Stable isotope data from eastern equatorial Pacific (EEP) core TR163-19 (2°15′N, 90°57′W, 2348 m) are presented for the surface-dwelling foraminifers Globigerinoides ruber and G. sacculifer and thermocline-dwelling Globorotalia menardii and Neogloboquadrina dutertrei. Using species-specific normalization factors derived from experimental and plankton tow data, we reconstruct a 360 kyr record of water column hydrography across the past three glacial cycles. We demonstrate that G. ruber maintains a mixed layer habitat throughout the entire record, while G. sacculifer records a mixture of thermocline and mixed layer conditions and G. menardii and N. dutertrei record thermocline properties. We conclude that G. sacculifer is not appropriate for paleoceanographic applications in regions with steep vertical hydrographic gradients. Results suggest that this region of the EEP had a thicker mixed layer and deeper δ13CDIC boundary between the surface and equatorial undercurrent during the last two glacial periods. A shift in N. dutertrei and G. sacculifer geochemistry prior to ∼185 kyr suggests water column structure and chemocline gradients changed, possibly due to a shift in the position of the undercurrent relative to this site. The timing and magnitude of glacial-interglacial δ13C variations between species indicates that near-surface carbon chemistry is controlled by changes in productivity, atmospheric circulation, and advected intermediate water sources north of the Antarctic polar front. These results demonstrate that when properly calibrated for species differences, multispecies geochemical data sets can be invaluable for reconstructing water column structure and properties in the past.
 An emerging view of the LGM EEP to the west of 110°W is that the thermocline depth was slightly shallower and the water column structure did not vary considerably from the present tropical Pacific subsurface structure [Andreasen et al., 2001; Andreasen and Ravelo, 1997]. Further to the east, LGM hydrography may have been considerably different as reflected by the presence of modern temperate and subpolar species in the Panama Basin water column [Andreasen et al., 2001] and changes in regional nutrient distribution and surface salinity [Loubere, 2001]. Radiolarian faunal assemblage changes and Mg/Ca analyses of the surface-dwelling planktonic foraminifera, Globigerinoides ruber (white), indicate reduced glacial surface temperatures (SST) [Koutavas et al., 2002; Lea et al., 2000; Pisias and Mix, 1997], which may be due to changes in low-latitude radiative forcing, advection of cool upwelled waters from the Peru Current to the south [Faul et al., 2000], intensified meridional winds driving deeper mixing of surface waters [Andreasen and Ravelo, 1997] or some combination of factors. These observations suggest that the eastern section of the EEP is very sensitive to atmospheric and ocean circulation changes on glacial-interglacial timescales.
 Modern hydrography in the EEP is controlled by four currents. The eastward flowing North Equatorial Counter Current (ECC) centered at about 6°N, the eastward flowing subsurface Equatorial Undercurrent (EUC) between 30–300 m depth at 0°, the westward flowing South Equatorial Current (SEC) between 8°S and 3°N and the Peru Current (PC) further south (Figures 1a and 1b) [Wyrtki, 1966]. The northerly component of the southeast trade winds drives the PC which flows northward along the Peru coastline and advects cool, nutrient-rich waters northwest into the SEC. Near the Galapagos Islands in the vicinity of site TR163-19 (2°15.5′N, 90°57.1′W, 2348 m water depth), high-salinity EUC water is deflected northward between 2° to 4°N and 95° to 92°W at 75 to 100 m depth [Lukas, 1986; Steger et al., 1998]. The water column in this region would therefore be expected to exhibit a strong halocline and thermocline (Figure 1c) [Millero et al., 1998].
 A suite of studies have previously examined the vertical distribution and isotope geochemistry of surface-dwelling (upper 100 m) foraminifera across the EEP [Curry et al., 1983; Fairbanks et al., 1982; Faul et al., 2000; Thunell et al., 1983; Watkins et al., 1998]. The basic premise behind these studies is that variations in the isotopic composition of foraminifera shells are primarily a function of physical properties of the water column such as temperature, salinity and δ13C of dissolved inorganic carbon (DIC), offset by environmental parameters such as pH or [CO32−] [Spero et al., 1997] or physiological processes such as foraminiferal respiration and symbiont photosynthesis [Bemis et al., 2000; Spero and Lea, 1993, 1996; Spero and Williams, 1988]. As depth increases, the δ18O of shell calcite increases due to decreasing temperature and increasing salinity (δ18Owater) (Figure 1d). In contrast, shell δ13C decreases with depth because phytoplankton elevate surface δ13CDIC as they discriminate against 13CO2 during photosynthesis, and heterotrophic organisms reintroduce 13C-depleted CO2 back into the DIC pool via respiration at depth [Kroopnick, 1974]. The result of community physiology on water column carbonate chemistry in this region is to create a surface to thermocline pH gradient [Millero et al., 1998] that will reflect the δ13CDIC gradient. In the modern EEP, the surface and EUC δ13CDIC end-members of this gradient are ∼1.8 and −0.2‰ [Fairbanks et al., 1982]. Because the salinity and therefore δ18Ow of the subsurface EUC is higher than that of the mixed layer, and these waters contain a low pH and δ13CDIC signature derived from the nutrient rich EUC and advected upwelled water of the PC, these geochemical gradients are steeper than are generally found elsewhere. The geochemical gradients found in the EEP should create an unambiguous signal when attempting to reconstruct changes in thermocline structure using multiple species of planktonic foraminifera.
 Four species that reside in the upper 150 m of the EEP water column are known to coexist through at least the past 360 kyr. Globigerinoides ruber (white) and G. sacculifer are spinose species that dwell in the mixed layer and upper thermocline from 0–50 m (Figure 1d) [Fairbanks et al., 1982, 1980; Faul et al., 2000; Ravelo and Fairbanks, 1992]. Two nonspinose species, Globorotalia menardii and Neogloboquadrina dutertrei, are common inhabitants of the EEP thermocline with partially overlapping depth habitats [Fairbanks et al., 1982; Faul et al., 2000]. All four species are known to possess symbionts [Bé et al., 1977; Gastrich, 1987; Mielke, 2001] which confines their primary life cycle to the euphotic zone. Whereas G. menardii resides within the upper to middle thermocline at depths between 25–85 m, N. dutertrei inhabits the middle to deep thermocline between 60–150 m. The habitats of these latter two species are thought to be closely coupled to the depth of the chlorophyll maximum [Watkins et al., 1998]. Together, the isotopic composition of these four species reflects the physical and chemical structure of the mixed layer and thermocline and can be used together to reconstruct changing properties of the near-surface EEP on G-I timescales.
 In this study, we combine laboratory and field data to explore the application of a multispecies approach to reconstruct the physicochemical properties of the EEP for the past 360 kyr at Cocos Ridge site TR163-19. Lea et al.  previously demonstrated that this site has maintained an average tropical SST between ∼23–28°C during the last three glacial cycles. Using the distinct vertical zonation differences between these four planktonic species, we reconstruct changes in thermocline structure, mixed layer depth and δ13CDIC gradients of the EEP water column back to marine isotope stage (MIS) 10. These results demonstrate that a multispecies approach to paleoceanographic problems, when combined with calibrated laboratory relationships, can yield new insights into water column and atmospheric processes that are not possible using the classical single species, single proxy approach.
 Fossil foraminifera were picked from Cocos Ridge core TR163-19 located just north of the cold upwelling water that characterizes much of the eastern equatorial Pacific (Figure 1). For downcore applications, G. ruber and G. sacculifer (without sac chamber) were picked from the 250–350 μm sieve fraction whereas G. menardii (600–850 μm shell length) and N. dutertrei (>500 μm shell length) were picked from the coarse (>350 μm) fraction. Because these latter two species are known to add a calcite crust of varying amount that can shift the isotopic composition toward deeper, colder environmental conditions [Bé et al., 1966; Hemleben et al., 1977; Sautter, 1998], we visually identified and selected specimens with minimal crusting for analysis. Oxygen and carbon isotope data from the larger size fractions of these two species show minimal variations with size [Billups and Spero, 1995; Bouvier-Soumagnac and Duplessy, 1985; Kroon and Darling, 1995]. For size comparison evaluation, large G. sacculifer (>650 μm shell length, without sac) were analyzed from a subset of intervals between the Holocene and the LGM. All G. ruber δ18O, Mg/Ca and SST data and the N. dutertrei δ18O data for 0–150 kyr were originally published by Lea et al.  and Spero and Lea . New data from G. sacculifer, N. dutertrei and G. menardii were obtained from the same sieved samples that were used for previous G. ruber and N. dutertrei analysis.
 The chronology for TR163-19 is based on correlation of the G. ruber oxygen isotope record to the standard SPECMAP chronology using 38 isotope substage tie points [Lea et al., 2000] with a small revision of the age model during the last 30 ka based on radiocarbon dates [Spero and Lea, 2002]. G. ruber was sampled at 5 cm resolution [Lea et al., 2000], equivalent to a potential resolution of about 2 kyr. G. sacculifer, N. dutertrei and G. menardii were sampled at 10 cm resolution throughout the core with the exception of the last 50 kyr and termination II (N. dutertrei and G. menardii) which were sampled at 5 cm resolution. These data are archived with the WDC Paleo at http://www.ngdc.noaa.gov/paleo/paleo.html.
 Between 10–20 shells of each species (30 shells for the larger G. sacculifer) were pooled, sonicated for 5–20 seconds s in methanol until clean, and roasted at 375°C for 30 min in vacuo. The samples were then reacted in supersaturated H3PO4 (sp. gr. 1.93) at 90°C using an Isocarb common acid bath autocarbonate device. The resulting CO2 was analyzed by a Micromass Optima isotope ratio mass spectrometer (IRMS). Oxygen and carbon isotope values are reported in ‰ notation relative to the V-PDB (Vienna Pee Dee Belemnite) standard where:
Analytical precision was ±0.05‰ and ±0.04‰ for δ18O and δ13C, respectively, (±1σ) based on repeat analyses of a NBS-19 calcite standard.
 Plankton tow samples of G. ruber (white), G. sacculifer and G. menardii were obtained from surface drift tows (10–30 m depth) in Exuma Sound, Bahamas (24°N, 76°W; August 1991) and southwest Puerto Rico (17.7°N, 67°W) in August 1997 and April–May 1999. Shells were measured, weighed and analyzed individually for δ18O and δ13C. Surface water samples from these collection periods were collected biweekly, poisoned with a few drops of a HgCl2 solution and sealed for δ13CDIC determination. For analysis, 5 mL of water sample was acidified in vacuo with ∼0.3 mL of H3PO4 and the resulting CO2 was cryogenically purified and analyzed with an Optima IRMS. Precision of the δ13CDIC measurements is ±0.03 ‰ as determined by repeat analyses of a laboratory DIC standard.
3. Results and Discussion
3.1. Stable Isotope Calibrations From Laboratory and Field Studies
 The use of “paleotemperature equations” to interpret δ18O data in paleoceanographic studies has been integrally linked to the concept of disequilibrium precipitation of calcite by foraminifera. Many correction factors have been developed from sediment trap samples to correct for species-specific offsets from equilibrium [Deuser, 1987; Deuser and Ross, 1989; Niebler et al., 1999]. An implicit assumption when using correction factors is that the equilibrium offset is constant across the full temperature range of the species. Given the range of corrections that have been published for each species, these offsets are poorly constrained, suggesting some of the basic assumptions are not valid. We contend that while the issue of equilibrium fractionation is important from a mechanistic point of view, the paleoceanographic community generally desires reproducibility and predictability. These objectives can only be attained with confidence using empirical relationships derived from laboratory experiments.
Erez and Luz  first demonstrated that experiments with living G. sacculifer could be used to generate empirical species-specific temperature:δ18O relationships for paleoceanographic applications. Although their relationship was similar to previously published “paleotemperature equations,” it did not adequately explain oxygen isotope data obtained from G. sacculifer in field studies [Bemis et al., 1998; Duplessy et al., 1981] and was not broadly applicable to other species without the application of “corrections” [Niebler et al., 1999]. Following the discovery that light (as it affects symbiont photosynthesis) and seawater carbonate ion concentration could affect shell δ18O [Spero, 1992; Spero et al., 1997; Spero and Lea, 1993], Bemis et al.  published a suite of empirical relationships from controlled experiments that produced the first field-applicable temperature:δ18O relationships for Orbulina universa and Globigerina bulloides. The average of their high-light and low-light O. universa relationships was identical to a relationship from Indian Ocean O. universa collected in plankton tows [Bouvier-Soumagnac and Duplessy, 1985]. Thunell et al.  further demonstrated that the laboratory based G. bulloides relationship is robust when applied to sediment trap δ18O data from the northeast Pacific. More importantly, their data showed that the high-light O. universa relationship was directly applicable to G. ruber (white) δ18O data, thereby establishing the species-specific relationship that we use here.
 We have now completed a new experimental temperature:δ18O calibration for G. sacculifer grown under high light (H. Spero et al., manuscript in preparation, 2003) and a new calibration for the nonspinose species G. menardii [Mielke, 2001] (Table 1). The G. sacculifer relationship predicts lower δ18O values at a given temperature than the original Erez and Luz  relationship, in agreement with field data. The G. menardii relationship is indistinguishable from an equation derived from G. menardii plankton tow data [Bouvier-Soumagnac and Duplessy, 1985] supporting its applicability to field samples. Because we have not yet cultured N.dutertrei, we use the plankton tow based temperature:δ18O relationship of Bouvier-Soumagnac and Duplessy  (Table 1) to establish the oxygen isotope offset of this species from the other three species in our study. The four empirical relationships (Figure 2) demonstrate that species-specific factors controlling calcification can create variable offsets of >0.5‰ depending on ambient temperature. When using a multispecies approach, one must normalize these differences in order to compare oxygen isotope data in a stratigraphic context (Table 2). We choose to normalize these oxygen isotope data to the G. ruber (O. universa high-light relationship) calibration. Although this is equivalent to applying correction factors, the fact that the slopes of the temperature:δ18O calibration relationships differ from one another precludes the use of a constant correction that does not consider calcification temperature.
Table 1. Temperature:δ18O Relationships Used to Compute Normalization Factors in Table 2
 In an ideal application of the normalization process, the ambient temperature for each species would be calculated using a proxy such as Mg/Ca [Lea et al., 1999b; Nürnberg et al., 1996], and the offset from G. ruber is determined for each interval in the core using the equations in Table 1. Questions related to water column temperature, salinity (as δ18Ow) and density gradients would then be possible as would questions regarding regional hydrological change through time [Lea et al., 2000]. Because Mg/Ca based temperature estimates are not yet completed for G. sacculifer, G. menardii and N. dutertrei from TR163-19, we assume that G. menardii and N. dutertrei grew at an average thermocline temperature of 15°C and G. sacculifer at 25°C (Figure 1c, Table 2). The error on the normalization factor for G. sacculifer and G. menardii would be <0.1‰ if temperature varied ±2°C around this estimate. For N. dutertrei, the factor error would be ∼0.15‰ for a variation of ±2°C.
 Laboratory experiments have demonstrated that the δ13C of foraminiferal calcite varies with symbiont photosynthesis, respiration and the carbonate ion concentration of seawater (see Spero  for review). The combined influence of these physiological processes shifts shell δ13C away from carbon isotopic equilibrium [Mulitza et al., 1999; Ortiz et al., 1996] which is generally thought to be approximately 1‰ enriched in 13C relative to that of δ13CDIC [Romanek et al., 1992]. In general, the practice of combining 10–30 foraminiferal shells for geochemical analysis in each interval of a deep sea core effectively averages the physiological component of the carbon (and oxygen) isotope signature, yielding values that are offset from δ13CDIC by some average amount. Although experiments can identify and quantify the influence of these processes under different environmental conditions, the sedimentary record reflects the combined, depth-integrated influence of different water column properties on an entire population. Carbon isotope data obtained from plankton tows or sediment traps are the best available sources of material for determining average population offsets from δ13CDIC.
 Plankton tow data from foraminifera collected in the tropical Atlantic and northern Caribbean demonstrate that species-specific population offsets from δ13CDIC exist in each species studied here (Table 2). These Δδ13Cshell-DIC offsets range from −0.94 ± 0.28‰ (n = 12) in white G. ruber (300–450 μm shell size) to +0.73 ± 0.23‰ (n = 28) in individually analyzed G. sacculifer from the 500–700 μm size range. Because individual analyses of G. menardii (>500 μm) show considerable scatter with a Δδ13Cshell-DIC of +0.18 ± 0.41‰ (n = 4) [Mielke, 2001] we will assign a zero offset from δ13CDIC. Data for N. dutertrei are not available from our studies. Therefore we use a Δδ13Cshell-DIC offset of +0.5‰ determined on South Atlantic specimens from plankton tows [Mulitza et al., 1999].
 For G. sacculifer, these field data must be further corrected to account for shell size differences between the plankton tows (500–700 μm) and fossil shells (250–350 μm sieve fraction). Carbon isotope analyses of fossil [Oppo and Fairbanks, 1989] and living G. sacculifer in both the field and laboratory [Hemleben and Bijma, 1994; Spero and Lea, 1993] demonstrate that larger G. sacculifer are enriched in 13C relative to smaller individuals. Spero and Lea  have argued that this size:δ13C relationship is due to growth under different light levels because of symbiont photosynthetic rate changes. Large 13C-enriched shells indicate growth in a shallow (warmer) high-light environment and vice versa. Cooler water temperatures found in deeper low-light environments, can also reduce shell size [Caron et al., 1987], but should have only a small effect on shell δ13C [Bemis et al., 2000]. Analyses of shells from plankton tows and core tops suggest the magnitude of the size offset is approximately 0.85‰ [Mulitza et al., 1999; Oppo and Fairbanks, 1989], yielding a final Δδ13Cshell-DIC offset of −0.12‰ for the small G. sacculifer analyzed in TR163-19.
 Although there is little effect of temperature on the δ13C of symbiotic foraminifera growing under low-light levels [Bemis et al., 2000] that might be found at depths below 30 or 40 m, shell δ13C in O. universa growing under high-light conditions (equivalent to <30 m) increases by 0.05‰ °C−1. We currently do not have a temperature:δ13C calibration for any of the species studied here and for this reason do not attempt to apply a temperature correction to the δ13C normalization factors in Table 2. Nevertheless, the potential for a temperature influence on shell δ13C should be considered when applying these factors elsewhere.
 Obviously, sieve dimension and shell size are not synonymous. Whereas foraminifera shells are measured along their longest dimension under the microscope, shells can slip through a sieve sideways thereby adding larger shells to the sieve size range. Because we did not measure the actual shell sizes of the fossil G. sacculifer prior to analysis, we recognize the computed offset may be slightly more negative than is appropriate for our samples. However, the potential error is small relative to omitting the size correction entirely. Shell size corrections are not necessary for the remaining three species because the calibration and downcore fossil size fractions are indistinguishable from each other.
3.3. TR163-19 Oxygen Isotope Records
 The oxygen isotope record in TR163-19 records three full glacial cycles through MIS 10 (Figure 3a). The four species show a clear depth ranking with G. ruber recording the lowest δ18O values indicating it occupies the shallowest habitat. G. sacculifer yields more positive δ18O values throughout the majority of the core suggesting it is inhabiting a slightly deeper/cooler [Dekens et al., 2002] and/or more saline environment than G. ruber. Both G. menardii and N. dutertrei record the most positive δ18O values in agreement with their known depth habitat in the cooler thermocline. Examination of the nonnormalized data would suggest that these two thermocline dwellers exchange depth habitats between MIS 7–9.
 Normalizing the oxygen isotope data of each species to G. ruber using the factors in Table 2 clarifies the relative depth stratification of the four species (Figure 3b). As in the nonnormalized data, G. ruber continues to exhibit the lowest (shallowest) values while G. sacculifer shows a slightly cooler and/or more saline habitat than indicated by the nonnormalized data. Midthermocline dweller, G. menardii records deeper environmental conditions than G. sacculifer and N. dutertrei occupies the deepest habitat as recorded by the most positive δ18O values in the grouping. These normalized data now reconcile the apparent G. menardii-N. dutertrei depth habitat reversal between MIS 7–9 by producing G. menardii values that are equal to or slightly lower (shallower) than those of N. dutertrei. During the last glacial cycle, the full range of δ18O values between G. ruber and N. dutertrei is ∼2.5–3‰ which reflects the combined influence of temperature and salinity differences between the surface mixed layer and thermocline EUC below 50 m. Interestingly, the range of oxygen isotope values prior to MIS 7 decreases to <2‰, with MIS 6 and 7 recording an intermediate range of values.
 The modified depth relationships produced by normalizing these δ18O data also clarifies their environmental interpretation. It is clear that the magnitude of the interglacial oscillations recorded by G. ruber are muted in G. sacculifer during the last glacial cycle (Figure 3b). Lea et al.  demonstrated that a large component of the G. ruber δ18O amplitude is due to SST variability in surface waters overlying the site of TR163-19, with the residual δ18Ow component dominated by the global ice volume signal [Lea et al., 2002]. Although a portion of the muted signal in G. sacculifer could be due to the lower sampling resolution relative to G. ruber (10 cm versus 5 cm) it is doubtful that this alone could explain the considerable reduction in δ18O amplitude during MIS 5. Rather, it is probable that a part of the difference between these two species is due to the addition of ∼30% gametogenic calcite on the G. sacculifer shell within the thermocline [Bé, 1980; Duplessy et al., 1981; Lohmann, 1995]. These data imply that a surface mixed layer signal is not fully recorded in the 250–350 μm size fraction of G. sacculifer as demonstrated previously [Dekens et al., 2002]. The normalized N. dutertrei record indicates it has maintained a cool, deeper environment throughout the record. Furthermore, the normalized Holocene N. dutertrei δ18O values agree well with the predicted δ18Ocalcite for a habitat depth of >100 m (Figure 1d).
 As noted above, the oxygen isotope data suggest there may have been a change in EEP water column structure near the MIS 6/7 boundary. Between the Holocene and ∼185 kyr, G. sacculifer records a muted surface signal and the habitat of G. menardii, as recorded by δ18O, is separated from that of N. dutertrei by ∼1‰. Prior to ∼185 kyr, G. menardii and N. dutertrei appear to coexist at the same habitat depth and G. sacculifer shows a consistent expression of δ18O amplitude that suggests it is recording the mixed layer SST component of the δ18O signal that is observed in G. ruber. For the thermocline species, the transition is abrupt and is driven by an increase in the absolute N. dutertrei δ18O values after 185 kyr (Figure 3b).
 Two interrelated scenarios could explain the shift in N. dutertrei δ18O around 185 kyr. If, the mixed layer were deeper in the earlier part of the record, we might expect a shoaling of the chlorophyll maximum to the top of the thermocline as phytoplankton respond to light limitation with increasing depth. Because the depth distribution of N. dutertrei and G. menardii are tightly coupled to the depth of the chlorophyll maximum and food availability [Fairbanks and Wiebe, 1980; Watkins et al., 1998], and the temperature and salinity gradients in the thermocline are steep, an average population shift on the order of only 10–20 m could be sufficient to produce the observed shift in N. dutertrei δ18O. Alternatively, it is possible that the gradient of the pycnocline below the mixed layer decreased due to a reduction in EUC/thermocline salinity and/or increase in temperature. These data would also be consistent with a shift of the EUC away from Site TR163-19 or extended El Niño-like conditions in the EEP between MIS 7 and 9. In the former case, the shift would most likely be toward the south because the current position of the core of the EUC is to the south of TR163-19. Although we cannot distinguish between these possibilities at this time, it may be possible to test these scenarios in the future by conducting Mg/Ca analyses on the different species to determine ambient temperature and to estimate δ18Ow [Lea et al., 2000].
3.4. TR163-19 Carbon Isotope Records
 Examination of the raw carbon isotope values (Figure 4a) demonstrates the primary dilemma facing researchers attempting to interpret nonnormalized multispecies data sets. Using a straightforward interpretation which assumes that the most positive δ13C values correspond to the surface-dwelling species, then the shallowest dwelling species would be G. menardii and N. dutertrei with G. sacculifer and G. ruber yielding similar values from a deeper environment. This ranking is the reverse of the oxygen isotope interpretation and is inconsistent with known depth distributions from plankton tow studies in this region [Fairbanks et al., 1982].
 Following normalization of the δ13C data to δ13CDIC based on the plankton tow offsets (Table 2), a different picture emerges (Figure 4b). The shallowest species, G. ruber, yields a core top δ13C of ∼2.5‰ and records δ13C oscillations of ∼0.8‰ throughout the record. Neogloboquadrina dutertrei has the most negative δ13C values in agreement with its deeper habitat. G. sacculifer records δ13C values that are either submixed layer or a mixture of mixed layer and thermocline information. The G. sacculifer values are generally equal to or slightly higher than those of N. dutertrei back to MIS 8. Prior to ∼200 kyr, G. sacculifer values vary between those of G. menardii and N. dutertrei. These data support our interpretation based on δ18O that the small size fraction (250–350 μm) G. sacculifer calcify a considerable part of their shell in the thermocline within the δ13CDIC chemocline (Figures 1c and 1d).
 Carbon isotope data from the midthermocline dweller, G. menardii, indicates it shifts between G. ruber surface values during glacial periods (e.g., MIS 2–4, 6) and intermediate (relative to N. dutertrei) δ13CDIC values during interglacial periods (e.g., MIS 5, 7 and 9) (Figure 4b). These shifts are most apparent on terminations I and II when the transition produces very low values that are identical to the δ13C of N. dutertrei. It is important to recognize that the oxygen isotope data do not support a G. menardii habitat transition into the mixed layer during glacial times (Figure 3b). Rather, because the δ13CDIC transition zone occurs across a narrow depth range that is situated within the upper pycnocline (Figure 1d) [Fairbanks et al., 1982; Millero et al., 1998], the δ13C shift we observe could be the result of a small change in water column habitat within the thermocline or shift in the depth of the δ13CDIC chemocline that is related to subtle changes in the structure of the pycnocline [Loubere, 2001]. We note that the excellent agreement between the δ13CDIC normalized values of G. menardii and those of G. ruber and N. dutertrei supports the application of the δ13C factors we use here.
 The thermocline signature recorded by G. sacculifer during the last glacial cycle raises serious questions about the use of this species as a recorder of surface water processes in regions of the EEP with steep hydrographic gradients. Laboratory data have demonstrated that G. sacculifer shell size is a function of irradiance with small shells growing under low-light levels and vice versa [Bé et al., 1982; Caron et al., 1981; Spero and Lea, 1993]. Hence the thermocline signature in the small (250–350 μm) size fraction specimens may be due to the unintentional selection of specimens that grew deeper in the water column under low-light conditions.
 To test this hypothesis, we analyzed large G. sacculifer (>650 μm shell size) from nine intervals across termination I for comparison to G. ruber and small size fraction G. sacculifer (Figures 5a and 5b). The results show the expected δ13C offset between G. sacculifer size fractions [Mulitza et al., 1999; Oppo and Fairbanks, 1989] that has been attributed to growth under different light levels and symbiont photosynthetic rates [Spero and Lea, 1993], with a mean Δδ13Clarge-small of 0.86 ± 0.28‰ (n = 9). This offset is identical to the size fraction correction we applied to the G. sacculifer normalization process (0.85‰; Table 2) based on the plankton tow data of Mulitza et al. , thereby confirming the assumptions that went into our size-based corrections. More importantly, the δ18O values from the larger G. sacculifer are considerably more negative than values for the smaller 250–350 μm G. sacculifer fraction. Although a small component of this offset (∼−0.3‰) may be due to the influence of symbiont photosynthesis on shell δ18O under high-light conditions [Bemis et al., 1998], this offset is too small to account for the observed difference between size fractions. We therefore conclude that the larger G. sacculifer inhabited a shallower, warmer, higher light environment as predicted by laboratory data. The mean offset between size fractions ranges from −1.25 ± 0.02‰ (n = 2) during termination I to −0.38 ± 0.17‰ (n = 7) during the LGM. The change in Δδ18O between size fractions during the last 30 kyr could relate to a change in mixed layer thickness during the LGM.
 After normalizing these oxygen and carbon isotope data with the appropriate size-dependent factors (assuming variations in symbiont photosynthesis control G. sacculifer size) (Table 2; Figures 5c and 5d), the δ18O data for large G. sacculifer place it near the habitat depth of G. ruber and shallower than the small G. sacculifer. We suggest that the difference between size fractions is due to a lower (warmer) δ18O value in the pregametogenic component of the larger shells. Because the final G. sacculifer shell δ18O values also include a substantial gametogenic calcite component that is added at a slightly deeper depth in the thermocline [Bé, 1980; Lohmann, 1995], we cannot confidently estimate the true pregametogenic depth habitat. However, we note that the normalized δ13C values for the two G. sacculifer fractions are indistinguishable. These combined data suggest that the small and large G. sacculifer are indeed depth segregated, but both are calcifying a considerable component of their shells within the δ13CDIC chemocline.
Oppo and Fairbanks  conducted a comparable study on different fractions of G. sacculifer in the Caribbean but did not observe a shell δ18O difference among size fractions. Unlike the EEP, the Caribbean is characterized by a deep mixed layer so habitat differences of 10–60 m should not lead to a δ18O difference. Interestingly, although the timing of the G-I δ18O change in the large fraction of G. sacculifer is similar to that of G. ruber (Figure 5a), the deglacial δ18O shift in the small size fraction begins ∼5 kyr later. We cannot explain this timing difference at present. Based on the combined G. sacculifer δ13C and δ18O datasets presented here, we must conclude that that this species is not appropriate for reconstructing surface water properties in the EEP or other tropical and subtropical regions with a shallow pycnocline and/or steep δ13CDIC chemocline.
 Two features of the TR163-19 δ13C stratigraphy for the three remaining species stand out. The first is the large rapid drop and subsequent recovery of G. menardii and N. dutertrei δ13C that occurs at the start of glacial terminations (Figures 6a and 6b). The magnitude of these events ranges between 0.5–0.7 ‰ in N. dutertrei and ∼1‰ in G. menardii. The second was noted earlier, a clear oscillation in the G. ruber δ13C record with spectral power at 111 and 40 kyr, that shows ∼0.8‰ lower values during the glacials than interglacials. Furthermore, the δ13C minima are absent in the surface G. ruber record during terminations I and II.
3.5. Decoupling Mixed Layer and Thermocline δ13C in the EEP
 Radiocarbon dating of the intervals bracketing the termination I δ13C minimum event shows that the initial decrease occurs shortly after 16,630 ± 50 14C years (19.8 ± 0.3 cal ka; 69 cm core depth) and reaches its minimum value at 13,250 ± 50 14C years (15.9 ± 0.2 cal ka; 54 cm core depth) (Figure 7) [Spero and Lea, 2002]. These radiocarbon ages were corrected by 635 yrs to account for the reservoir age in the vicinity of the Galapagos Islands. Whereas the event initiation age is consistent with the timing of initial Antarctic warming at the end of the LGM [Sowers et al., 1993], the age of the absolute minimum corresponds to the initial appearance of high-δ13C North Atlantic Deep Water (NADW) in the deep South Atlantic (13.1 14C ka) [Charles and Fairbanks, 1992]. Based on this temporal agreement, Spero and Lea  proposed that the gradual increase in δ13C in the N. dutertrei record reflects the onset of North Atlantic thermohaline circulation and the increasing contribution of low-nutrient, high-δ13C NADW to SAMW at the polar front. The fact that the timing of the N. dutertrei δ13C minimum at TR163-19 is identical to the age of the minimum event in upwelled South Atlantic waters off Namibia (13.2 14C ka) [Schneider et al., 1992] and the arrival of NADW in the South Atlantic, attests to the rapid propagation of this geochemical signal into the EEP via SAMW and the EUC.
 As noted earlier, the G. menardii δ13C minimum events across the last two glacial terminations are nearly twice as large as those of N. dutertrei (Figure 7). Interestingly, G. menardii δ13C begins to decrease at ∼21.5 ka, or ∼1.5 ka prior to the initial N. dutertrei decrease. We hypothesize that the magnitude of the G. menardii signal is due to the combination of a shoaling of the δ13CDIC chemocline that would accompany a decrease in mixed layer depth, combined with a change in the δ13CDIC of the thermocline/EUC via the mechanism discussed above. However, because the G. menardii δ13C signal begins to decrease prior to that of N. dutertrei on both terminations, the hydrographic structure of the mixed layer and thermocline may have begun to change prior to the initiation of mixed layer temperature rise. A shift in tropical hydrography that predates SST rise in the EEP and Southern Ocean/Antarctic warming [Lea et al., 2000; Spero and Lea, 2002] could reflect late glacial changes in low-latitude atmospheric circulation that have been reported from Lakes Titicaca and Junin in Peru [Seltzer et al., 2002] and a transition from a glacial El Niño-like to post-glacial La Niña state for the tropical Pacific [Koutavas et al., 2002; Stott et al., 2002]. The early timing of these transitions supports hypotheses that argue for the tropical Pacific as the initiator of global climate change during glacial cycles [Cane and Clement, 1999].
 The δ13C minima events recorded by N. dutertrei and G. menardii at TR163-19 are conspicuously absent in the G. ruber δ13C record from terminations I and II (Figures 6b and 7). Rather, G. ruber δ13C oscillates with ∼0.3–0.4‰ amplitude during the δ13C minimum event that is followed by a 0.8‰ increase that is comparable to the δ13C increase observed in the thermocline dwellers. On orbital timescales, G. ruber displays a quasi 41 kyr periodicity with reduced values during glacials (MIS 2–4, 6) and elevated values during interglacials (MIS 1, 5, 7). During the last glacial cycle at least, we can eliminate local upwelling as the cause for the cyclic change in G. ruber δ13C because the N. dutertrei δ13C record remains relatively constant between 100 kyr and the initiation of the δ13C minimum event. One possibility is that the G. ruber δ13C values decrease due to an increase in surface water pH or [CO32−] [Spero et al., 1997] that may have accompanied the reduction of atmospheric pCO2 to glacial levels at ∼70 kyr [Lea et al., 1999a; Petit et al., 1999]. Unfortunately it is not possible to use G. ruber and G. sacculifer δ13C values to deconvolve the carbonate ion effect (CIE) from the δ13CDIC of surface waters [Spero et al., 1999] because these two species live in different regions of the water column with different ambient δ13CDIC values. However, the slope of the CIE on G. ruber, −0.0089‰/μmol [CO32−] [Spero et al., 1999], would produce a ∼0.8‰ decrease in shell δ13C if we assume an average increase of ∼90 μmol kg−1 [CO32−] for glacial surface waters [Lea et al., 1999a]. Although the influence of [CO32−] could explain the low glacial G. ruber values, we note that boron isotope data from core V19-28 (2°22′S, 84°39′W) suggests that the surface pH in the EEP was unchanged within error (±0.1 pH unit) across both termination I and II [Sanyal et al., 1997]. If these δ11B results also reflect surface conditions slightly to the northwest at site TR163-19, they would argue against a CIE contributing to the δ13C shifts we observe.
 Because the upper water column at TR163-19 is strongly stratified most of the year [Levitus and Boyer, 1994], and DIC exchange between the surface and thermocline is minimized, reduced δ13C values during cold glacial periods could derive their source from low-δ13C water that advects into this region from the south [Feldberg and Mix, 2002]. Data from the Cariaco Basin in the Caribbean suggests that trade wind strength was greater and the position of the intertropical convergence zone (ITCZ) was further south during glacials [Yarincik et al., 2000]. Elevated wind speeds over the EEP may have strengthened the Peru upwelling system, thereby permitting low-δ13C water to advect northward into the mixed layer above TR163-19. However, this mechanism would be inconsistent with the recently proposed El Niño-like state for the glacial Pacific [Koutavas et al., 2002; Stott et al., 2002] and our earlier conclusion that mixed layer depth was deeper at TR163-19 during the glacial period. This leaves us with a dilemma that requires additional data and a greater regional distribution of G. ruber δ13C patterns to explain.
 The oscillation in G. ruber δ13C across the last glacial termination begins in the same interval that G. menardii δ13C starts to decrease, 21.5 cal kyr, and continues until ∼14 cal kyr when G. ruber δ13C increases continuously into the early Holocene. Data collected from gravity core P7GC in the Panama Basin (2.6°N, 84.0°W) show that organic carbon mass accumulation rates (MAR) increased abruptly at ∼22 ka, peaked at 16 ka, and finally decreased to Holocene levels by 12 ka [Farrell et al., 1995]. We speculate that the δ13C oscillations recorded by G. ruber across termination I could reflect the influence of regional productivity variations. Interestingly, the initial MAR increase in P7GC is consistent with the timing of G. menardii δ13C reduction suggesting changes in Panama Basin MAR may have been influenced by the changing water column hydrography that we see at TR163-19.
 Data from termination III and IV yield a different picture of δ13CDIC gradients in the older section of TR163-19. Here G. ruber records the δ13C minima spikes that are present in the thermocline dwellers, and G. menardii maintains an intermediate depth which does not completely cross the δ13CDIC chemocline as it does during terminations I and II. Because all 4 species record the δ13CDIC minima across terminations III and IV, these minimum events must represent a common environmental signal that mixes into the surface layer. For this signal to be absent from the mixed layer during termination I and II would imply counter-balancing productivity and/or water column mixing changes, both of which are possible [Andreasen et al., 2001; Farrell et al., 1995; Loubere, 2000].
 In agreement with the δ18O records (Figure 6a), these δ13C data suggest that the water column overlying the site of TR163-19 appears to have experienced a structure change during the last 360 kyr. Jian et al.  have suggested a similar idea noting that opal content in the South China Sea doubled or tripled after MIS 7. The abrupt decrease in N. dutertrei δ13C and increase in δ18O at ∼180 kyr could reflect a change in the nutrient content and/or source of the EUC/thermocline waters (e.g., North Pacific versus Southern Ocean intermediate water) at this site. Alternatively, the abrupt shift in isotope values could reflect a change in the position of the northern EUC jet away from Site TR163-19. Further studies examining the differences between the last four glacial terminations in the tropical Pacific may help identify which changes are likely to have affected this region.
 Utilizing laboratory and field-derived normalization factors, we demonstrate that oxygen and carbon isotope data from multiple species of foraminifera can be used to reconstruct water column hydrography in the EEP over the last 360 kyr. The combination of data from mixed layer and thermocline-dwelling species yields a dynamic view of local thermocline changes on sub-Milankovich timescales that is not possible using the more typical single species approach. Although the geochemistry of G. ruber, G. menardii and N. dutertrei yield depth-specific information that can be used to reconstruct upper water column physiography and carbon isotope chemistry in the EEP, interpretation of data from G. sacculifer is equivocal because the shell geochemistry is controlled by a mixture of surface and thermocline conditions. We conclude that G. sacculifer is not appropriate for paleoceanographic applications in regions with steep vertical hydrographic gradients.
 We demonstrate that the EEP in the vicinity of TR163-19 underwent a major shift in water column stratification at ∼185 kyr when the geochemistry of N. dutertrei and G. menardii indicate a large increase in chemocline and thermocline gradients that are similar to those seen today. Based on G-I shifts in G. menardii carbon isotope chemistry, we hypothesize that the mixed layer was thicker and δ13CDIC chemocline boundary was deeper during glacial periods than during warm interglacials. The timing and magnitude of G-I δ13C variations between species suggests that near-surface carbon chemistry is controlled primarily by changes in productivity, atmospheric circulation and source waters for SAMW/AAIW at the Antarctic polar front.
 We thank J. Kennett for samples, J. Bijma, R. Zeebe, A. Wischmeyer, S. Barker and B. Hönisch for comments and suggestions, D. Gates, A. Schilla, P. Dekens, A. Davé, P. von Langen, S. Duncan, and M. Thomas for sample preparation and L. Juranek and D. A. Winter for mass spectrometer operation and maintenance. The reviews and comments of K. Faul and an anonymous reviewer are gratefully acknowledged. This material is based upon research supported by the National Science Foundation under grants OCE-9903632 and OCE-9729203 (HJS) and OCE-9729327 (DWL) and a fellowship to HJS by the Hanse Institute for Advanced Study, Delmenhorst, Germany.