Research Article

Super ENSO and Global Climate Oscillations at Millennial Time Scales

See allHide authors and affiliations

Science  12 Jul 2002:
Vol. 297, Issue 5579, pp. 222-226
DOI: 10.1126/science.1071627

This article has a correction. Please see:


The late Pleistocene history of seawater temperature and salinity variability in the western tropical Pacific warm pool is reconstructed from oxygen isotope (δ18O) and magnesium/calcium composition of planktonic foraminifera. Differentiating the calcite δ18O record into components of temperature and local water δ18O reveals a dominant salinity signal that varied in accord with Dansgaard/Oeschger cycles over Greenland. Salinities were higher at times of high-latitude cooling and were lower during interstadials. The pattern and magnitude of the salinity variations imply shifts in the tropical Pacific ocean/atmosphere system analogous to modern El Niño–Southern Oscillation (ENSO). El Niño conditions correlate with stadials at high latitudes, whereas La Niña conditions correlate with interstadials. Millennial-scale shifts in atmospheric convection away from the western tropical Pacific may explain many paleo-observations, including lower atmospheric CO2, N2O, and CH4 during stadials and patterns of extratropical ocean variability that have tropical source functions that are negatively correlated with El Niño.

The discovery that the Earth experiences large, abrupt climate variations that have no clear external (solar) forcing has stimulated the pursuit of highly resolvable climate archives such as ice, marine, and terrestrial deposits that can provide clues about the origin of these rapid climate events. Until recently, too few records existed from the tropics to establish whether millennial climate variability is inherently tied to tropical climate dynamics. This void is being filled with new sediment cores collected by the IMAGES program (1).

Western Tropical Pacific Proxy Records

Here, we present δ18O and Mg/Ca paleothermometry results [see supporting online material (SOM)] from surface-dwelling planktonic foraminifera from a site with a high sediment accumulation rate (MD98-2181, hereafter MD81) located at the eastern edge of the Indonesian archipelago at 6.3°N and 125.83°E (at a 2114-m water depth) (Fig. 1). The data are used to investigate whether sea surface temperatures (SSTs) in the Pacific warm pool varied in association with high-latitude climate during the past 70 thousand years (kyrs). If they did, this would have important implications for understanding millennial climate variability, because the Pacific warm pool is an important center of atmospheric convection. During the northern summer, modern SSTs in the warm pool average 29° to 30°C. In winter, SSTs cool to 27° to 26°C. Precipitation is highest in summer, with roughly 80% of annual rainfall occurring between June and October (2). Rainfall averages between 300 and 400 mm/month during summer and between 50 and 100 mm/month during winter. As a result of this precipitation pattern, sea surface salinities vary seasonally by about 1.5 per mil (‰). The annual precipitation pattern is tied to the migration of the Intertropical Convergence Zone (ITCZ) over the site and the timing of the northern monsoon. During modern El Niños and La Niñas, the western tropical Pacific experiences dramatic differences in precipitation. For example, summer precipitation over northern Indonesia can be diminished by as much as 60% during a major El Niño.

Figure 1

Map showing location of MD98-2181.

ENSO-induced changes in surface temperature and precipitation leave a measurable signal in the trace element and δ18O composition of biogenic carbonate precipitated in surface waters. Shallow water corals are excellent recorders of this geochemical variability and provide unparalleled resolution for reconstructing ENSO (3–5). However, there are no coral records from the Indonesian region that extend throughout the entire last glacial cycle, and most glacial age corals are now well below sea level and beyond the reach of conventional coring, except where they have been uplifted by tectonism (6). Planktonic foraminifera are also excellent archives of surface water hydrography and can be recovered from long sediment cores that extend continuously through the last glacial. Globigerinoides ruberand Globigerinoides sacculifer, two surface-dwelling species of foraminifera that have been used extensively in paleoceanographic reconstructions, occur in marine sediments of the western tropical Pacific. In plankton tow studies (7), G.ruber is abundant in the warm summer surface waters of the western tropical Pacific that can reach 30°C. Conversely, G. sacculifer does not produce shells readily at temperatures higher than about 27°C (8). On the other hand, this species does occur in the sediment records from Indonesia as a distinct but low abundant constituent, implying that it produces shells primarily during cooler winter months when temperatures average 26° to 27°C. Analyses of recent and late Holocene G. ruber andG. sacculifer extracted from site MD81 display δ18O and Mg/Ca paleotemperatures that reflect these seasonal preferences [see SOM (fig. S1)]. Calcite preservation is excellent in these cores. The core sits above the lysocline, as attested to by the occurrence of aragonite pteropods throughout the core. The consistency between the isotopic- and Mg/Ca-based foraminiferal SST reconstructions suggests that each proxy provides a good estimate of seasonal surface water conditions.

The age model for the early Holocene and late glacial intervals of MD81 is based on 10 accelerator mass spectrometry dates [see SOM (fig. S2 and table S1)]. The age model for the glacial portion of this core is based on correlation of the δ18O of G.ruber and the Greenland Ice Sheet Project Two (GISP2) ice core δ18O stratigraphy (9). We use discrete horizons in the δ18O stratigraphy in MD81 to correlate directly with GISP2 (see SOM).

Using a minimum number of tie points and extrapolating a nearly constant rate of sediment accumulation between them produces a δ18OVPDB stratigraphy for MD81 that is remarkably similar in character to the δ18OSMOW stratigraphy in GISP2 (Fig. 2). Within the glacial portion of MD81, each of the major Dansgaard/Oeschger (D/O) events can be identified in the δ18O record (Fig. 2). Both planktonic foraminiferal species record the isotopic variability associated with the D/O cycles, although the δ18O of G.sacculifer is offset from that of G.ruber, and the sample resolution is not yet as high. The close correspondence between Greenland and MD81 δ18O records implies that, on time scales of D/O cycles, the tropics were undergoing changes in both summer and winter conditions, with warmer seasonal temperatures and/or lower local salinities occurring during interstadials.

Figure 2

The δ18O of Greenland Ice (GISP2), as compared to the δ18O of planktonic foraminifera from MD81. Interstadial numbers and the Younger Dryas are indicated at the top.

Reconstructing Tropical SSTs

The Mg/Ca paleotemperatures for G. ruber andG. sacculifer (Fig. 3) display systematic shifts between the Holocene and last glacial that are similar to other tropical Pacific records (10). Summer G. rubertemperatures were between 26° and 27°C during the last glacial maximum (LGM), or approximately 2°C cooler than Holocene temperatures (Fig. 3). Winter G. sacculifer temperatures were approximately 3°C cooler during the glacial; however, the warmestG. sacculifer temperatures occurred in the early Holocene. The magnitude of summer and winter SST variability was quite small throughout the past 70 kyrs, averaging only 1° to 2°C throughout most of the last glacial period.

Figure 3

Mg/Ca paleo-SST reconstructions for summer and winter.

The Local Salinity Record

The small amplitude of temperature variability in the glacial portion of MD81 does not account for the larger changes in planktonic foraminifer δ18O or the strong correspondence between the MD81 and Greenland δ18O records. The planktonic record does contain a component of δ18O variability that reflects changes in ice volume. However, this has a longer time constant than the millennial-length variations, and the magnitude of this effect can be constrained from the δ18O of atmospheric O2 from ice cores (11,12). The close correspondence between MD81 δ18O variations and the GISP2 record must reflect local hydrographic changes that affected local surface water δ18O and salinity.

To estimate how surface-water δ18O and salinity varied during the last glacial, we differenced the Mg/Ca-derived and δ18O-derived SST estimates (holding the δ18O of surface water constant at the modern value of [–0.4‰ Pee Dee belemnite (PDB) standard]. The difference in these two estimates is transformed to an isotopic difference between modern and glacial seawater δ18O (Δδ18O) by dividing the difference between the δ18O-derived and the Mg/Ca-derived paleotemperature (Δ temperature) by the temperature-dependent fractionation of 0.22‰/°C. We refer to this as Δδ18O to reflect the change in surface water isotopic composition relative to the modern value. Where Δδ18O exceeds the ice volume component, we interpret this to reflect higher (higher surface water δ18O) or lower (lower surface water δ18O) surface salinities at this site (Fig. 4).

Figure 4

The Δδ18O of planktonic foraminifera, as compared to the GISP2 δ18O record (D/O events are labeled at the top). The component of Δδ18O due to ice volume differences at MIS III and MIS II is shown. The Δδ18O variations that exceed the ice volume value reflect local surface water δ18O differences relative to today. Increased Δδ18O reflects increased salinity. Summer salinities were as much as ∼2‰ higher during the LGM, whereas winter salinities were similar to modern values. The largest salinity variations in both summer and winter were associated with the D/O cyclicity. Stadials were associated with higher salinities during both summer and winter. During the interstadials, surface salinities were similar to modern or, in the case of winter values, were slightly lower than today.

It is evident from the comparison of Δδ18O and the GISP2 temperature record that much of the Indonesian planktonic foraminiferal calcite δ18O signal was associated with changes in local salinity. The modern δ18O-salinity relationship in the tropical western Pacific is approximately 0.2‰ (δ18O)/‰ salinity (13), although this relationship varies slightly from region to region and may be lower in the western tropical Pacific. After accounting for the ice volume effect and applying the modern δ18O-salinity relationship, the Δδ18O results imply that summer surface salinities during the glacial were as much as 2‰ higher than they are today. A similar estimate of glacial-Holocene salinity contrast has been observed in other warm pool records (14). Thus, higher sea surface salinities appear to have been widespread within the warm pool. Because surface water salinities are strongly correlated with precipitation patterns in the warm pool and because precipitation in northern Indonesia is heavily weighted toward the summer months, we interpret these results to reflect a shift in mean precipitation that affected the local hydrologic balance during summer, producing the higher surface water δ18O. Consequently, the summer/winter salinity contrast was smaller.

The most remarkable feature of the MD81 record is the correlation between Δδ18O and the D/O cycles in Marine Isotope Stage III (MIS III) (Fig. 4). Positive excursions in Δδ18O, reflecting increased tropical salinities, coincided with stadial conditions at high latitudes. The interstadials were associated with lower tropical salinities. The magnitude of Δδ18O variability during the D/O cycles was on the order of 0.5 to 1.0‰, equivalent to 1 to 2‰ salinity (approximately the magnitude of the modern seasonal variability). Comparing the Δδ18O variations in the MD81 to the global ocean Δδ18O record derived from the ice core O2record (12) indicates that the positive excursions in MD81 Δδ18O reflect increased surface salinities relative to modern values. This implies that, in the western tropical Pacific, salinities were, on average, saltier throughout much of the past 70 kyrs than they are today. Only during interstadials did summer salinities increase modestly, but they never returned completely to modern values.

A Pleistocene Pattern of El Niño and La Niña on Millennial Time Scales

The magnitude of SST and salinity variability in the western tropical Pacific implicates a pattern of shifting atmospheric convection away from the western tropical Pacific when the high latitudes were colder. This is analogous to modern ENSO, where the region of strongest vertical convection shifts away from Indonesia toward the central equatorial Pacific during an El Niño. However, today strong El Niños typically last less than a year (15, 16). The MD81 record points to longer term shifts in the mean state of atmospheric convection over the western tropical Pacific that lasted for millennia. Because of the way foraminifer samples average a signal, the millennial changes could also reflect more frequent and perhaps more severe El Niños, whereas the interstadials were associated with less frequent and less severe ENSO events. In this way, the signal left in the sediments is weighted toward the mean or more frequent condition. We apply the term “super-ENSO” to distinguish our inference about the millennial patterns seen in the tropics from our knowledge of modern ENSO, which is explicitly tied to interannual variability.

The correlation between stadial conditions at high latitudes and El Niño in the tropics is surprising, because previous proposals have suggested a possible El Niño–stadial relation (17–19). This stems from observations that El Niño can have a far-reaching effect, including generally warmer conditions in the extratropics (20). We submit that a review of existing data for the last glacial indicates that tropical and extratropical climate/ocean variability is consistent with a dominant El Niño state in the western tropical Pacific during stadials and that the brief interstadials coincided with near-normal or La Niña conditions. Below, we review records that exhibit D/O associations and point out how these records provide a consistent link between El Niño and stadials.

Tropical Records

In the Indian Ocean, upwelling along the Oman margin results from southerly monsoon winds during the summer. Productivity is highest in summer in response to wind-driven upwelling, leading to denitrification in the water column (21). Recently, Altabet et al. (21) showed that wind-driven upwelling and productivity along the Oman margin varied in concert with the D/O cycles. Denitrification was reduced during stadials and increased during interstadial times. A δ18O stalagmite record from Hulu cave in eastern China has also revealed a pattern of reduced summer monsoon moisture at times when Greenland was experiencing stadial conditions, consistent with the Oman margin record of millennial-scale monsoon failure (22). Today, monsoon failure and El Niño are strongly correlated, and we submit that this provides a coherent explanation for the coincident changes in the Indian and Asian summer monsoons during stadial periods.

The Pleistocene record of denitrification in the eastern tropical North Pacific (23–25) is very similar to that of the Indian Ocean (21). Denitrification was reduced during stadial periods, implying a less intense oxygen minimum zone (OMZ). Because the OMZ in this region is maintained by high productivity due to trade wind–induced upwelling, the diminution of OMZ during stadial periods suggests that the trade winds were weaker at those times. Modern El Niños are associated with weakened trade winds and reduced productivity in the eastern tropical Pacific.

The glacial record from Cariaco Basin in the western equatorial Atlantic exhibits variations between laminated and bioturbated sedimentation that correlate with the millennial climate oscillations in Greenland (26). Intervals of bioturbated sedimentation coincided with stadials, and laminated intervals coincided with interstadials. The bioturbated sediments reflect periods of lower productivity and reduced rainfall and runoff (26). Disruptions in convection over the western tropical Pacific during modern ENSO results in reduced rainfall and runoff in Cariaco as the seasonal migration of the ITCZ is interrupted. The brief return of laminated sediments during interstadials reflects periods of normal or higher rainfall, runoff, and higher productivity, a pattern associated with the normal or La Nina phase of ENSO.

In the tropical Pacific, Lea et al. (10) used Mg/Ca and δ18O to reconstruct changes in surface water δ18O during the last 130,000 years. Although the record they studied from Ontong Java Plateau is of low resolution, they documented a smaller glacial-Holocene Δδ18O (0.7‰) than we observe in Indonesia and smaller than they observed in their eastern tropical Pacific site. They suggested that the smaller change in the western Pacific could be explained by stronger trade winds and enhanced vapor transport to the western Pacific. These results would appear to be in conflict with those presented here. However, the Ontong Java Plateau site lies at the edge of the central equatorial salinity anomaly associated with modern ENSO (4,27). During an El Niño, the locus of atmospheric convection shifts eastward across the Ontong Java Plateau (27). This site may have been within the zone of enhanced precipitation as convection shifted eastward away from Indonesia during the stadials.

Extratropical Records

In the northeastern Pacific, the Santa Barbara Basin contains a record of marine variability that matches the Greenland temperature record (28). In this basin, periods of laminated sedimentation correlate with interstadials, whereas bioturbated intervals correlate with stadials. We again call upon a tropical link to explain this sedimentation pattern. Today, low sea level pressure and strong atmospheric convection in the western tropical Pacific are accompanied by higher sea level pressure in the northeastern Pacific (29). During an El Niño, reduced convection over the western tropical Pacific leads to a weakened NE Pacific high. As a result, northerly wind stress along the California margin is reduced, leading to diminished productivity, less carbon oxidation, and higher dissolved oxygen in the marginal basins. Shifts such as this have been implicated in recent reversals in the laminated sedimentation patterns of the Santa Barbara and Santa Monica Basins (30). An El Niño–stadial relation provides a logical explanation for reduced productivity and carbon oxidation that resulted in higher oxygen levels within the Santa Barbara Basin during the cold periods.

Atmospheric CO2 and N2O

The most compelling argument for an El Niño–stadial link may be the atmospheric CO2, CH4, and N2O records from ice cores (31–33). The atmospheric concentration of each of these gas species increased during interstadials, although the variations were small. The primary source of preindustrial atmospheric N2O, CH4, and CO2 is from the tropics. Upwelling in the tropical Pacific is the largest atmospheric source of CO2 and a major source of N2O (32). Tropical soils constitute the other primary source of N2O, and these are major sources of CH4. El Niño events disrupt the flux of CO2 and N2O to the atmosphere as upwelling in the eastern equatorial Pacific is reduced or shut down. Similarly, shifts in precipitation from land to ocean, as occurs during El Niño, can have a profound effect on tropical soils, as witnessed by the severe droughts over Indonesia and failure of the monsoon. We submit that the El Niño–stadial association provides a plausible explanation for the lower atmospheric CO2, CH4, and N2O observed in ice core records.

We conclude that the strongest case can be made for an El Niño–stadial linkage during the last 70 kyrs. At times of cooling at high latitudes, the tropical Pacific was experiencing either less-frequent or less-persistent El Niños. The notion that El Niño would have been the dominant state during a glacial contrasts sharply with previous modeling and low-resolution observational studies that have predicted stronger trade winds and a larger tropical thermocline tilt under glacial conditions (34, 35). The new perspective on millennial climate variability in the tropics presents an important challenge to the climate community to find the physical forcing that causes reduced atmospheric convection in the western tropical Pacific and reduced trade wind strength at times of high-latitude cooling. The answers will likely shed important light on how and why the Earth's climate undergoes abrupt climate changes and the likelihood that an abrupt change will occur in the future.

  • * To whom correspondence should be addressed. E-mail: stott{at}


View Abstract

Navigate This Article