). The island of Tahiti Nui (French Polynesia) is located in the southern tropical Pacific and is distant from locations of glacial ice sheets. Sea-level change at Tahiti during deglaciation is therefore dominated by the addition of meltwater to the oceans rather than by the effects of ice mass redistribution and isostacy. Steady subsidence of 0.25 m per 1000 years (4), resulting from the load of the island on the underlying oceanic plate coupled with a location distant from ice loading, makes Tahiti an ideal site to reconstruct past sea levels. Material from before the Last Glacial Maximum was recovered at each of the three locations where Tahiti drilling was performed (Faaa, Maraa, and Tiarei) (6) (fig. S1) and seven separate cores have yielded pre-LGM corals suitable for U/Th dating from 113 to 147 m below sea level (mbsl).

Corals were screened for secondary calcite and aragonite by x-ray diffraction (XRD) and thin-section petrography. Of the 25 pre-LGM corals analyzed for U-Th isotopes (7), 12 had values of (234U/238U)i (234U/238U ratios corrected for decay since deposition) between 137 and 151 per mil (‰), which we take as a reasonable range on the basis of known variability of past seawater 234U/238U ratios during the glacial-interglacial cycle (5, 8). These 12 are considered pristine and are discussed further here; replicate measurements that differ significantly have been excluded from discussion (but are illustrated in Fig. 1B as small circles).

Fig. 1

Sea-level plot for MIS 3, the penultimate deglaciation, and MIS 6. (A) Age versus subsidence-corrected depth plot, showing in situ samples measured in this study as solid red circles. Paleo–water depth estimate of greater than 20 m is illustrated with a red bar; the dashed continuation illustrates the possibility that water depth may have been even greater. Subsidence rate used is 0.25 m per 1000 years for consistency with the previous coral record from Tahiti (4). Shown for comparison are ages versus reconstructed depth for corals from the Huon Peninsula [blue diamonds (9, 10, 30)]. Subsidence/uplift rates are illustrated for both Tahiti and a section of the Huon Peninsula by orange lines. Sea-level reconstructions from the Red Sea are shown in green (11) and purple (12). The deglacial coral sea-level record from Tahiti (4) is shown in red diamonds. (B) Subsidence/uplift-corrected coral elevations versus age, for the penultimate deglaciation. New data from this study are reported as solid red circles; small circles are corals that are interpreted to have undergone some alteration, whereas large circles are considered to have robust chronology. Red boxes represent paleodepth estimates based on fossil assemblages and lithologies (13); the dashed continuation illustrates the large possible depth range of these facies. Existing coral data (1, 2, 5, 3135) are shown as open symbols. A coral from Bard et al. (5) is shown (green-filled square), from which the timing of MIS 5e in Lisiecki and Raymo (18) is constrained. Two proposed sea-level curves are shown for the deglacial: The green line is a sea-level reconstruction based on an ice sheet model and high-latitude air temperature (15), but with the chronology adjusted to be 4500 years older on the basis of Tahiti coral data; the blue line is a RSL reconstruction from the Red Sea (16) with the chronology adjusted +2500 years. The gray dashed vertical lines are timing constraints of the deglaciation for the suggested start (142 ka) and the time by which sea level must have risen above 85 mbsl at Tahiti (137 ka).

Corals of marine isotope stage 3 (MIS 3) age, after a correction for subsidence [0.25 m per 1000 years (4)], occur at 105 to 130 mbsl with ages of 29,600 to 33,000 years ago (29.6 to 33.0 ka) and are considered to be in situ (table S1 and fig. S2). They have (234U/238U)i values that agree closely with one another, and there is no indication of alteration. These new coral data, once corrected for subsidence, are at substantially greater depth (by ~30 to 40 m) than corals of similar age from the Huon Peninsula (9, 10)—corrected for uplift—and are also at a similarly greater depth with respect to δ18O-based reconstructions of sea level from the Red Sea (11, 12) (Fig. 1A). This difference is too large to be accounted for by variability of the non-eustatic component of relative sea level (RSL) at far-field sites (distant from ice sheets), or error in the subsidence/uplift rates.

The most plausible explanation for the discrepancy between the new Tahitian coral data and sea-level reconstructions from the Red Sea and Huon Peninsula is that they grew in deeper water. The MIS 3 corals are part of an assemblage (encrusting Montipora and foliaceous Pachyseris, thinly encrusted by coralline algae, with the absence of infilling muds) suggesting a deeper forereef subfacies and indicative of paleo–water depths greater than 20 m (13). Such paleodepths are consistent with sea level remaining at ~80 mbsl until at least 29.6 ka, in agreement with records from the Huon Peninsula (9, 10) and Red Sea (11, 12).

Subsidence-corrected depths of two MIS 6 corals (310-M0009D-25R-2W-41,49 and 310-M0009D-25R-2W-52,55) are at 109 mbsl (Fig. 1B). The abundance of tabular Acropora and massive Porites, coupled with the presence of thick algal crusts on the upper surfaces of corals and their incorporation in a coarse sandy matrix, suggests a facies similar to “robust-branching coral” (13) and indicates a probable depth of 0 to 6 m for these samples. These samples therefore provide the first coral-based estimate of sea level during the fully glacial portion of MIS 6. From 153.4 (±0.5) to 152.7 (±0.7) ka, RSL at Tahiti was <109 mbsl and likely within 6 m of this (i.e., in the range 103 to 109 m). Glacial isostatic adjustment modeling has indicated that, at the LGM, RSL at Tahiti was 6 (±6) m lower than the purely eustatic ice volume equivalent sea level (ESL) (14). If ice loadings during MIS 6 and at the LGM were broadly similar, as seems a reasonable approximation, a similar discrepancy would have occurred and ESL would be 97 to 103 mbsl at 153 ka. It is likely that sea level fell slightly further after 153 ka toward the penultimate glacial maximum as suggested by δ18O (15) (Fig. 1B).

Two corals are from early in the penultimate deglacial. Replicate analyses of coral 310-M0019A-29R-1W-0,4 are in agreement: 136.9 (±0.9) and 137.8 (±0.4) ka. As with the MIS 3 and MIS 6 samples, this coral has no calcite, no aragonite overgrowths, and no signs of visible alteration. The two analyses have (234U/238U)i values within error (2 standard deviations) of one another and within the expected seawater range. The age of this coral indicates that RSL must have risen to <85 mbsl by 137 ka. Another coral (310-M0019A-27R-1W-62,83) has a similar uplift-corrected depth but younger replicate ages of 133.1, 133.2, and 134.0 ka (Fig. 1B). This may reflect drowning of the reef after 137 ka, with accretion unable to keep up with rising sea level across the deglaciation. Alternatively, sea level may have risen to an early highstand shortly after 137 ka and then fallen back to within 20 m of the 310-M0019A-27R-1W-62,83 sample by 133 ka. Such a millennial-scale sea-level fluctuation has been suggested previously on the basis of Huon Peninsula corals (2) and Red Sea δ18O (16). This interpretation is supported by the lithology and fossil assemblage of the M00019A cores. Sample 310-M0019A-29R-1W-0,4 is from a unit containing a coralgal framework dominated by Acropora, suggestive of shallow water depth, whereas 310-M0019A-27R-1W-62,83 is from a framework of massive Porites indicating a water depth from 0 to 25 m (13) (fig. S3). Although no U/Th ages were successful in the core section between these two dated samples, this section contains a coralgal framework of encrusting Porites that suggests development in deeper water during an early sea-level highstand. This sedimentological evidence suggests a reversal of sea level during the penultimate deglacial, in agreement with earlier studies (2, 16). The magnitude of this early highstand—about two-thirds of the total deglaciation—makes it substantially larger than millennial-scale variability such as that seen during MIS 3 (17).

The Tahiti coral data alone indicate that global ice volume must have reduced markedly between 153 and 137 ka. A more detailed reconstruction of the deglaciation may be constructed by combining continuous sea-level curves with the chronological constraints from the coral data. When comparing different localities it is important to consider that there may be regional differences in RSL, and between RSL and ESL, that strictly indicate deglaciation. RSL at far-field sites such as Tahiti provides a reasonable approximation of ESL, although the isostatic component is important at the start and end of deglaciations. RSL at Tahiti is expected to be lower by 0 to 15 m during the early deglacial and higher than ESL by 1.5 to 3.5 m during the early interglacial highstand (14). Tahiti RSL during the early deglacial may therefore lag ESL slightly.

To assess the magnitude and duration of the penultimate deglacial sea-level change, we used the sea-level reconstruction of Bintanja et al. (15) (based on modeling of the ice volume component of deep-sea benthic δ18O) and draped this over our far-field coral data (Fig. 1B), maintaining the duration of the deglaciation and altering only the timing. The original time scale of the Bintanja et al. record is based on the Lisiecki and Raymo (18) time scale, so comparison of our timing of the deglaciation also provides some assessment of the error in that time scale. The Bintanja et al. reconstruction uses a 3000-year mean of the input data, so any millennial-scale structure to the deglaciation is not represented. For consideration of millennial-scale variability, we compared the coral data to modeled sea level from δ18O of planktonic foraminifera from the Red Sea (16), although that record does not extend to the glacial maximum.

Adjusting the Bintanja et al. time scale to be consistent with Tahiti corals requires a shift to ages that are 4500 years older. This suggests that the midpoint (the time by which sea level rose to half of the total glacial-interglacial rise) of the penultimate deglaciation occurred at ~136 ka, consistent with the previous assessment from some highstand corals (1, 19) and from U/Th dating of Bahamas sediments (20). Our data also require the early deglacial portion of the Red Sea sea-level curve (16), which has a chronology based on the highstand age in an “open-system” coral compilation of Thompson and Goldstein (21), to be ~2500 years earlier. The timing presented here is also slightly earlier than the North Atlantic benthic δ18O change placed on a chronology based on correlation of ice-rafted debris events with weak Asian monsoon intervals, precisely dated in Chinese speleothems (Fig. 2) (22).

Fig. 2

The timing of the penultimate deglacial, illustrated with the data of Bintanja et al. (15) (in green), with the chronology adjusted (+4500 years) to match new coral data of this study (Fig. 1B), and the Red Sea record of Siddall et al. (16) (in dark blue), adjusted by +2500 years. The Dome Fuji ice core δ18O (23) and North Atlantic bottom water δ18O (22) records are shown for comparison in gray and purple, respectively. Local summer insolation is shown for 65°N (light blue) and 65°S (red) (36). The gray dashed vertical lines are the timing constraints of the deglaciation from Fig. 1B. Note that deglaciation must start when NHI is at or close to a minimum.

The sea-level rise at the penultimate deglacial can also be compared with the accompanying rise in atmospheric CO2 on the independent chronology of Kawamura et al. (23) for the Dome Fuji ice core record based on matching N2/O2 variations to local insolation at the ice core site (Fig. 3). This comparison indicates no resolvable difference in timing between sea level and CO2. This is in disagreement with a sea-level lag of 4000 years previously inferred with the use of ice-core atmospheric δ18O as a lagged response to sea-level change (24). This discrepancy suggests that atmospheric δ18O is not a reliable indicator of sea level, and that the 1‰ shift seen at the deglaciation (although of similar size to that occurring in seawater during deglaciation) is controlled by changes in the Dole effect across the deglacial (25). The apparent synchroneity of sea level and CO2 change at this deglacial [and the slight sea-level lag during the last deglaciation (3, 23)] means that mechanisms involving sea-level changes as drivers of CO2 change are no longer falsified by timing constraints, as had previously been suggested (24).

Fig. 3

Timing of the increase of atmospheric CO2 and decrease of atmospheric δ18O compared to sea-level rise across the penultimate deglaciation. Sea-level rise is illustrated by the curves of Bintanja et al. (15) and Siddall et al. (16), adjusted to fit the new coral constraints on this study (Fig. 1B). The timing of CO2 rise is based on the Dome Fuji record, on the time scale of Kawamura et al. (23). The uncertainty of this time scale, shown by the horizontal bar, is ±2000 years at 134 ka (23). The Vostok CO2 time series is aligned to this Dome Fuji record to provide the timing of atmospheric δ18O (involving a shift of 1600 years at the midpoint of the deglaciation relative to the Vostok glaciological time scale, GT4) (37). Note the clear lag between sea level and atmospheric δ18O. The gray dashed vertical lines are the timing constraints of the deglaciation from Fig. 1B.

Determining the start of sea-level rise is useful to elucidate the cause of the deglaciation, but is harder to define than the midpoint of deglaciation. The coral age at 137 ka constrains the start of the deglaciation to be at least that old. A more gradual start to the penultimate deglaciation, as is suggested by the Bintanja et al. curve, would place the onset of ice sheet collapse at ~142 ka (Fig. 1B).

Early orbitally tuned chronologies (26, 27) followed the suggestion of Milankovitch (28) and assumed that the rate of ice sheet collapse should be greatest when Northern Hemisphere summer insolation (NHI) was at a maximum because of enhanced summer melting. Depending on the precise month used to define the peak of summer, this suggests a maximum of melting rate close to 129 ka. Tahiti corals indicate that sea level was clearly rising before this at the penultimate deglacial, indicating that orbitally tuned chronologies (26) may be incorrect by up to 7000 years. It is clear that the phasing of the penultimate deglacial was very different from that of the last deglacial, and that tuning approaches assuming a constant phasing are inappropriate if millennial-scale accuracy is required.

Melting of the penultimate deglaciation started during a minimum of NHI (Fig. 2). This requires the climate system to be unusually poised for deglaciation if it were driven by NHI, with the subsequent rise in NHI being responsible for pacing the rate of melting throughout the deglaciation (somewhat at odds with the millennial variability seen during the deglaciation). It is interesting that the initiation of both this and the last deglaciation occurred during Southern Hemisphere summer insolation maxima, albeit with slightly different phasing, suggesting a possible role for the Southern Hemisphere. The different phasing of this deglaciation relative to the last one, however, indicates a more complex relationship between insolation and deglaciation, perhaps involving a stochastic response to insolation (29) or control by more than a single season and latitude. Any mechanism proposed to link insolation to orbital–time scale climate change must be tested against the difference in phasing of the two most recent deglaciations, as constrained by U-Th dating of corals.

Supporting Online Material

Materials and Methods

Figs. S1 to S3

Table S1


References and Notes

  1. See supporting material on Science Online.
  2. Supported by UK Natural Environment Research Council grant NE/D001250/1 and the Gary Comer Abrupt Climate Change Fellowship (A.L.T.). We thank the Expedition 310 scientists for the use of samples and data provided by the IODP, and R. Pearce and R. Williams at the National Oceanography Center Southampton for their assistance with XRD measurements.
View Abstract

Stay Connected to Science

Navigate This Article