Main

It is well established that past rates of sea-level rise due to ice-volume reduction have reached up to 5 m per century23,24,25. However, such values relate to deglaciations, dominated by disintegration of the now-absent Laurentide ice sheet, which questions their suitability for projections of future sea-level change within a well-developed interglacial period. So far, no detailed information exists about the rates of sea-level change associated with fluctuations within interglacial periods in general, and above 0 m in particular. This focuses attention on MIS-5e, the most recent (best dated) interglacial period during which sea level stood several metres above 0 m between roughly 124 and 119 kyr (thousand years ago) (see the Supplementary Information).

MIS-5e warmth was caused by orbital forcing of insolation, rather than the predicted greenhouse forcing of the near future, so that MIS-5e ice-volume responses may have differed in detail from future responses. However, we do not consider the MIS-5e warming as a straight analogy to the future, but instead aim to provide an observational context that quantifies the potential range of sea-level change rates above 0 m, to inform the debate about ice-volume reduction/sea-level rise in the next century that currently relies entirely on theoretical projections21,22.

By dating fossil corals, previous studies have established that MIS-5e sea level reached an average highstand around +4 (±2) m, with individual maxima up to +7 or +9 m (refs 2–13) (see the Supplementary Information). Up to 5 m of this signal derived from reduction of the Greenland ice sheet14. There may also have been contributions from (West) Antarctica, which is responsive to climate change26, although models suggest a slow and gradual response22.

Although dated corals yield impressive insight into sea-level position and absolute age, they lack the tightly constrained relative-age framework needed to calculate rates of change. We overcome that problem by combining coral data with a new, continuous, highly resolved sea-level reconstruction through MIS-5e from the recent calibration method for central Red Sea stable oxygen isotope (δ18O) records15,16,17,18, which offers tight stratigraphic control of relative ages.

Figure 1a shows new high-resolution δ18O data from the MIS-6–5e deglaciation15 through MIS-5e in central Red Sea sediment core GeoTü-KL11, superimposed on the previously published low-resolution record of KL11 for the entire MIS-6–4 interval17,27 (see the Methods section). Also shown is a new 0.5–2-cm-resolution δ18O record through MIS-5 from nearby core GeoTü-KL09, superimposed on a whole-core 10-cm-resolution record for that core (see the Methods section). Throughout this letter, records from cores KL11 and KL09 are correlated in the depth domain on the basis of simple linear matching between MIS-4 and the onset of MIS-5e, without any subjective ‘fine-tuning’ (Fig. 1a–d).

Figure 1: Stable isotope and derived sea-level records for central Red Sea cores KL11 and KL09.
figure 1

a, δ18Oruber throughout MIS-5e–4 for KL11 (black filled circles, line), alongside high-resolution data (red filled circles, grey shading) and a low-resolution series (red open circles) for KL09 (see the Methods section; VPDB: Vienna PeeDee Belemnite). Black bars indicate amplitudes of δ18O change equivalent to about 2 ∘C temperature change and 10 m of sea-level change. The numbers identify Marine Isotope Stages. b, As a, but for sea level as derived from δ18Oruber (refs 1718). c, As a, but magnified for a detailed comparison of trends within MIS-5e. d, As c, but for derived sea level.

Full size image

Figure 1b,d shows the δ18O data converted into sea-level estimates, using the calibration developed for KL11 (1σ=6 m) (see the Methods section)15,17,18. A first important observation is that sea level according to KL09 reached a shallowest value of about −17 m during MIS-5c, compared with about −40 m (or −30 m on the basis of low-resolution results of KL11) during MIS-5a (Fig. 1b). The MIS-5c value agrees with Barbados coral terraces, which suggest that sea level reached up to −16 m during both periods28, but the MIS-5a value seems low relative to Barbados. We have yet to fully resolve MIS-5a, and consider our MIS-5c estimate more definitive than the MIS-5a value.

Figure 2 concentrates on MIS-5e in both KL11 and KL09, compared with a detailed series of coral-based sea-level estimates from Barbados4. The Red Sea results are shown versus depth in each core, and the KL11 series was calibrated to age by linearly matching the ‘inverted U’ shape of its long-term average sea-level trend to that in the dated corals4 (Fig. 2a). Within MIS-5e, we maintain our straightforward linear KL11–KL09 correlation (Fig. 1a), so that the correlation of KL11 to the dated corals (Fig. 2a) implicitly yields an approximate age model for KL09 (Fig. 2b).

Figure 2: MIS-5e high-resolution Red Sea sea-level reconstruction for KL11 and KL09 versus coral data.
figure 2

KL11 and KL09: black lines/filled circles, 1σ=6 m error bars17,18; coral data: orange squares, with age/altitude error bars4. a, Long-term average trends: 3,000 year gaussian filter through KL11 (thick blue line); long-term trend in coral data (dashed red line). b, As a, for KL09. c, Short-term sea-level trends, from a 750 year gaussian filter through KL11 (thick blue line). Dashed blue lines show statistical 1 standard error for this record. The first time derivative of the thick blue record yields rates of change (purple step plot, with maximum rates of rise (positive) and lowering (negative)). d, As c, for KL09.

Full size image

A 3,000 yr gaussian filter through KL11 reveals a long-term mean MIS-5e highstand up to +6 m, in agreement with the coral data (Fig. 2a). The same filter through KL09 reveals a comparable pattern (Fig. 2b) (see the Methods section). Shorter-term fluctuations are evaluated using a 750 yr filter through both records (Fig. 2c,d). Here, the two Red Sea records differ: the distinct fluctuations seen in KL11 are poorly/not represented in KL09 (Fig. 2c,d). Although this apparent discrepancy probably reflects different impacts of bioturbation in the two cores (see the Methods section), it does highlight a need for qualification of our results in terms of ‘robustness’. Accordingly, we first identify a detailed record of sea-level oscillations within MIS-5e on the basis of KL11, then investigate to what extent this record is supported by previous work, and finally identify which of the inferred features are most robust in that they are also obvious in KL09.

The 1σ=6 m uncertainty in the Red Sea sea-level method includes a ±1 ∘C uncertainty in sea surface temperature (SST) and a generous uncertainty for net evaporation17,18. To minimize the influence of any bias in individual points, we infer MIS-5e sea-level fluctuations after smoothing the data with a moving 750 yr gaussian filter (Fig. 2c). With 1σ=6 m for individual points, the standard error of the filtered mean sea-level record in KL11 is determined in a statistical sense for this particular data set (Monte Carlo method) at 1 standard error=3.5 m (Fig. 2c). This yields the first sea-level record with the essential attributes (high temporal resolution and stratigraphic continuity) for quantification of rates of change associated with millennial-scale oscillations within MIS-5e. First, however, we consider whether there is support from other, independent sea-level indicators for the inferred oscillations (Fig. 2c). Ages reported for the Red Sea record are subject to the about ±1 kyr uncertainty margin of the coral datings that underlie our age model4 (Fig. 2a). However, the tightly constrained stratigraphic order of our records stipulates that such age uncertainties will be systematically distributed along the record, so that age uncertainties are much smaller than ±1 kyr where relative sample ages (hence, durations) are concerned.

Although some studies of MIS-5e corals favour a single highstand rather than a compound structure2,3,13, there is a hiatus in the Western Australian record3 that may agree with suggestions of intra-MIS-5e sea-level instability around 122±1 kyr in fossil reef records from Barbados4, the Bahamas3,8,9 and the Red Sea5,6,7,11,12 (see Supplementary Information, Table S1). KL11 suggests two main highstands around 123 and 121.5 kyr, separated by a relative lowstand around 122.5 kyr (Fig. 2c). A third minor highstand is suggested around 119.5 kyr, after a relative lowstand centred on 120.5 kyr. The amplitude of the latter fluctuation, as well as that of the sea-level drop following the 119.5 kyr high, is well supported by the coral data4 (Fig. 2c). The same holds for the early MIS-5e rise into the first highstand of 123 kyr (Fig. 2c), and it was previously found that detailed sea-level fluctuations inferred from the Red Sea record for the MIS-6 to 5e deglaciation agree with independent coral and speleothem data15. Furthermore, we note that the coral-based sea-level estimates4 may be suggestive of a sea-level fluctuation of similar amplitude to our inferred sea-level drop around 122.5 kyr (Fig. 2c). The second MIS-5e highstand inferred from the Red Sea record (∼121.5 kyr) resides within a conspicuous gap in the Barbados data4, but similar work around the Gulf of Aqaba reports terraces up to +10 m that date to 121–122 kyr (ref. 29).

Despite some apparent agreements, the lack of temporal resolution and stratigraphic continuity clearly limits the use of dated coral series for comprehensive validation of our record’s finer structure. In addition, the validity of detailed time-series comparisons (as above) might be questioned, given the dating uncertainties. Finally, it has been suggested that rapid sea-level fluctuations may not be represented by well-developed corals30. Detailed stratigraphic descriptions of reef/coastal architecture sequences may help, as they yield better insight into the temporal sequence of events.

Such work along the Egyptian coast of the northern Red Sea reveals two main highstand peaks (around +5 to +9 m), separated by a brief 7–10 m sea-level drop5,6,7 (see Supplementary Information, Fig. S1). The absolute altitude of the deposits might be affected by uplift, but due to its gradual/long-term nature, uplift cannot account for the centennial-scale relative sequence of two peaks separated by a sharp drop, or for the amplitude of the fluctuations. Note that Red Sea sea-level records based on fossil reefs and coastal deposits5,6,7,10,11,12 are conceptually, technically and tectonically independent from our reconstructions. The reef/coastal sequences5,6,7 thus offer strong independent support to the main intra-MIS-5e sea-level fluctuation suggested by our record, with two main highstands (123 and 121.5 kyr) separated by a sharp drop (122.5 kyr) (Fig. 2c). Even the inferred amplitudes of change agree.

Reef-architecture studies on the Eritrean coast of the southern Red Sea11,12 seem to corroborate not only the two main highstands in our record, but even the minor third peak11 (Fig. 2c). Although only qualitatively described, the work identifies an initial rapid sea-level rise (their subunit 5e1), a drop, a second sea-level rise (5e2), another minor drop, and finally a minor peak with patch reef development (5e3) before the final drop at the end of MIS-5e11 (see Supplementary Information, Fig. S2). The various reef/coastal sequence studies describe the intra-MIS-5e sea-level drop(s) as erosive/consolidation phases without apparent new reef formation5,6,7,11. Such transient events would probably be missed in time series of dated corals.

Both the structure and amplitude of our KL11 record of MIS-5e sea level (Fig. 2c) are therefore empirically validated (within error margins) by four main observations. These are: the general consensus that sea level reached a longer-term mean position up to +6 m; the observations of individual (peak) deposits up to about +9 m; the suggestions of intra-MIS-5e sea-level variability in a variety of records; and the agreement between our reconstruction and detailed reef/coastal architecture studies regarding the sequence of events within MIS-5e. Moreover, our reconstruction is robust with respect to sedimentation rate within KL11, isostatic effects and regional fluctuations in temperature and net evaporation (see the Supplementary Information).

We now use our record to estimate the rates of MIS-5e sea-level rise. The first time derivative of the short-term smoothed KL11 record shows peak rates of rise of 2.5, 2.0 and 1.1 m per century (Fig. 2c). Mean coral values suggest comparable rates of rise of 1.7 and 1.0 m per century through the intervals 124.7–123.3 and 120.5–119.8 kyr, respectively (Fig. 2). Because overall chronology is critical to the derived rates of sea-level rise, we derive a representative maximum duration for the MIS-5e highstand above 0 m from the literature (see Supplementary Information, Table S1). On the basis of an age range of about 119–128 kyr, this maximum duration is about twice that indicated in Fig. 2. This conservative way of assessing uncertainty in the rates of sea-level rise yields the very minimum estimates for the rates of MIS-5e sea-level rise, being 1.3, 1.0 and 0.6 m per century. We thus infer a full potential range for the rates of rise between 2.5 and 0.6 m per century. We also note that sea-level lowering events reveal a remarkably consistent rate between −1.3 and −1.8 (−0.7 to −0.9) m per century (values in brackets refer to a doubled MIS-5e duration).

Although the intra-MIS-5e oscillations in KL11 seem to be corroborated by coral and reef architecture data, we consider them as only strongly suggestive, because they are not fully replicated in KL09 (see the Methods section). However, two events are common to both records and consequently seem particularly robust (Fig. 2c,d). These are the rise above 0 m at the onset of MIS-5e around 123.5 kyr and the drop ending the main highstand around 119 kyr. The rate of rise above 0 m around 123.5 kyr is 1.5–2.5 (0.8–1.3) m per century, whereas the rate of lowering around 119 kyr is −1.3 to −1.6 (−0.7 to −0.8) m per century. On the basis of only these robust events, therefore, the rate of rise above 0 m from the Red Sea data would be 1.6±0.8, similar to the 1.6±1.0 m per century estimate based on all MIS-5e fluctuations in KL11.

A 1.6 m global sea-level rise per century would correspond to disappearance of an ice sheet the size of Greenland in roughly four centuries (modelling suggests 1,000 years or more20). During MIS-5e, such rates of sea-level rise occurred when the global mean temperature was 2 ∘C higher than today, as expected again by AD 2100 (refs 1922). Using MIS-5e to gain insight into the potential rates of sea-level rise due to further ice-volume reduction in a warming world, our data provide an observational context that underscores the plausibility of recent, unconventionally high, projections of 1.0±0.5 m sea-level rise by AD 2100 (ref. 21).

Methods

We present stable isotope data for the surface-water dwelling planktonic foraminiferal species Globigerinoides ruber (white) in two central Red Sea sediment cores: GeoTü-KL11 (18∘ 44.5′ N, 39∘ 20.6′ E, 825 m waterdepth) and GeoTü-KL09 (19∘ 57.6′ N, 38∘ 08.3′ E, 814 m waterdepth), which we refer to in abbreviated form as KL11 and KL09. Samples for KL11 were prepared in Tübingen and analysed in Kiel, Germany, on a MAT251 mass spectrometer, over the period 1988–1996. For KL09, a low-resolution pilot series of samples was prepared in Tübingen and analysed in Southampton on a Europa Geo2020 mass spectrometer, in 2006. The high-resolution series for KL09 was prepared and analysed in 2006–2007 in Southampton, using the Geo2020. Each analysis represents a set of at least 10 and usually more than 20 specimens, and external precision (1σ) is 0.06%. Despite the different equipment, lab protocols, operators and sampling and preparation methods in the two separate sediment cores, the records of KL09 and KL11 agree well, especially when allowing for the external precision (Fig. 1a).

Calibration of δ18Oruber variations in KL11 into sea-level change is described in detail in refs 1518. The scaling used in that method infers a regional mean annual MIS-5e SST that is up to 0.5 ∘C warmer than the present (±1 ∘C at the 1σ level), which is within the range suggested by alkenone-based reconstructions for the nearby easternmost Mediterranean31. An extensive pilot study of MIS-5e SST in KL09 is discussed in the Supplementary Information. Calibration of KL09 δ18Oruber to sea level was done using the calibration method developed for KL11 (refs 1518). It may be expected, however, that absolute δ18Oruber values at different sites may be offset from one another (for example, as a function of latitude within the basin16,18 or temperature differences between sites). We use the mean long-term sea-level trend in KL09 in comparison with the coral data to compensate for the potential artefact that was introduced when calibrating KL09 data using the KL11-specific calibration between δ18Oruber and sea level. A reasonable fit is found when shifting inferred KL09 sea-level values down by about 6 m (Fig. 2b, dashed arrow). As the 1σ margin of the sea-level calibration is about 6 m (refs 1718), this inferred offset remains comfortably within the overlapping 1σ bands around the KL11 and KL09 sea-level reconstructions. In terms of δ18O, the inferred 6 m offset would imply that KL11 δ18Oruber is systematically about 0.2% heavier than contemporaneous KL09 δ18Oruber, which would be equivalent to less than 1 ∘C temperature difference between the sites.

We attribute the difference between the shorter-term fluctuations apparent within MIS-5e in KL11 and KL09 (Fig. 2c,d) to different impacts of bioturbation in the two cores. Sediment accumulation rates through MIS-5e are about 1.5 times higher in KL11 than in KL09 (compare depth axes in Figs 1, 2). As KL11 was sampled in 1 cm steps and KL09 in 0.5 cm steps, the resultant temporal resolution might seem better in KL09 than in KL11 (Fig. 2), but the low-pass filtering effect of bioturbation during sediment accumulation would cause considerable smoothing of high-frequency signals in lower accumulation settings (KL09), whereas these may be preserved under higher accumulation conditions (KL11).