Deep-Sea Temperature and Ice Volume Changes Across the Pliocene-Pleistocene Climate Transitions

See allHide authors and affiliations

Science  17 Jul 2009:
Vol. 325, Issue 5938, pp. 306-310
DOI: 10.1126/science.1169938


Earth has undergone profound changes since the late Pliocene, which led to the development [~2.7 million years ago (Ma)] and intensification (~0.9 Ma) of large-scale Northern Hemisphere ice sheets, recorded as transitions in the benthic foraminiferal oxygen isotope (δ18Ob) record. Here we present an orbitally resolved record of deep ocean temperature derived from benthic foraminiferal magnesium/calcium ratios from the North Atlantic, which shows that temperature variations are a substantial portion of the global δ18Ob signal. The record shows two distinct cooling events associated with the late Pliocene (LPT, 2.5 to 3 Ma) and mid-Pleistocene (MPT, 1.2 to 0.85 Ma) climate transitions. Whereas the LPT increase in ice volume is attributed directly to global cooling, the shift to 100,000-year cycles at the MPT is more likely to be a response to an additional change in ice-sheet dynamics.

Two pronounced shifts in the composition of the global benthic foraminifera isotope record (δ18Ob) mark the intensification of Northern Hemisphere glaciation (NHG) during the period from the mid-Pliocene to the late Pleistocene (1). The first, starting at ~3.2 to 2.7 million years ago (Ma), marks the late Pliocene transition (LPT) from warm, relatively ice-free conditions to a colder and more glaciated climate in the NH, manifested by increased amplitude of the 41-thousand-year (ky) obliquity cycles in the δ18Ob record (Fig. 1). The onset of large NH glacial/interglacial cycles is marked by the appearance of ice-rafted debris (IRD) in North Atlantic and North Pacific sediments at 2.7 Ma (2, 3), associated with increased stratification of North Pacific surface water (4). It has been debated, however, whether these singular events are linked to a threshold response to a shift in background climate state, driven by a decrease in atmospheric carbon dioxide concentration [a decrease in the partial pressure of CO2 (Pco2) (5)], or just a regional expression of a gradual global cooling as suggested by tropical and subtropical records (6, 7). A second shift from ~1.2 to 0.7 Ma, termed the mid-Pleistocene transition (MPT), marks the transition in the dominant periodicity from low-amplitude 41-ky cycles to large-amplitude 100-ky cycles. Mechanisms proposed to explain the MPT include a threshold response to deep ocean cooling and changes in sea-ice growth (8), hypothetically caused by a secular decrease in atmospheric Pco2 (9). An alternative mechanism known as the regolith hypothesis maintains that the exposure of basal rocks under the ice sheets, after long and intense erosion, enabled the buildup of taller ice sheets, independent of changes in background climate (10).

Fig. 1

Down-core records from western North Atlantic DSDP site 607. (A) Benthic oxygen isotope record [in ‰, using the Pee Dee belemnite (PDB) standard] from (13) and (14). (B) Evolutionary spectra of BWT between 500 and 1500 ka, showing that the BWT record is dominated by 41-ky cycles before the MPT and by 100-ky cycles after the transition. (C) Benthic foraminiferal Mg/Ca record, derived from C. wuellerstorfi and O. umbonatus, converted to BWT with the equation Mg/Ca = 0.15 × BWT + 1.16. (D) Benthic carbon isotope record (δ18Ob ‰ PDB) from (13) and (14). (E) SST record from DSDP site 607 based on census counts from (13). The MPT and LPT are highlighted by shading. Black lines represent average signals for each variable, determined by applying a Gaussian filter with a cutoff frequency of 400 ky.

Choosing between the different mechanisms hinges, however, on our ability to separate the δ18Ob record into its temperature and ice volume components. To date, there is only one orbitally resolved reconstruction of Pliocene-Pleistocene variability in deep ocean temperature (11). The study, based on a ostracode Mg/Ca record from the deep North Atlantic, shows an increase in glacial-interglacial peak-to-peak amplitude [∆(G-I)] from 3.6°C during the late Pliocene 41-ky glaciations to 4.5°C during the mid- to late Pleistocene 100-ky glaciations, driven mainly by a decrease in glacial temperatures. Because the ostracode record of bottom-water temperature (BWT) covers only the late Pliocene (2.3 to 3.2 Ma) and late Pleistocene [0 to 220 thousand years ago (ka)] intervals, a comprehensive assessment of the evolution of deep ocean temperature, its relationship to NHG, and the transition to the 100-ky G-I cycles has been difficult.

We determined BWT in the deep North Atlantic from Mg/Ca ratios in the benthic foraminifera Cibicidoides wuellerstorfi and Oridorsalis umbonatus. We used this record to deconvolve changes in the oxygen isotopic composition of seawater (δω) and estimate ice volume variations as related to changes in deep ocean temperature. The 3.2-million-year (My) record was derived from Deep Sea Drilling Project (DSDP) site 607 (41°N, 32°W; water depth 3427 m) and supplemented with measurements from a nearby piston core (Chain 82-24-23PC; 43°N, 31°W; water depth 3406 m) (Fig. 1) (12). Both sites are currently bathed in North Atlantic Deep Water [NADW, temperature (T) = 2.6°C, salinity = 35] and are thus linked to the hydrographic conditions at the high latitudes (where NADW forms), which are under the direct influence of NH ice sheets. The location and relatively high sedimentation rate at site 607 make it an ideal site for evaluating the role of the deep ocean in Pliocene-Pleistocene climate evolution at orbital-scale resolution (3-ky resolution) (13, 14).

We converted Mg/Ca ratios to paleotemperature using a regional Mg/Ca-temperature calibration (Mg/Ca = 0.15 × BWT + 1.16; fig. S2) that accounts for both temperature and carbonate saturation effects on foraminiferal Mg/Ca (15, 16). It has recently been suggested that the ~0.2 mmol/mol Holocene (HL)–Last Glacial Maximum (LGM) change in Mg/Ca observed in another deep sea core north of site 607 (BOFS 5K) was largely (>70%) driven by a ~20 μmol/kg decrease in CO3 concentration [CO3] (17). In contrast, mid- to late Pleistocene Mg/Ca ∆(G-I) at site 607 varies between 0.4 and 0.6 mmol/mol, and [CO3] variations may account for only 30 to 40% of the signal. The estimated error of BWT estimation is ±1.1°C (12).

Our temperature record suggests that climate cooling over the past 3.2 My occurred primarily through two distinct events associated with the LPT and MPT shifts in the global δ18Ob record. BWT variations generally covary and are coherent with the δ18Ob record in frequency, average trend, and G-I amplitude (Fig. 1). Across the LPT (2.5 to 3.0 Ma), average BWT decreases from 4.5° to 2.5°C, whereas average δ18Ob increases by 0.71 per mil (‰) (Table 1). In contrast with the ostracode record (11), we showed that the decrease occurs in both interglacial and glacial temperatures. Spectral analysis of the BWT record from 1000 to 2500 ka shows a dominant 41-ky peak, coherent with the δ18Ob record, signaling the development of obliquity-paced G-I temperature variability after the LPT (fig. S5) (12).

Table 1

Statistical summary of the primary features of the Mg/Ca-BWT and δω record, specifically the mean (M), glacial (G), and interglacial (I) trends and glacial-interglacial ∆(G-I) amplitude from different time intervals.

View this table:

Across the MPT (1150 to 850 ka), average BWT decreases from 2.5° to 1.2°C, whereas average δ18Ob increases by 0.39‰. As during the LPT, the MPT cooling occurs in both glacial and interglacial temperatures. An evolution spectrum reveals that the 100-ky frequency starts to appear as a broad band around 800 ka but becomes the dominant frequency only at the culmination of the cooling trend, around 700 ka (Fig. 1B). The fact that the cooling and ∆(G-I) amplitude increase precede the frequency shift suggests that either Earth’s climate passed a critical threshold or that cooling might not solely have driven the MPT. Subsequently, G-I BWT variability resembles the asymmetric sawtooth pattern of the δ18Ob record, with gradual cooling into glacial periods followed by abrupt warming during deglaciations.

The BWT record demonstrates that cooling of the deep ocean is a substantial portion of the increase in δ18Ob at both transitions. For the average long-term temperature trend, cooling accounts for ~70 ± 5% of the positive δ18Ob shifts across both transitions. During the LPT, BWT ∆(G-I) increases from 2° to 3.3°, whereas δ18Ob ∆(G-I) increases from 0.6 to 1.2‰, suggesting that on orbital scales, the contribution of temperature to δ18Ob variability is also ~70 ± 5%. During the MPT, BWT ∆(G-I) increases from 3.3° to 3.9°C, whereas δ18Ob ∆(G-I) increases from 1.2 to 1.9‰, indicating that temperature accounts for only ~50% of the δ18Ob ∆(G-I) variability. The decrease in the BWT/ice ratio suggests an increase in the sensitivity of NH ice sheets to changes in global temperature (i.e., more ice per degree of cooling) during the mid- to late Pleistocene.

Cross-spectral analysis between δ18Ob and BWT reveals that BWT leads δ18Ob during the late Pliocene to early Pleistocene by 3 ± 2 ky in the obliquity band and by 11 ± 5 ky during the mid- to late Pleistocene in the eccentricity band (Fig. 2). Around 700 ka, the lead increases in association with the appearance of a narrow band of 100-ky variability. The BWT lead in the mid–late Pleistocene is in agreement with other deep ocean temperature studies (11, 15). The phase relationship between BWT and δ18Ob represents the time constant of ice sheet response to changes in insolation (18). The increase in the phase relationship, coincident with the emergence of 100-ky cycles, indicates that there was a switch in the response time of the NH ice sheets related to a change in their size. Before the MPT, ice sheets were smaller and responded linearly to summer insolation forcing; however, after 700 ka, with the emergence of the 100-ky climate regime, ice sheets grew larger, which delayed their response time to changes in high-latitude temperatures.

Fig. 2

Phase and coherency relation between δ18Ob and Mg/Ca-BWT at the eccentricity (100-ky) and obliquity (41-ky) bands. Intervals that are coherent at 80% and 95% confidence levels are shown with gray bars and black bars, respectively. Before coherency and phase analysis, all records were interpolated to even intervals of 3-ky resolution. Phases were computed with the use of the ARAND program iterative spectra feature, with a 300-ky window and 250-ky lags.

The BWT record at site 607 reflects a combination of changes in global temperature and local changes due to water mass variability (13, 14, 19). The δ13Cb record (Fig. 1D) shows that for the past 3.2 My, site 607 was predominantly bathed during interglacials by relatively warm, salty, and carbonate [CO3]–saturated NADW, mainly reflecting northern Atlantic sea surface conditions. During glacial periods, the site was bathed, to a large extent, by the relatively cold, fresher, and less saturated Antarctic Bottom Water (AABW), which is linked to surface conditions in the Southern Ocean. The relative proportions of NADW (T = 2.6°C) and AABW (T = 0°C), estimated from δ13Cb (9), vary on both long and orbital time scales. From 3.2 to 2.7 Ma, NADW dominated site 607, reflecting warmer NH polar surface temperature, probably due to strong meridional overturning circulation (20). The decrease in BWT across the LPT occurred in both interglacial and glacial temperatures and thus is attributed mostly to global cooling related to the initiation of NH glaciations. After 2.7 Ma, NADW production decreased but was not completely replaced by AABW, and thus the increase in BWT ∆(G-I) primarily reflects global variability related to the waxing and waning of NH ice sheets. Our record, showing conspicuous cooling during the LPT, contrasts with the ostracode-based reconstruction of (11), which shows no discernible cooling trend across the LPT, and with the results of (7) who, based on an alkenone-derived record of sea surface temperature (SST) from the eastern equatorial Pacific, suggested that climate cooled monotonically from the Pliocene to the late Pleistocene. Here we suggest that our record more accurately reflects a global temperature trend, whereas tropical SST records are also imprinted by regional effects.

From 1150 to 850 ka, the variability in BWT ∆(G-I) reflects high-latitude surface water cooling with a growing influence of water mass variability. After accounting for HL-LGM change in [CO3], we estimate a ∆(G-I) BWT change of ~3.1°C, which is entirely consistent with the ∆δ18Ob change of about 1.5‰ in Chain 82-24-23PC, assuming HL-LGM ∆δω =0.8‰ in the North Atlantic (12). Support for our interpretation of the Mg/Ca record in terms of temperature rather than changes in saturation comes from several lines of evidence. Whereas Mg/Ca shows strong similarity with the δ18Ob record, it differs significantly in both the general trend and timing of transitions from changes in the deep Atlantic [CO3] saturation as reflected in the δ13Cb and CaCO3 preservation records (12). On an orbital scale, BWT leads changes in δ13Cb (table S2), suggesting that circulation-driven changes in carbonate saturation are not the dominant control of Mg/Ca. We estimate that changes in ocean circulation contributed <0.5°C to the total glacial cooling of 1.3°C, suggesting an average decrease of BWT of about 0.9°C during both glacial and interglacial maxima due to high-latitude cooling. Indeed, the trend, ∆(G-I), and frequency changes are entirely consistent with SST records from site 607 (Fig. 1E), Ocean Drilling Program (ODP) site 982 (21), and tropical upwelling regions, suggesting that we are capturing a global climate shift (13, 22).

We use the paleotemperature equation [T = 16.9 – 4.0 (δ18Ob – δω)] to calculate the δ18O composition of seawater (23). We assume that δω variability mainly reflects variations in global ice volume, where a 10-m change in sea level equates to a 0.1‰ change in δω (24), which is on the upper side of, though consistent within, uncertainties with estimates of changes in δω (25), and yields the best fit with sea level reconstructions over the past 450 ky (Fig. 3) (21). Given an error of ±1.1°C in BWT estimation and an uncertainty of ~0.2‰ in δ18Ob (12), the uncertainty of sea level reconstruction is ±32 m for ∆(G-I) amplitude and ±21 m for the mean trend (12). The δω record shows a gradual increase in continental ice from the mid-Pliocene to the late Pleistocene, primarily due to an increase in ice volume during glacial intervals. Between ~2.7 and 1.2 Ma, average ∆(G-I) sea level fluctuations are about 60 to 80 m, increasing to 120 ± 32 m in the late Pleistocene. It is noteworthy that interglacial ice extent inferred from the δω record has been relatively constant since 3.2 Ma, in contrast with the benthic δ18O record without temperature correction.

Fig. 3

(A) Sea level record from 0 to 450 ka derived from Mg/Ca-BWT estimates and δ18Ob from Chain 82-24-23PC and DSDP site 607, compared with other published sea level records (17, 3234). (B) δω record from 0 to 3.2 Ma (‰, using standard mean ocean water). The solid block line in (A) and (B) represents a three-point smoothed curve of the δω record (12).

Our sea level reconstruction is generally consistent with previous studies suggesting 40 to 70 m of peak-to-peak ∆(G-I) in the late Pliocene and the increase in glacial ice toward the late Pleistocene (11, 12, 26). However, our new record provides critical details necessary for assessing the evolution of NH ice sheets in relation to changes in the climate background state. Our record shows that the LPT cooling is associated with an average sea level drop of ~22 ± 21 m due to 44- and 15-m drops during glacial and interglacial intervals, respectively (Table 1). Within errors, these estimates are consistent with previous estimates of 30 to 35 m of sea level fall (11, 26, 27, ). Across the MPT, glacial sea level drops by ~20 ± 21 m, whereas interglacial sea level stays relatively stable, resulting in an average increase in ice volume of about 8 ± 21 m, associated with a global cooling of ~0.9°C. The glacial estimate is consistent with a reconstruction of low sea level stands, based on depositional sequences from shallow marine sediments, which found a 20 to 30 m decrease from 1000 to 900 ka (28). An additional and comparable drop in glacial sea level occurs in the late Pleistocene, after the transition to 100-ky periodicity. However, our records show that this last interval is also characterized by more extreme deglaciations manifested by higher (15 ± 32 m) interglacial sea level stands. Apparently, after the MPT, less NH ice persisted in interglacial periods despite the cooler climate.

The divergence in sea level and similarity in deep ocean temperature history across the LPT and MPT suggest that each transition may be responding to a different set of forcings. The global cooling of the deep ocean and concomitant increase in ice volume across the LPT may reflect a high-latitude response to a secular decrease in atmospheric Pco2 (5). Our 2°C cooling across the LPT is consistent with modeling studies suggesting that the inception of NHG may be linked to a substantial drop in air temperature during the late Pliocene (29, 30). Deep ocean cooling associated with the MPT has been highlighted as another threshold response to radiative forcing driven by a decrease in atmospheric Pco2, which resulted in a larger areal extent of NH ice sheets. This would have led to larger fluctuations of planetary albedo, resulting in enhanced BWT ∆(G-I) and sea level amplitudes due to colder and “icier” glacials and warmer, less icy interglacials. Alternatively, the regolith hypothesis maintains that changes in basal rock conditions enabled the buildup of taller (rather than expanded) ice sheets, independent of G-I changes in albedo (10). Our record shows that deep ocean cooling from 1150 to 825 ka precedes the major expansion of ice sheets and the frequency shift from 41-ky to 100-ky glaciations at 700 ka, thereby suggesting that although cooling might have played an important role, it was not sufficient to explain the MPT. Two lines of evidence support this hypothesis. First, the decrease in G-I BWT–to–ice volume ratio during the MPT suggests an increase in the sensitivity of NH ice sheets to changes in global temperature, which cannot be directly attributed to increased albedo. Second, although the glacial increase in ice volume suggested by the δω record is consistent with the cooling of glacial intervals, the interglacial decrease of ice volume is inconsistent with the cooling of interglacial intervals. Both observations point to a fundamental change in ice-sheet dynamics rather than just a threshold response to global cooling. This corollary is consistent with the fact that the frequency change occurs after the major cooling interval.

Our data support the hypothesis that the MPT represents a fundamental change in ice-sheet dynamics that is consistent with the growth of thicker, more unstable ice sheets that fully deglaciate during interglacial periods (10, 30). The increased abundance of IRD after the MPT is consistent with an increase in ice-sheet thickness (31) and with geological evidence suggesting that ice sheets grew in height rather than areal extent (10). We conclude that the cooling associated with the MPT might have preconditioned the NH to allow the growth of taller glacial ice sheets but cannot account for the shift in frequency or sea level history.

Supporting Online Material

Materials and Methods

Figs. S1 to S5

Tables S1 and S2


  • Present address: Research School of Earth Sciences, Australian National University, Canberra, ACT, 0200, Australia.

References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. We thank J. Wright for assistance with isotope measurements and suggestions, M. Raymo for numerous discussions, and W. Zhang and K. Lawrence for assistance with time series analysis. Two anonymous reviewers provided insightful suggestions that substantially improved the manuscript. We acknowledge the ODP and Woods Hole Oceanographic Institution Seafloor Samples Laboratory for supplying sediment samples. This work was supported by a USSSP Schlanger ODP Fellowship to S.S. and NSF award OCE 02-20922 to Y.R.
View Abstract

Navigate This Article