Research Article

Tropical Ocean Temperatures Over the Past 3.5 Million Years

See allHide authors and affiliations

Science  18 Jun 2010:
Vol. 328, Issue 5985, pp. 1530-1534
DOI: 10.1126/science.1185435

Birth of the Cool

Over the past 4 million years or so, tropical sea surface temperatures have experienced a cooling trend (see the Perspective by Philander). Herbert et al. (p. 1530) analyzed sea surface temperature records of the past 3.5 million years from low-latitude sites spanning the world's major ocean basins in order to determine the timing and magnitude of the cooling that has accompanied the intensification of Northern Hemisphere ice ages since the Pliocene. Martínez-Garcia et al. (p. 1550) found that the enigmatic eastern equatorial Pacific cold tongue, a feature one might not expect to find in such a warm region receiving so much sunlight, first appeared between 1.8 and 1.2 million years ago. Its appearance was probably in response to a general shrinking of the tropical warm water pool caused by general climate cooling driven by changes in Earth's orbit.


Determining the timing and amplitude of tropical sea surface temperature (SST) change is an important part of solving the puzzle of the Plio-Pleistocene ice ages. Alkenone-based tropical SST records from the major ocean basins show coherent glacial-interglacial temperature changes of 1° to 3°C that align with (but slightly lead) global changes in ice volume and deep ocean temperature over the past 3.5 million years. Tropical temperatures became tightly coupled with benthic δ18O and orbital forcing after 2.7 million years. We interpret the similarity of tropical SST changes, in dynamically dissimilar regions, to reflect “top-down” forcing through the atmosphere. The inception of a strong carbon dioxide–greenhouse gas feedback and amplification of orbital forcing at ~2.7 million years ago connected the fate of Northern Hemisphere ice sheets with global ocean temperatures since that time.

Understanding the links between tropical sea surface temperatures (SSTs) and changes in high-latitude climate across ice age cycles presents a challenge to paleoclimatologists. On the one hand, the tropical oceans should be shielded from processes that produce large temperature sensitivity in the high latitudes, including ice-albedo feedback on land, sea-ice feedbacks, and steering of wind fields by ice sheets. Modeling studies suggest that the direct effects of large continental ice sheets should extend equatorward on a length scale of ~2000 km from the (primarily Northern Hemisphere) edge of polar ice sheets and sea ice (1, 2)—a substantial radius, but not nearly enough to extend climate change throughout the tropics. However, the high latitudes may possess other potent influences on tropical ocean temperatures. High-latitude oceanographic processes determine the properties of subsurface waters that help to determine vertical density stability, mixing, and upwelling in the tropical oceans (35). Most ocean carbon cycle models also give the high-latitude oceans the dominant role in modulating the carbon dioxide (CO2) greenhouse effect that helps determine global surface temperatures (68).

Changes in tropical ocean SST could in turn lead to large positive feedbacks that enhance the global expression of glacial-interglacial climate cycles (9, 10). The tropical oceans provide the major storage of heat for the global climate system and are also the source of most of the atmospheric water vapor, the single most important greenhouse gas. Evaporation of water from the surface ocean depends directly on SST (11). Through their dominant role in the global hydrological cycle, tropical oceans can also exert effects on global cloud cover, on terrestrial vegetation and soil moisture, and on the release of the greenhouse gases methane (CH4) and nitrous oxide (N2O) from land (12), both of which depend in part on tropical temperatures and hydrology.

The magnitude of tropical ocean temperature change over the course of the Ice Ages has been debated since the inception of modern quantitative paleoclimatology. The first global reconstructions of ocean SST at the heart of the last ice age, the seminal Climate: Long-range Investigation, Mapping, and Prediction (CLIMAP) study based on marine microfossils, suggested that glacial ocean cooling was restricted to mid and high latitudes (some tropical oceanic areas in fact yielded estimates of warmer-than-modern SST) (13). Newer geochemically based SST proxies, as well as revisions of microfossil-based SST calibrations, now give a consensus that the CLIMAP result—although qualitatively correct in producing a picture of polar amplification of glacial temperature changes—underestimated tropical cooling during the last ice age (1417). It is now clear that substantial glacial-interglacial sensitivity exists in tropical SST over the course of at least the past ~1 million years (My) in every region examined to date (1620).

Over the combined ~800-thousand-year (ky) length of the Vostok and European Project for Ice Coring in Antarctica (EPICA) ice core records (12, 2124), tropical SST varies coherently with measured glacial-interglacial variations in the greenhouse gases CO2, CH4, and N2O. Indeed, it has been argued that time series of tropical ocean temperatures may provide one of the best available templates for reconstructing the effect of past greenhouse gas levels on global climate (14). This study presents a composite record of tropical SST variations from the mid-Pliocene [3.5 million years ago (Ma)] to the present, a span of time that allows us to assess tropical high-latitude climate linkages over the course of the onset and intensification of cyclic Northern Hemisphere ice ages at 2.7 Ma (25). We used the alkenone unsaturation index, recorded at ~3-ky resolution, at each of four tropical ocean sediment sites so as to provide a self-consistent picture of SST changes on both broad and orbital scales. Tropical SST records are synchronized via benthic foraminiferal δ18O that was measured in the same sediments at the study sites and aligned to the global LR04 (26) stack. The benthic oxygen isotope record, which records a combination of deep-ocean temperature changes and variations in continental ice volume over time, also allows us to compare the timing and amplitude of tropical SST changes with climate evolution in the high latitudes.

We argue that the tropical SST data suggest that a coherent, substantial CO2-glaciation feedback began at about 2.7 Ma, synchronizing to first order the tropical SST variations at the orbital time scale in late Pliocene and all of Pleistocene time. Before 2.7 Ma, the combination of glacial and CO2 feedbacks may have been much weaker, suppressing the similarity of tropical SST between the different ocean basins and lessening their sensitivity to high-latitude processes.

Evolution of tropical SST over 3.5 million years. We chose to examine four tropical sites located in dynamically distinct regions of the three ocean basins [Fig. 1 and supporting online material (SOM) text S.1]. We produced alkenone SST records at each site over intervals of Plio-Pleistocene time in which supporting stratigraphic data indicate continuous sedimentation. The alkenone paleotemperature proxy relies on the unsaturation index of biomarkers produced by restricted species of coccolithophorid algae, organisms that must live in the photic zone. Modern ecological studies and data from recent sediments support the proposition that alkenone temperature estimates closely approximate mean annual temperature in the surface ocean mixed layer (27). By combining alkenone records from widely dispersed tropical locations, we hope to minimize biases from local SST responses and seasonal biases in alkenone production (SOM text S.1). The alkenone proxy reaches saturation at about 28°C, which prevents us from producing records from the “warm pool” locations of the tropical Indo-Pacific and Atlantic basins.

Fig. 1

Location of long sediment cores analyzed for this study in relation to mean annual SST (SOM text). SST estimates from site 806 were made with the Mg/Ca method (19); all others were made by use of alkenone paleothermometry.

Our tropical records show very strong similarity to each other and to the pattern of global ice volume and deep-sea temperatures over the course of the Plio-Pleistocene glaciations (Figs. 2 and 3). Within the pattern of rhythmic SST changes, unusually intense coolings punctuate the record at ~2.5, 2.1, and 1.7 Ma at all the sites. The similarity of tropical SST extends beyond the pattern of individual ice age cycles; our data reveal coherent long-wavelength (300 to 500 ky in duration) trends in SST that may be related to orbital forcing (Figs. 2 and 4). The absolute values of SST variance, centered on a 200-ky moving average, appear to converge strongly in the late Pleistocene (past 1 My), which is indicative of an increasingly uniform ice-age tropical SST response as we approach the present (Fig. 3). More surprisingly, changes in SST variance appear timed to the 404-ky component of orbital eccentricity; variance in SST is highest during periods of lower-than-average eccentricity. The existence of eccentricity-related (~100- and 400-ky spectral components) in the δ18O record has long been controversial (2831), but this expression in tropical SST variance seems hard to dismiss. Low eccentricity apparently favors prolonged excursions to low temperatures during ice ages, even during the lengthy interval from ~2.7 to 1 Ma when ice age cycles follow the 41 ky obliquity pacing.

Fig. 2

Estimates of SST at the four sites on the basis of alkenone paleotemperature determinations. The gap in the ODP site 662 record comes from an interval disturbed by slumping that was not sampled. Data are archived at

Fig. 3

(A) Variance in δ18O, computed in 200-ky moving windows. (B) 404-ky eccentricity cycle (inverted). (C) Variance in SST, computed in 200-ky moving windows, at the four sites with alkenone paleotemperature estimates.

Fig. 4

Comparison of the tropical stack (four alkenone records plus Mg/Ca record from ODP site 806 of Medina et al. (19) to the global benthic δ18O stack of Lisiecki and Raymo (26).

The similarity of our tropical SST records led us to construct an averaged representation (“stack”) in order to compare the aggregate tropical temperature changes with the global δ18O curve (Fig. 4 and SOM text S.5). We have included a shorter (1.4 My) Mg/Ca–based record from the Pacific warm pool in an effort to better represent the evolution of the warmest regions of the tropical ocean over time. The stack enhances the signal-to-noise ratio of the proxy measurements, emphasizes the common components of tropical temperature trends, and averages the phasing of SST changes in disparate parts of the tropical ocean. It reveals the strong similarity in timing and amplitude of tropical temperature changes in relation to ice age climates, although the SST data contain more long-wavelength (~400 ky) variability and also more intense coolings at ~1.66 and 2.07 Ma (marine oxygen isotope stages 58 and 76, respectively) than would be inferred from the benthic δ18O record of glaciation. The stack estimates only one aspect of the tropical response to glacial cycles over time; other important variables such as the tropical hydrological cycle remain to be reconstructed on Plio-Pleistocene time scales.

The tropical SST records resemble the evolution of high-latitude climate in important ways. On the long time scale, the rising variance, with an exponential time scale (averaged over the 3 longest SST records) of 0.58 My recalls the pace of the 0.68 My time constant of the exponentially increasing amplitude in δ18O toward the present (32) (Fig. 3). Shifts in spectral content, most notably the mid-Pleistocene transition from 41- to 100-ky cycles, occur in both tropical SST and benthic δ18O in the interval 1.0 to 0.8 Ma (Figs. 2 and 4, SOM text S.2, and figs. S2 and S3). In detail, the phases of tropical SST with respect to global ice volume do differ between sites by a few thousand years, implicating the operation of regional-scale processes in addition to the common component. In all records, however, phase analysis indicates that tropical SST slightly leads (by about 2 to 5 ky, depending on site and time) glacial cycles as recorded by δ18O. These phase relations indicate that one cannot regard tropical SST as being forced by continental glaciation itself. Because the benthic δ18O signal is a composite of the relatively slow process of high-latitude ice sheet growth and decay and the potentially much more rapid process of deep ocean cooling and warming on glacial-interglacial cycles (3335), the net phase of the δ18O signal most likely is weighted toward the slow ice sheet component. Our data therefore leave open the possibility that tropical temperatures changed essentially synchronously with those of the deep sea (35) over much of the past 2.7 My.

A milestone in climate evolution occurred at ~2.7 Ma, when linkages both between the tropical sites and from the tropics to the high latitudes (represented by the δ18O curve) increased. The increase in 41-ky SST variance at 2.7 Ma coincides with the appearance of ice-rafted debris in the open North Atlantic and Pacific oceans at the same time, which is widely interpreted to reflect the appearance of large Northern Hemisphere continental ice sheets that would characterize glacial climates from that time onward (25, 36). This obliquity-related component of tropical SST variability not only increases in amplitude beginning at about 2.7 Ma (fig. S3), but becomes highly coherent between the tropical sites at this time (Fig. 5). Other tropical proxies such as the 41-ky pacing of dust delivery to the Arabian Sea (37) and an increase in C4 vegetation in northeast Africa (38) also show the incursion of high-latitude signals into the tropics at about the same time. For the next ~2 My, ice age cycles and tropical SST varied coherently at the 41-ky time scale, following the pacing of the earth’s orbital obliquity.

Fig. 5

Coherence between tropical sites at the 41-ky obliquity frequency, across the time span of intensification of Northern Hemisphere glaciation. The coherence at the 41-ky band rises sharply around 2.7 Ma and remains high from that time to the present. Coherence was determined by means of cross-spectral analysis using a 500-ky window with one-half lags.

Global Connections. The similarity of SST responses in dynamically dissimilar regions of the tropical ocean indicates a common forcing of glacial-scale temperature changes. However, the observed SST changes are opposite in sign to the annual tropical radiation forcing at the obliquity periodicity (18) and lack the 23- and 19-ky spectral components that should dominate seasonal insolation forcing of the study regions (39). The observed patterns therefore cannot be considered as simple adjustments of tropical SST to solar insolation forcing. Alternative mechanisms for changing SST within tropical climatology (such as changes in the strength of the Easterlies, changes in equatorial thermocline tilt, or changes in the intensity of monsoonal winds) would not produce synchronous responses in the tropical oceans or would produce very different amounts of warming and cooling in different tropical regions. Certain mechanisms, such as a trade wind–induced change in the tilt of the equatorial Pacific thermocline, would produce antiphased SST changes between the East and West that have not been observed [fig. S4 of (19) compares the similarity of the alkenone record of SST in the eastern Pacific Ocean Drilling Program (ODP) site 846 to the Mg/Ca record from ODP site 806 in the western Warm Pool].

It seems more than coincidence that both the amplitude of tropical SST cycles, and the coherence of the 41-ky component in tropical SST to orbital pacing rose at the same time that the Northern Hemisphere high latitudes began to sustain major glaciations. Because it is unlikely that Northern Hemisphere ice sheets directly drove the SST responses, a parsimonious interpretation of the climatic changes over the past 3.5 My would instead invoke a common cause for both the intensity of glacial cycles and their expression in the tropics. The most likely candidate lies in high-latitude oceanographic processes tied to the global carbon cycle, which would provide a common driver for global temperature changes and for glaciation. No clear consensus exists yet to explain the late Pleistocene CO2 and greenhouse gas cycles captured in polar ice cores, but most plausible models concentrate on mechanisms in the high-latitude oceans (such as changes in stratification, wind mixing, or biological productivity) that could sequester CO2 in the deep sea during ice ages and then release it in interglacial periods (6, 8, 40). However, a tropical SST change in parallel with CO2 forcing would provide a powerful positive feedback to glacial-interglacial climate change by enhancing the initial CO2 greenhouse forcing with additional water vapor effects (11, 41). Tropical temperature feedbacks could easily spill over into forcings on methane and nitrous oxide emissions from tropical wetlands as well, as suggested by the record of these gases in polar ice cores (12, 23).

We therefore used the rhythm and amplitude of the tropical SST stack to infer the following sequence of greenhouse gas–climate relations over the past 3.5 My. Before the onset of cyclic Northern Hemisphere glaciation, CO2 and tropical temperatures varied dominantly at the ~404-ky cycle, with only modest amplitude. At about 2.7 Ma, poorly understood climatic processes began to couple high-latitude Northern Hemisphere glaciation to deep-ocean CO2 sequestration at the obliquity pacing, which led to the step-function increase in coherency at the 41-ky obliquity component between the tropical SSTs and also between the tropical records and deep ocean δ18O (Fig. 5 and fig. S5). Since that time, every glacial cycle was sustained to a large degree by a reduction in CO2 and other greenhouse gases, as evidenced by the consistent pattern of tropical cooling and warming exhibited in close temporal association with increases and decreases in the δ18O record. The amplitude and pacing of the greenhouse gas cycles has changed sympathetically with high-latitude climate as well: We would infer that the late Pliocene–early Pleistocene declines in CO2 were much less severe than those of the late Pleistocene, as evidenced by the smaller drops in tropical SST. The transition from the 41-ky regime to the late Pleistocene 100-ky regime—although it led to deeper glacial extremes—hardly seems like a revolution. Tropical SSTs were strongly coupled to high-latitude change before this transition, and a pervasive ~400-ky (possibly eccentricity-related) modulation of SST variance preceded the spectral shift by at least 1.5 My. The increasingly similar pattern of SST variance (Fig. 3) reflects the imposition of an increasingly powerful greenhouse gas component of tropical SST forcing that overrode potentially divergent regional dynamical responses of tropical SST to the ice ages. The pattern we sketch seems consistent with the less continuous but more direct evidence on the evolution of CO2 over the past 2 My (42) on the basis of the boron isotope proxy in marine plankton shells. As reconstructed in that study, CO2 excursions also occurred with the ice age cycles of both the “41 ky” and “100 ky” worlds, although their amplitude during the 41 ky world of the early Pleistocene was approximately one half (30 to 40 parts per million) of typical late Pleistocene 100-ky ice age cycles.

If greenhouse gas feedbacks largely coordinated the intensity of the glacial effect on global climates, then how can one explain the evidence to this point that global CO2 levels have declined by only a very small amount from the mid-Pliocene to the present (4346)? The answer seems to us that our view of climate evolution over this time has been confounded by two types of trends in the data. The first is the tendency for glacial cycles after 2.7 Ma to become more intense toward the present, which is evident in both the benthic δ18O record and in our tropical SST data. Many studies have implicitly assumed that this dynamic represents a second process, a progressive drift toward cooler and more glaciated conditions. Linear fits to long climatic time series appear to support such a drift (Table 1), but they are quite misleading. If we look simply at the long-term trend of extremes of the ice ages and pick the maxima and minima of the benthic δ18O curve for succeeding glacial and interglacial periods, we see that the behaviors are quite different from each other (Fig. 6A). The depth of glaciations increases almost monotonically toward the present. Interglacials, however, have changed relatively little over the past 3.5 My. After a period of progressive cooling and/or increase in continental ice in the late Pliocene, interglacial climates in fact reached their coldest/most glaciated extent in the early—not late—Pleistocene. This interpretation is consistent with the ice volume reconstruction of (35) that suggests more-intense deglaciations for the late as compared with the early Pleistocene. The story from the tropics is quite similar (Fig. 6, B and C). Interglacial cooling has been modest (Fig. 6C). Interglacial temperature decline at our sites has not been uniform but ranges from about 0.15°C per million years to 0.7°C per million years (Table 1). Only the eastern equatorial Pacific cold tongue (represented by ODP site 846) shows a strong progressive cooling over time (we have not tried to model oscillatory components of the data in Fig. 6). Evidence from the Indo-Pacific warm pool based on a different (Mg/Ca) SST proxy suggests essentially no cooling over the past 3.5 My (19, 47, 48). We therefore estimate the aggregate tropical interglacial cooling to lie in the range of 0.5 to 0.75°C over the past 3.5 My. In contrast, glacial SST cooling has intensified strongly and consistently between the sites (Fig. 6B).

Table 1

Statistical analysis of alkenone SST trends shown in (Fig. 6). T, temperature in degrees Celsius; r, correlation coefficient; t, age in millions of years.

View this table:
Fig. 6

(A) Evolution of benthic δ18O for interglacial and glacial periods over the past 4 My. (B) Evolution of tropical SST for glacial periods over the past 3.5 My, with least squares line fit superimposed. (C) Evolution of tropical SSTs for interglacial periods over the past 3.5 My, with least-squares line fit superimposed.

The story of the evolution of global climate in relation to Northern Hemisphere glaciation turns out to concern mostly the ice ages themselves: It is a story of how and why, beginning at about 2.7 Ma, the climate began to plunge ever more deeply into ice ages over time while recovering to a similar interglacial state in between ice ages. A modest (1 to 2°C) change in deep-ocean temperatures over this interval, and essentially no change in interglacial sea levels (35), together with the small interglacial tropical ocean surface cooling we estimate above, require only subtle reductions in CO2 and greenhouse gases in the interglacial state of climate since 2.7 Ma. This picture is consistent with reconstructed glacial-interglacial contrasts in partial pressure of CO2 (Pco2) (42), which display no systematic reduction in interglacial CO2 levels over the past 2.1 My but rather an intensification of Pco2 drawdown in glacial periods toward the present, as would be inferred from the asymmetrical evolution of our continuous tropical temperature stack (Fig. 3). Provided that consistent, orbital-resolution paleo-CO2 reconstructions can be obtained in the future, our tropical temperature stack can help to constrain the equilibrium sensitivity of tropical climate to CO2 forcing over a range of paleo-CO2 perturbations since the mid-Pliocene (4246).

Supporting Online Material

References and Notes

  1. This research was supported by National Science Foundation (NSF) grants OCE9986760 and OCE-0351599 to T.D.H. and OCE0623487 and OCE0623310 to T.D.H. and K.T.L. and by grants from the Evolving Earth Foundation to K.T.L. This research used samples provided by ODP and the Integrated Ocean Drilling Program (IODP). ODP was sponsored by NSF and participating countries under the management of Joint Oceanographic Institutions. IODP is supported by NSF; Japan’s Ministry of Education, Culture, Sports, Science and Technology; the European Consortium for Ocean Drilling Research; and the People’s Republic of China, Ministry of Science and Technology.
View Abstract

Navigate This Article