Persistent El Niño–Southern Oscillation variation during the Pliocene Epoch



[1] There is an urgent requirement to understand how large fluctuations in tropical heat distribution associated with the El Niño–Southern Oscillation (ENSO) will respond to anthropogenic emissions of greenhouse gases. Intervals of global warmth in Earth history provide a unique natural laboratory to explore the behavior of ENSO in a warmer world. To investigate interannual climatic variability, specifically ENSO, in the mid-Piacenzian Warm Period (mPWP) (3.26–3.03 Ma), we integrate observations from the stable isotopes of multiple individual planktonic foraminifera from three different species from the eastern equatorial Pacific with ENSO simulations from the Hadley Centre Coupled Model version 3 (HadCM3), a fully coupled ocean-atmosphere climate model. Our proxy data and model outputs show persistent interannual variability during the mPWP caused by a fluctuating thermocline, despite a deeper thermocline and reduced upwelling. We show that the likely cause of the deeper thermocline is due to warmer equatorial undercurrents rather than reduced physical upwelling. We conclude that the mPWP was characterized by ENSO-related variability around a mean state akin to a modern El Niño event. Furthermore, HadCM3 predicts that the warmer Pliocene world is characterized by a more periodic, regular-amplitude ENSO fluctuation, suggestive that the larger and deeper west Pacific warm pool is more easily destabilized eastward. These conclusions are comparable to the observed trend over the last 40 years to more regular and intense ENSO events. Future research must resolve whether global warming alone, or in concert with tectonic factors, was sufficient to alter ENSO variability during warm intervals of the Pliocene.

1. Introduction

[2] As a result of the large heat reservoir of the tropical oceans, changes in ENSO dynamics may amplify global warming [Cane, 1998]. Tracing the history of ENSO variability through the mid-Piacenzian Warm Period (mPWP), a recent interval in Earth history with a warmer than present-day climate analogous to our own anthropogenically changed future, is key to understanding future climate change. The mPWP comprised a continental configuration similar to today's but was on average 3°C warmer [Haywood and Valdes, 2004] despite being at climatic equilibrium [Crowley, 1990] with only moderately higher pCO2, (350–400 ppm [Van Der Burgh et al., 1993; Raymo et al., 1996; Pagani et al., 2010]), suggestive of high Earth System sensitivity to pCO2 [Lunt et al., 2010].

[3] During the mPWP and the Pliocene Epoch, an inferred reduction of both the east-west zonal and mixed layer temperature gradients [Dowsett et al., 1999; Chaisson and Ravelo, 2000; Wara et al., 2005; Dekens et al., 2008], a state akin to a modern strong El Niño event [Molnar and Cane, 2007], has lead to the eastern equatorial Pacific (EEP) being described as having a permanent El Niño–like state [Wara et al., 2005; Ravelo et al., 2006]. Such a description is ambiguous, failing to distinguish between a situation where there is ENSO related interannual variability around a mean El Niño–like state [Haywood et al., 2007; Brierley and Fedorov, 2010] and one where the climatic system is in a continual El Niño state with little or no interannual variation [Philander and Fedorov, 2003; Fedorov et al., 2006]. Current Pliocene studies either lack the resolution to determine a 2–7 year ENSO signal through time-averaged signals from multiple crushed foraminifera per data point or may be biased toward recording only warm periods due to biases in plankton growth [Wunsch, 2009]. Recently, ENSO-like interannual variation has been shown in a high-resolution Poirites coral record in the western equatorial Pacific during the Pliocene [Watanabe et al., 2011].

[4] The Ocean Dynamic Thermostat model [Clement et al., 1996] has been shown to cause an inverse relationship between solar forcing (and therefore global temperatures) and eastern Pacific sea surface temperature (SST) during the Holocene on millennial timescales [Marchitto et al., 2010]. The potential for this mechanism to interact with longer timescale Pliocene dynamics has been hypothesized [Rickaby and Halloran, 2005] but the degree to which this occurs is, as of yet, unclear due to the interplay of both tectonic and climatic forcings on the EEP warm pool during the mPWP and any differential responses between solar and greenhouse forced warming.

[5] Here we combine the Hadley Centre Coupled Model, version 3 (HadCM3), output with the stable isotopic composition of multiple individual forams, an approach pioneered by Koutavas et al. [2006] and Leduc et al. [2009], to provide the first direct test of interannual variability in the EEP during the mPWP and investigate mechanisms responsible for change in ENSO behavior during global warming.

2. Methods

[6] To investigate interannual variability in the Eastern equatorial Pacific during the Pliocene we analyzed individual foraminifera from 8 targeted time slices from ODP site 846 (3.10°S, 90.82°W, 3296 m) over the past 4 Myr (Table 1). We used the history of sea surface temperature (SST) [Lawrence et al., 2006] to select snapshots throughout the Plio-Pleistocene to compare various conditions in the upwelling system. This site is situated directly in the East Pacific Cold Tongue (EPCT) in the region of maximum sea surface variation in an El Niño year and sufficiently far south to avoid a bias toward small fluctuations in the Intertropical Convergence Zone (ITCZ) and equatorial countercurrent. Calculated backtrack paths show only east-west translation during the study period and consequently broadly consistent hydrological conditions [Pisias et al., 1995].

Table 1. Name, Age and Temperature of Selected Time Slices
NameAgea (ka)SSTa (°C)

[7] We analyzed the isotopic composition of multiple individual planktonic foraminifera, with a life span of 2–4 weeks [Spero, 1998], within a single time slice to capture the full variability of surface water column conditions, that is, that due to seasonality and interannual fluctuations. We use the mixed layer dwelling Globigerina ruber (white) (0–20 m), Globorotalia menardii, from the steep part of the thermocline (45–70 m), and Neogloboquadrina dutertrei from the deep thermocline (75–135 m) [Faul et al., 2000]. All three species calcify within a narrow depth range and in equilibrium with δ18Osw, or with a constant offset from equilibrium [Fairbanks et al., 1982]. N. dutertrei has been shown to calcify in the thermocline when this overlaps with the photic zone, and slightly above it when the thermocline is deeper than the photic zone [Steph et al., 2009]. In order to demonstrate whether ENSO variation persists, we must investigate whether any individual foraminiferal δ18O values lie outside the expected seasonal variance for different time slices through the Pliocene and Pleistocene.

[8] A sediment trap study from the Gulf of California shows how month-averaged interannual foraminiferal δ18O variation closely follows predicted δ18O values for a variety of different species at different depth habitats in the water column in a hydrological regime with both strong seasonality and interannual variation controlled primarily by ENSO [Wejnert et al., 2010]. We believe that the monthly aliquots used in this sediment trap approach, parallels the use of single specimens from past sediment cores in our study, and demonstrates the success of the single-specimen approach to capturing interannual variability.

[9] From ∼24g of sediment core per time slice (<2 cm core length, between ≈388 and 609 years depending on the sedimentation rate [Shackleton and Shipboard Scientific Party, 1992; Shipboard Scientific Party, 1992], individual foraminifera of the target species were picked from the 355–425 μm and 300–355 μm size fractions and ultrasonically cleaned. Up to 40 individual foraminifera, where possible, were selected per species per sample and weighed on a Sartorius microbalance. Due to the operational limits of the mass spectrometer any foraminifera weighing under 10 μg were discarded. The vast majority of foraminifera did not show, through visual inspection, any signs of dissolution. Type specimens, inspected under SEM, confirmed that there were no small-scale dissolution effects or additional calcite. Foraminifera showing signs of dissolution, additional calcite, infilling or being partially incomplete were not weighed or analyzed.

[10] Oxygen and carbon stable isotopes were measured on a Delta V advantage mass spectrometer with a Kiel IV carbonate device preparation using orthophosphoric acid at 75°C. The Oxford in-house speleothem standard, SPOX, was used, with NBS-18 and NBS-19 to calibrate the results to the PeeDee belemnite standard. The reproducibility of standards is better than 0.1‰ and typically 0.06‰ for δ18O and is better, approaches 0.01‰, for δ13C.

2.1. Model Description

[11] The GCM simulations described in this paper are carried out using the UK Met Office coupled ocean-atmosphere GCM HadCM3, version 4.5 [Gordon et al., 2000]. The resolution of the atmospheric and land components is 3.75° in longitude by 2.5° in latitude, with 19 vertical levels in the atmosphere. The resolution of the ocean model is 1.25° by 1.25° with 20 levels in the vertical. Parameterizations include the radiation scheme of Edwards and Slingo [1996], the convection scheme of Gregory et al. [1997], and the MOSES-1 land-surface scheme, whose representation of evaporation includes the dependence of stomatal resistance on temperature, vapor pressure and CO2 concentration [Cox et al., 1999]. The ocean model uses the Gent and McWilliams [1990] mixing scheme. There is no explicit horizontal tracer diffusion in the model. The horizontal resolution allows the use of a smaller coefficient of horizontal momentum viscosity leading to an improved simulation of ocean velocities compared to earlier versions of the model. The sea ice model uses a simple thermodynamic scheme and contains parameterizations of ice concentration [Hibler, 1979] and ice drift and leads [Cattle et al., 1995]. In simulations of the present-day climate, the model has been shown to simulate SST in good agreement with modern observations, without the need for flux corrections [Gregory and Mitchell, 1997].

[12] The ability of the HadCM3 to simulate ENSO variability has obvious implications for this work. It has been established that HadCM3 has a good simulation of present-day ENSO [Collins et al., 2001; AchutaRao and Sperber, 2002]. In a CMIP (Coupled Model Intercomparison Project) ensemble of flux adjusted and nonflux adjusted models, HadCM3 performed consistently well against observational data, capturing the variability in tropical Pacific surface air temperatures (SAT) as well simulating correctly the timing of the annual cycle of Pacific SSTs and the amplitude of recent ENSO events [AchutaRao and Sperber, 2002]. Both the SST mode (characterized by an east-west propagation of temperature anomalies at the surface and low-amplitude events) and the thermocline mode (characterized by a west-east propagation of temperature anomalies in the subsurface and high-amplitude events) are present in HadCM3 similar to observations [Neelin et al., 1998; Fedorov and Philander, 2001; Guilyardi, 2006].

2.2. Pliocene Boundary Conditions

[13] Required boundary conditions were supplied by the U.S. Geological Survey PRISM2 (Pliocene Research Interpretations and Synoptic Mapping project) 2° × 2° digital data set. The particulars of the PRISM2 data set have been well documented in previous papers [Dowsett et al., 1999; Haywood et al., 2000, and references therein]. In brief, the prescribed boundary conditions cover the time slab between 3.29 and 2.97 Ma according to the geomagnetic polarity time scale [Berggren et al., 1995]. Boundary conditions integrated into the model that are specific to the Pliocene include: (1) continental configuration, modified by a 25 m increase in global sea level, (2) modified present-day elevations, (3) reduced ice sheet size and height for Greenland (∼50% reduction) and Antarctica (∼33% reduction), (4) Pliocene vegetation distribution, and (5) Pliocene SSTs and sea ice distributions. The global mean surface temperature in the mPWP simulation is 3°C warmer than the preindustrial simulation, with deep ocean temperatures 0.5–1°C warmer. In addition, the mid-Piacenzian simulation has a mean state warming in the EEP of up to 3°C and a reduced west-east SST gradient at the surface. This is similar to the hypothesized permanent El Niño–like state (El Padre) for the mPWP derived from alkenone and Mg/Ca SST records [Wara et al., 2005; Lawrence et al., 2006; Cannariato and Ravelo, 1997; Steph et al., 2006].

[14] The geographical extent of the Greenland and Antarctic Ice Sheets within the PRISM2 data set was based on global sea level estimates derived for the Pliocene by Dowsett and Cronin [1990]. The PRISM2 reconstruction uses model results from Michael Prentice (personal communication to Harry Dowsett [Dowsett et al., 1999]) to guide the areal and topographic distribution of Antarctic and Greenland ice. A 25 m sea level rise is equivalent to a maximum decrease in global average salinity of approximately 0.25 psu, which is small and therefore was not included in our Pliocene coupled ocean-atmosphere simulations. For a more detailed description of the PRISM2 data set and how it differs from earlier PRISM data sets, see Dowsett et al. [1999].

2.3. Model Predictions

[15] HadCM3 simulations, showing temperature anomaly plots for March, April, and May (MAM), the peak time for absolute SST anomalies in the EEP (Figure 1), also suggest that ENSO persists throughout the mPWP. The notable feature of the model prediction for the mPWP is a more regular and periodic behavior compared to the preindustrial simulation, with an increase in ENSO frequency (Figure 1a). The leading empirical orthogonal function (EOF) for the tropical Pacific SST shows increased ENSO variability in the mPWP simulation (∼20%) with the majority of this variability focused over the equatorial Pacific, although there is a smaller meridional extent compared to the preindustrial simulation (Figure 1c). At site 846, SSTs in both the mPWP and preindustrial simulations appear to show a slight bias toward La Niña, with a larger number of temperature anomalies between −0.5 to −1.0°C. However, above a ±1°C anomaly, there are ∼35% more warm anomalies in the mPWP than cold and over ∼20% more compared to the preindustrial simulation (Figure 1b). The number of extreme events (>1.5°C) remains the same for both simulations.

Figure 1.

(a) ENSO Index plots for March, April, and May mean (MAM) at ODP site 846 from HadCM3 running (left) preindustrial and (right) mid-Piacenzian Warm Period (mPWP) (PRISM interval 3.264–3.025 Ma) simulations at two different depths, 5 m and 200 m. Dotted line represents a 1.5°C temperature anomaly (a strong El Niño event), that is, those likely to be recorded using the individual foraminifera technique. Model anomalies are averaged over 5 months; therefore, a 1.5°C model anomaly approximately equates to a 2.25‰ range in foraminifera isotopes. (b) Frequency of SST anomalies over the 200 year time series at 5 m and 200 m depth (grey, preindustrial; red, mPWP). (c) Leading EOFs for the (left) preindustrial and (right) mPWP simulations. The EOF analysis decomposes the signal using orthogonal basis functions determined from the SST data from each simulation (minimizing residual variance) to find spatial and temporal patterns of variability. At ODP site 846 variances are 18.98% and 22.87%, respectively, with the seasonal cycle removed, indicating an increase in nonseasonal variance from the preindustrial to the mPWP simulations.

3. Determining Interannual Variability

[16] In order to determine the true signal of interannual variability from individual foraminifera, which may be compared to the model simulations, it is necessary to dissect the isotopic signal into its constituent parts. We define the term extraseasonal as referring to an ENSO event that produces conditions that are more extreme than the natural range caused by seasonality and non-ENSO event interannual variability (i.e., with ENSO indices between −0.5 and 0.5). Many ENSO events will not produce extraseasonal conditions (see Figure 2). An extraseasonal foraminifera is therefore one that records these extraseasonal events by producing a δ18O signal outside that produced by the non-ENSO variation described above.

Figure 2.

Water temperature seasonality from TOA buoy (2°S, 95°W) at (a) 0 m and (b) 80 m water depth (Tropical Atmosphere Ocean project, 2010, Dashed and dotted grey lines indicate the most extreme warm and cold temperatures recorded by the buoy since 1992 during a non-ENSO event month (ONI <0.5 and >−0.5). Significant years are labeled. Note that the large relative temperature anomalies may not be considered an abnormal absolute temperature anomaly (e.g., November 1997).

3.1. What Counts as an Extraseasonal Foraminifera?

[17] The climatic record of the isotopic composition of individual foraminifera from a package of sediment contains a mixture of signals from a variety of different sources and decoding the interannual variability from the seasonal record is one of the major challenges of interpreting signals from single specimens. There is an interplay between natural variation in seasonality, both from temperature and salinity; interannual variation such as ENSO; changes in the dynamics of the ITCZ; and how and when these signals are recorded by the foraminifera through the oxygen isotopes of their carbonate shells (δ18OCaCO3).

[18] Previous studies [Koutavas et al., 2006; Leduc et al., 2009] use the World Ocean Atlas to predict the amplitude of an expected seasonal cycle in δ18O carbonate [Antonov et al., 2006; Locarnini et al., 2006], incorporating the effects of temperature and salinity from a typical year. However, typical years do not occur in nature with a small degree of interannual variation occurring from multiple sources at all times. As a result it is a rare event when the Oceanic Niño Index (ONI) is at zero, with some measure of ENSO related variance occurring at all times.

[19] As a consequence we apply a broader seasonal range to include the natural variability of all conditions not defined as an actual ENSO event (Niño Index >0.5 or <−0.5) using data from the nearest TAO array buoy (Tropical Atmosphere Ocean project, 2010,, located both on the site backtrack path and also in the EPCT (at 2°S, 95°W) [Pisias et al., 1995]. We combine this data with O'Neil et al.'s [1969] paleotemperature equation to gain predicted δ18O ranges for the target foraminifera (Figure 2). Recording since 1992, the buoy data encompasses a large range of interannual variability but fails to include a strong La Niña event while it may overrepresent El Niño due to the inclusion of the 2009–2010 event. Ten and fifteen year running means fail to encompass the full natural variability so we include all available data.

[20] We are forced to exclude the effects of salinity on the data due to a paucity of data from the TAO buoy. In the east equatorial Pacific ENSO related salinity variation is typically small with no salinity difference between El Niño months and non-ENSO months (0.63 psu range ≈ 0.3‰ ≈ 1.5°C). A slight freshening (by a further 0.08 psu ≈ 0.38‰ ≈ 1.8°C), albeit only in the top 40 m, during La Niña events will serve only to lighten δ18Osw partially cancelling out the increase in δ18O due to temperature. While seasonal salinity reinforces the temperature effect on δ18O it only has a significant (>0.3‰) change during the boreal spring seasonal freshwater lid at depths less than 20 m.

[21] Not all foraminifera growing during extraseasonal conditions will record an extraseasonal isotopic signature due to the foraminifera's location in both space and time. Due to the uncertainties in the depth habitat of the foraminifera within the species range it is possible that a foraminifera growing in extraseasonal conditions at the base of the species habitat would record an isotopic signature that would be considered normal at the top of the species habitat. Only δ18O signatures outside the full seasonal range for all depths of the species habitat can be considered as indisputably extraseasonal (Figure 3). This false negative scenario reduces the number of extraseasonal foraminifera recorded compared to those that existed.

Figure 3.

Expected depth habitat and δ18OCaCO3 recorded by each foraminifera species during a typical year defined by the World Ocean Atlas [Locarnini et al., 2006; Antonov et al., 2006] (darkest color, dotted line); during non-ENSO event months, that is, the full seasonal range (middle color, dashed line) from TAO buoy data (Tropical Atmosphere Ocean project, 2010,; and encompassing all interannual variation (lightest color, solid line). Species colors as before with red and blue lines on the chart indicating warm and cold extremes. Fill areas show the foraminifera that would be recorded as having the corresponding δ18OCaCO3 rather than the conditions the foraminifera lived under, hence the reduced number of recorded foraminifera with extreme values owing to uncertainty as to where individual foraminifera calcified within the species depth habitat. Absolute δ18OCaCO3 may vary between buoy and core sites through both space and time due to δ18Osw differences. Relative changes are expected to show only minor variations.

[22] Similarly foraminifera growing during peak ENSO events may also not record an isotopic signature outside that seen during the normal seasonal cycle. Due to the temporal discrepancy between peak ENSO conditions (typically boreal winter) and peak seasonal conditions (boreal spring and autumn for hot and cold seasonal extremes, respectively) many large temperature anomalies are not absolute temperature anomalies above the seasonal cycle. At the surface only the strongest ENSO events such as 1997/1998 that last well into the boreal spring produce an absolute temperature anomaly (Figure 2a). This effect is reduced at depth due to the smaller seasonal cycles and at 80 m depth smaller ENSO events such as 2002/2003 and 2009/2010 are also potentially recorded (Figure 2b).

[23] Due to these restrictions, the probabilities of a month occurring with conditions that would be recorded as extraseasonal at each of the foraminifera species depth habitats are 0.04, 0.12, and 0.13 for G. ruber, G. menardii, and N. dutertrei, respectively. Assuming a near constant foraminifera population due to the equatorial setting with minimal seasonality we expect 4%, 12% and 13% of the individual foraminifera for each species as indisputably extraseasonal.

[24] We do not expect any preservational bias from the foraminifera. Given that seasonal foraminifera blooms are controlled by the environmental conditions rather than foraminifera life cycle [Thunell and Reynolds, 1984], in an equatorial environment there is little seasonal blooming of the foraminifera, which would increase both number and preservation potential of a particular month of the year within the sedimentary record. If this does occur it will only act to mask any interannual signal. While El Niño conditions are associated with reduced productivity [Barber and Chavez, 1983] the thicker mixed layer can provide a larger habitat range and therefore increased numbers of mixed layer dwelling foraminifera [Thunell and Sautter, 1992].

3.2. Non-ENSO Variation

[25] The absolute δ18O values will also change through time due to changing sea level and local salinity; indeed the difference between the mean values of PIA2 and MAMC G. ruber can be explained by predicted sea level changes alone. Tectonic changes also play a role; calculated backtracked paths [Pisias et al., 1995] predict little north-south movement, keeping the core site in the EPCT and so similar hydrological conditions. However, east-west changes may also contribute to absolute temperature differences over time and partly account for absolute offsets between the TOA buoy and the core site. As a result we focus primarily on relative differences between foraminifera at different depths rather than absolute changes between time slices. Despite these absolute fluctuations we have several lines of evidence to suspect that δ18O seasonality would be largely unchanged.

[26] There remains the strong possibility that many tropical ocean changes through time could be attributed to ITCZ fluctuations as opposed to changes in ENSO, potentially leading to greater salinity seasonality. In both El Niño events and a warmer Pliocene world, higher east equatorial Pacific SSTs removes the blocking cold tongue and allows a more southerly migration of the ITCZ. If the ITCZ were over the eastern Pacific equator during the Pliocene then we might expect reduced seasonality from more stable temperatures and precipitation patterns; indeed no salinity anomaly is recorded near the core site during a modern El Niño event with a more southerly ITCZ, while underneath the east Pacific ITCZ (WOA05 data from 90.5W, 5.5N) δ18OCaCO3 seasonality is just 0.6‰ at the surface compared to 1.6‰ near the core site. It is likely, however, that without a cold blocking tongue the ITCZ would have a large seasonal latitudinal shift. The salinity difference between current ITCZ and EPCT δ18Osw would increase the seasonality by a further 0.2‰ at the surface. It is important to note however, that at the core site the salinity works against temperature: the warm El Niño events with low δ18Osw have δ18Osw increased by high salinity and the cool La Niña events with high δ18Osw have δ18Osw decreased by the associated low salinity. The influence of fresher conditions at the core site from a more southerly ITCZ is counteracted by warmer sea surface temperatures required for a more southerly ITCZ.

[27] HadCM3 predicts little change in seasonality at the core site, with a slight increase occurring at the surface (0 m) (Figure 4a) and a slight decrease at depth (200 m) (Figure 4b). These changes are small however at less than 1°C additional range.

Figure 4.

Modeled mean annual cycle at ODP site 846 for the mPWP (dashed) and preindustrial (dotted) at (a) the surface and (b) 200 m depth. At the surface the seasonal cycle is larger in the mPWP compared to the preindustrial; however, it is smaller at 200 m.

[28] Furthermore, HadCM3 predicts a small increase in salinity seasonality. A more southerly ITCZ results in greater boreal winter (DJF) precipitation turning from net evaporation to net precipitation. Movement of the DJF ITCZ is minimal, failing to move fully into the southern hemisphere and over the core site. The result of this is a freshening of the entire mixed layer by less than 0.3 practical salinity units (psu). During the boreal summer (JJA) there is no change in precipitation rates or in precipitation-evaporation. The salinity gradient is lost resulting in more saline conditions in the top 5 m (up to +0.5 psu) and fresher conditions below 20 m (−0.3 psu). This homogenization of mixed layer salinity will serve to reduce the range of δ18Osw seen by G. ruber. Without associated precipitation changes this destratification is likely due to reduced seasonality in oceanic mixing and wind stress. The year-round slight freshening would result in no change in salinity seasonality at depth. At the surface Pliocene seasonality is less than 0.6 psu, a δ18Osw change of, at most, just 0.15‰.

4. Results

4.1. Interannual Variation

[29] Assuming modern seasonality ranges for the depth habitats of the Pliocene foraminifera and despite our assumption toward the upper limits of seasonality, interannual variation continues to produce notable extreme warm and cold events (Figures 5a and 6). Both mixed layer dwelling G. ruber and deeper dwelling N. dutertrei record extraseasonal δ18O throughout the Pliocene (Figures 5a and 6) showing four and eleven extraseasonal foraminifera, respectively, accounting for 5% and 17% of the analyzed population, comparable to predicted percentages (Table 2). There is little change in the δ18O range of the population through time: both species show no correlation (Figure 5b, r2 < 0.05) between the standard deviation of sample δ18O and SST, a proxy for EEP climatic state. G. ruber has a Brown-Forsythe Test score of 0.5476, indicating a highly likely (test threshold of 0.05) probability of equality of variance between samples. N. dutertrei has a Brown-Forsythe Test score of 0.0874, likely lowered by the smaller population samples in PIA1 and PIA2 but still remaining above the test threshold.

Figure 5.

(a) Stable isotope analyses of individual foraminifera (G. ruber, blue open squares; G. menardii, green open triangles; N. dutertrei, red open diamonds; along with corresponding means (black Xs) tied with appropriate colored lines). Alkenone-derived sea surface temperatures estimates from ODP core 846 (black line) from Lawrence et al. [2006] with emphasis (yellow arrows) of sample ages. (b) Standard deviation of δ18O of each species at each depth showing trends versus temperature (symbols and colors as before but filled) with black dotted line showing trend of N. dutertrei if sample PIA1 is omitted. (c) Relative abundances of the three target species of foraminifera. For absolute counts, see Table A1.

Figure 6.

Spread of δ18O values for each species centered on the sample mean (colors and symbols as in Figure 5). Horizontal lines represent the extreme seasonal deviation from the sample mean integrated over the entire foraminifera depth habitat (red is warm, and blue is cold) calculated from buoy data from the Tropical Ocean Atmosphere project (Tropical Atmosphere Ocean project, 2010, Far left black symbols indicate the spread of the three target species from Wejnert et al.'s [2010] modern-day study with appropriate warm limit defined. The data set is not long enough for a cold limit to be defined. Black circles show individual G. ruber (1.6 ka) from Koutavas et al. [2006] with two sets of seasonal limits, those defined by Koutavas et al. (inner) and the seasonal limits defined by our study (outer). Black stars indicate the spread of N. dutertrei (1.6 ka) from Leduc et al. [2009] with site-appropriate seasonal limits.

Table 2. Predicted and Observed Percentages of Extraseasonal Foraminifera
SpeciesPredicted Extraseasonal Foraminiferaa (%)Observed Extraseasonal Foraminifera During Warm Pliocene Samplesb (%)
  • a

    From the percentage of months that would produce equilibrium δ18OCaCO3 values outside the seasonal range (including non-ENSO interannual variability).

  • b

    Samples PIA1, PIA2, and ZANC.

  • c

    Percentages of extraseasonal foraminifera using current depth habitat seasonality and likely Pliocene depth habitat seasonality.

G. ruber45
G. menardii121
N. dutertrei13/3c17/6c

[30] These results are comparable to studies containing more recent foraminifera under modern hydrological conditions. Koutavas et al.'s [2006] most recent sample 1.66 ka records 21% extraseasonal G. ruber, which is reduced to 6% under our more stringent criteria, while Leduc et al.'s [2009] 1.61 ka record shows 5% extraseasonal N. dutertrei, albeit at a location with different hydrological conditions.

4.2. Impacts of Changing Seasonality

[31] Typical seasonality experienced at the calcification depth of each of our Pliocene species may have been different during the Pliocene, because the thermocline was deeper in the EEP at this time [Cannariato and Ravelo, 1997]. Three lines of evidence suggest a greater degree of stratification with reduced upwelling of colder nutrient rich subthermocline waters during the Pliocene (Figures 5a and 5c). N. dutertrei, an upwelling preferring species, decreased in relative abundance from >60% in the Pleistocene to <15% in PIA1 and PIA2. The decreased δ18O difference between the shallow and deep foraminifera in the warm Pliocene (PIA1, PIA2, ZANC), and increased δ13C gradient, suggests a reduced vertical temperature gradient by up to ∼6°C (1.33‰) between PIA2 and S5E and a more efficient biological pump drove a nutrient gradient throughout the photic zone in more stratified waters [Lynch-Stieglitz et al., 1995]. Qualitatively, the deeper dwellers show a greater degree of warming, consistent with a deeper thermocline, during the Pliocene, relative to the Pleistocene, up to 1.4‰ for G. menardii and 1.6‰ for N. dutertrei, compared to ∼0.6‰ for G. ruber.

[32] The Pliocene seasonal cycle in suprathermocline waters was likely smaller because the seasonal cycle in Ekman-driven upwelling sourced warm suprathermocline waters, thereby reducing the amplitude of the wind-driven seasonal variation. That the majority of warm Pliocene G. ruber and G. menardii individual foraminifera lie well within ±0.6‰ compared to the seasonal Pliocene limits ±1.1‰, supports a reduced seasonal cycle in suprathermocline water temperature. The deeper Pliocene thermocline also means that the seasonal range of the N. dutertrei habitat was analogous to that of the modern midthermocline G. menardii. Applying this range to the Pliocene N. dutertrei, indicated by the shaded bars in Figure 6, we still find 4 extraseasonal N. dutertrei, 6% of the population, compared to the expected 3%.

[33] There is a hint that the peak heights in the histogram of the Pliocene (Figure 7) are higher which we interpret as supportive of reduced seasonality. Since statistical analysis of the distribution of the data is easily biased due to the relatively small population sizes (n = 40), a single additional extraseasonal foraminifera or lack of can significantly change any histogram descriptive statistics. Quantitatively, the Brown-Forsythe Test and analysis of the standard deviations show that the distribution of the individual foraminifera of G. ruber and N. dutertrei remain reasonably similar throughout the study period. Qualitatively, however, there is a noticeable change in histogram peak height between Pliocene and Pleistocene samples for both species, the Pliocene samples having peak heights above 0.33 and the Pleistocene samples below 0.33 (Figure 7). This could be interpreted as showing reduced interannual variance, but the presence of extraseasonal outliers and the lack of statistically significant changes in standard deviation, range or variance (Brown-Forsythe Test) regarding the whole population distribution leads us to conclude that we are recording a reduced seasonality at both depth habitats, but that the full interannual variation for these two species is not reduced. This result is at odds with the HadCM3 prediction of a slight increase in seasonality at the surface (Figure 4).

Figure 7.

Histograms showing the proportion of individual foraminifera per sample in 0.3‰ δ18O bins, centered on the sample mean, for each species: (a) G. ruber, (b) G. menardii, and (c) N. dutertrei. HOLO is open diamond, S5E is open square, CALA is open triangle, GELA is open circle, PIA1 is closed diamond, PIA2 is closed square, MAMC is closed triangle, and ZANC is closed circle.

[34] A statistically significant trend in population range of δ18O of intermediate dwelling G. menardii exists, which is proportional to SST (r2 value of 0.71; Figure 5b). A Brown-Forsythe Test score of 0.0001 indicates the difference in variance is significant. Notably, there is a larger “Pleistocene-like” G. menardii δ18O standard deviation during the Mammoth Cooling event (Marine Isotope Stage M2, sample MAMC) and the high 86% relative abundance of the upwelling species N. dutertrei, indicate similar rates of upwelling to more recent times and large interannual variability, akin to “Pleistocene-like” conditions. This event is a global perturbation to the warmth of the Pliocene and may be a failed attempt at Northern Hemisphere Glaciation [Haug and Tiedemann, 1998; De Schepper et al., 2009]. There is a hint therefore, that the depth of the thermocline and the superimposed interannual variance appear to vary together with warm and cold cycles in the Pliocene and Pleistocene.

[35] The peak height of the G. ruber MAMC is below 0.33, plotting with the Pleistocene samples, while for N. dutertrei the MAMC sample plots with the warmer Pliocene samples. This indicates that seasonality is dependent on temperature at the surface, potentially related to ITCZ fluctuations, but in the thermocline, seasonality is dependent on larger-scale processes discussed later.

4.3. Mechanisms of Interannual Variability

[36] The sign of the N. dutertrei extraseasonality hints at the mechanism of the interannual variability. Of the N. dutertrei δ18O which exceed 1.0‰ from the sample mean, five of the seven Pliocene extraseasonal foraminifera are exceptionally δ18O positive (cold), while three of the four extraseasonal Pleistocene individuals are exceptionally δ18O negative (warm). This observation is supported by HadCM3: at 200 m depth, there is a clear shift toward stronger cold anomalies (+1°C), with ∼50% more cold events in the mPWP simulation compared to warm and ∼30% in the preindustrial (Figure 1). We propose a fluctuating but deeper thermocline during the Pliocene, which brings cold subthermocline waters to the habitat depth of N. dutertrei, constrained to live close to the photic zone due to its association with photosynthetic symbionts [Hemleben et al., 1988]. Indeed, the closure of the Panama gateway by 4.2 Ma is thought to have caused the local thermocline to shoal to a depth where it could start to interact with the photic zone [Chaisson and Ravelo, 2000; Steph et al., 2006, 2010]. Between 2.0 and 1.5 Ma, the thermocline shoaled sufficiently that divergent trade winds could upwell cooler intermediate waters establishing the east Pacific cold tongue [Ravelo et al., 2004, Fedorov et al., 2006]. Between these times as the thermocline gradually shoals into the photic zone, N. dutertrei inhabits cool photic zone waters when available. N. dutertrei initially inhabits warmer suprathermocline waters in the mean state, and only thermocline shoaling events (La Niña) will be recorded. Later, once the thermocline has shoaled sufficiently, N. dutertrei inhabits the cooler subthermocline waters, with extraseasonality recorded as warmer thermocline deepening events (El Niño). Consequently we can attribute the extraseasonal variability in N. dutertrei to the thermocline, which shoals into the photic zone from the Pliocene to the Pleistocene, but which is persistently fluctuating due to ENSO-like variability.

4.4. Potential Biases

[37] Our result of persistent interannual variability appears to be unaffected by size or weight artifacts due to ontogeny or preservation. Sample mean normalized δ18O versus weight plots show no overall trend with r2 values always <0.14 and typically <0.02. With maximum r2 values of 0.251, and typically <0.1, there is no correlation between the sediment accumulation rate [Shackleton and Shipboard Scientific Party, 1992] and the range of δ18O. Furthermore, mean weights of both warm and cold extraseasonal foraminifera are within one standard deviation of the species mean weight, and with no trend between weight and climatic state, ruling out any influence of changing deposition rate or bioturbation with adjacent glacial intervals. Good preservation of the foraminifera in all samples bar HOLO makes bioturbation an unlikely complicating factor. To avoid running the risk of capturing millennial-scale climatic variations that would obscure higher-frequency lower-amplitude interannual variation such as ENSO while still capturing the largest possible signal a balance is required between as small a time slice as possible and as many foraminifera as possible. We deem forty individuals as sufficient to capture the full range of seasonal and interannual variability. Given the predicted percentages of extraseasonal foraminifera (12 and 13% for G. menardii and N. dutertrei), forty foraminifera is sufficient to confirm a null hypothesis at 99% confidence that the absence of extraseasonal foraminifera means none existed. With lower predicted percentages of extraseasonal foraminifera, G. ruber would require significantly more foraminifera than are present in a narrow time slice to achieve a statistically significant null hypothesis. For two smaller populations, however, there may be little difference between the N. dutertrei results of samples PIA1 and PIA2, the former fortuitously sampling only more extreme foraminifera and the latter more “typical” foraminifera.

[38] False positive results may occur through individual foraminifera calcifying outside our prescribed range. While individual G. ruber and N. dutertrei can be found in plankton tows as deep as 1750 m [Fairbanks et al., 1982] it is unlikely that these individuals calcify at such depths, and that calcification can be constrained to narrow depth intervals associated with particular hydrological conditions, such as the deep chlorophyll maximum, the thermocline (N. dutertrei) or the base of the photic zone (Globorotalia tumida) [Fairbanks and Wiebe, 1980; Ravelo and Andreasen, 1999; Mortyn and Charles, 2003]. There is no consistent relationship between the temperature anomalies recorded by the extraseasonal foraminifera and their shell weights, a potential indicator of depth of calcification within the water column. This implies that the extraseasonal foraminifera are not calcifying outside of the natural depth range of the seasonal foraminifera. Previous individual foraminifera analysis studies [Oba, 1990, 1991] from sediment traps in the Japan trench found no individuals outside of equilibrium calcite δ18OCaCO3 from the seasonal range in the surface waters (Globigerinoides sacculifer δ18OCaCO3 range of 3.0‰ from 185 individuals with an SST range of 12°C). A narrow (<1.8‰ δ18OCaCO3) range of thermocline dwelling Globorotalia inflata from the same study also confirms a lack of deep calcifying foraminifera.

[39] Vertical migration of the foraminifera through time is unlikely to create false positives. On a seasonal and interannual scale vertical migration is associated with attempts by individuals to remain in consistent hydrological conditions [Sautter and Thunell, 1991], which would serve to reduce the recorded δ18OCaCO3 values. Similarly, transect core top studies show that our foraminifera species maintain similar hydrological habitats rather than depth habitats with a deepening thermocline [Steph et al., 2009]. Long-term evolutionary vertical migration cannot be ruled out of any foraminiferal study, except where additional nonforaminiferal proxies such as alkenone SSTs exist. The similarity between G. ruber δ18O and the Lawrence et al. [2006] alkenone record from the core site suggests no change in this species depth habitat. We conclude that deep calcifying false positives potentially exist but are unlikely to play a significant role in our results considering the good agreement between the expected and measured percentages of extraseasonal foraminifera.

5. Conclusions

[40] In summary, evidence from individual foraminifera and HadCM3 simulations confirm the persistence of ENSO-like variability during the mPWP around a mean state characterized by a deeper thermocline, akin to a modern El Niño event. Mean thermocline depth changes accompany Pliocene climate cycles. HadCM3 provides additional insight into the more periodic and regular amplitude of mPWP ENSO-like variability compared to the more sporadic and unpredictable size of preindustrial events. Whether such variability will characterize the EEP in an anthropogenically warmed world requires an understanding of the mechanism driving this more periodic nature of the mPWP ENSO.

[41] So why is the thermocline deeper at this time? If the N. dutertrei remain thermocline dwellers in the deeper thermocline [Steph et al., 2009] then the mean δ18O change in the thermocline dwellers reflects a warming or change in location of the source waters of the Equatorial Undercurrent (EUC) which ventilates the equatorial thermocline. Rather than a lack of physical upwelling [Dekens et al., 2007] which likely began by 4.0 Ma [Steph et al., 2010]. HadCM3 simulates distinct warming (up to 3°C) at the location of modern subducting waters to the EUC. Despite the “Pleistocene-like” temperature and foraminifera assemblage of our MAMC Pliocene sample, the vertical gradient in δ13C appears more “Pliocene-like”: with a distinctive gradient between N. dutertrei and G. ruber of ∼1‰, comparable with other Pliocene samples. This MAMC sample suggests that the δ13C, although indicative of a stratified upper water column, is overprinted by an additional time-dependent process. The Pliocene EUC may be characterized by a more positive δ13C (nutrient poor) preformed signature [Cannariato and Ravelo, 1997], which is independent of the degree of upwelling in the EEP. Such a time evolution in the source waters may reflect tectonically induced changes in the Indonesian throughflow, which affect the source of the EUC [Cane and Molnar, 2001]. The trend over the last 40 years toward more regular, intense El Niño events [Trenberth et al., 2002] appears to parallel our Pliocene scenario. A future challenge is to unpick the different sensitivity of the variability in past ENSO to warming source waters of the EUC, a process likely to be induced by the nonequilibrium anthropogenic warming, compared to the more geologically equilibrated, tectonically driven changes in source waters to the EUC.

Appendix A

[42] The size of each analyzed population is a balance between gaining enough numbers to be sure of the result to a significant level whilst minimizing the time span over which the population is spread. There was a relatively low foraminiferal density in the core sediment and so 24 g wet weight of core sediment was used per time slice. Intact individual foraminifera suggest that bioturbation is not a significant factor at this core site. The populations described in Table A1 were collected and the number of successfully completed analyses, and consequently data points, is given in brackets. Small populations from HOLO are most likely due to the poor physical preservation of this sample, with large numbers of broken foraminifera in the large size fractions and foraminifera debris pervading even the smallest size fractions (<2 μm).

Table A1. Absolute Counts of the Three Target Foraminifera Species and the Number Analyzed in Parentheses
  • a

    Picked from the 355–425 μm and 300–355 μm size fractions from ∼24 g of core sediment (wet weight).

G. ruber0 (0)53 (32)57 (40)22 (17)51 (33)23 (17)39 (30)32 (27)
G. menardii3 (2)121 (39)71 (38)71 (39)62 (34)85 (30)51 (22)47 (36)
N. dutertrei67 (40)268 (40)253 (40)217 (40)14 (9)18 (15)572 (40)64 (40)


[43] We are grateful to Norman Charnley for help with analysis, Owen Green for foraminiferal expertise, and Paul Halloran and Jerome Groeneveld for helpful discussions. R.E.M.R. and A.M.H. acknowledge the Leverhulme Trust for the award of Philip Leverhulme prizes. S.G.B. acknowledges the Natural Environment Research Council for financial support during her Ph.D. This work used samples provided by the Ocean Drilling Program (ODP). The ODP (now IODP) is sponsored by the U.S. National Science Foundation and participating countries under management of the Joint Oceanographic Institutions (JOI), Inc. Practical work and writing were done by N.S., and modeling was done by S.G.B. and A.M.H. R.E.M.R. and A.M.H. provided insight, direction, and ideas to the work. All four discussed results and contributed to the manuscript. S.L. and M.H. provided significant lab assistance and data generation. The raw data has been deposited in Pangaea (Publishing Network for Geoscientifc and Environmental Data) at (doi:10.1594/PANGAEA.744735). The authors declare no financial competing interests.