Optimal use of Mg/Ca as a paleotemperature proxy requires establishing calibrations for different species of foraminifera and quantifying the influence of dissolution. To achieve this goal, we have measured Mg/Ca and δ18O in a series of tropical and subtropical core tops, including four depth transects: the Ceara Rise, the Sierra Leone Rise, and the Rio Grande Plateau in the Atlantic, and the Ontong Java Plateau in the Pacific, focusing on spinose mixed layer dwelling species Globigerinoides ruber and Globigerinoides sacculifer, and nonspinose thermocline dwelling Neogloboquadrina dutertrei. Shell Mg/Ca in G. sacculifer is 5–15% lower than in G. ruber, while N. dutertrei Mg/Ca is 49–55% lower than in G. ruber. This statistically significant offset has allowed us to establish different calibrations for each species. Multilinear regression analysis was used to develop calibration equations that include a correction term for the dissolution effect on Mg/Ca in foraminiferal calcite. Presented in this paper are two sets of calibrations; one set using core depth as a dissolution correction and another using ΔCO32− as a dissolution parameter. The calibrations suggest that G. ruber is the most accurate recorder of surface temperature, while G. sacculifer records temperatures below the surface at 20–30 m. The depth habitat of N. dutertrei is more uncertain, owing to the wide range in habitat depths depending on hydrographic conditions, but on average, Mg/Ca and δ18O data suggest it is at ∼50 m. Of the three species, N. dutertrei is the most sensitive to dissolution (up to 23% decrease in shell Mg/Ca per km), while G. sacculifer is the most resistant.
 Several proxies for sea surface temperature (SST) have been developed for use in paleoceanographic studies, including foraminiferal transfer functions, δ18O of foraminiferal calcite, and alkenone unsaturation ratios (U37k′). Because of the limitations of these tools, there has been considerable interest in developing additional temperature proxies. Specifically, the δ18O of foraminiferal calcite is complicated because it combines signals of both calcification temperature and the δ18O of the water in which the calcite was precipitated [Epstein et al., 1953]. The δ18O of water changes with glacial-interglacial cycles owing to the influence of ice volume changes [Shackleton and Opdyke, 1973]. Therefore using δ18O of foraminiferal calcite as a temperature proxy requires an estimation of ice volume. Regional differences in the precipitation-evaporation balance also influence salinity, and hence δ18O of water, thereby further complicating the signal. The alkenone temperature proxy is limited to regions with sufficient input of organic carbon [Herbert, 2001], and foraminiferal transfer functions suffer from the so- called “no modern analogue” problem [Mix et al., 1999].
 Considerable interest exists in the role of the tropical oceans in climate change, and, in particular, there is still significant controversy regarding the magnitude of tropical SST change on glacial-interglacial timescales [Crowley, 2000]. Because the previously mentioned temperature proxies arrive at different estimates of SST change and because tropical SST changes are likely to be small, an additional SST proxy is needed. Mg/Ca in foraminiferal calcite is proving to be a promising temperature proxy in tropical settings due to its sensitivity to temperature changes as well as its applicability in low organic carbon sediments. In addition to providing a less complicated SST signal, Mg/Ca is also potentially useful in separating the temperature and δ18O of water signal in calcite δ18O.
 Although culturing experiments provide good support for using Mg/Ca as a paleo-temperature proxy, there is a need to develop calibrations from sediment samples [Elderfield and Gannssen, 2000]. Fossil foraminifera in marine sediment are the source for paleoceanographic studies, and it is therefore necessary to have a clear understanding of how postdepositional processes affect Mg/Ca in foraminifera. Of particular concern is evidence that dissolution lowers Mg/Ca in sediment samples [Lorens et al., 1977; Rosenthal et al., 1993; Russell et al., 1994; Brown and Elderfield., 1996; Hastings et al., 1998]. Box core samples from the Ontong Java Plateau, for example, demonstrate that Mg/Ca in some species of planktonic foraminifera decreases with increased water depth because of the lower calcite saturation state of deeper waters [Brown and Elderfield, 1996]. Experiments simulating dissolution also demonstrate decreasing shell Mg/Ca with increasing dissolution in planktonic foraminifera [Hecht et al., 1975].
 Dissolution of calcite can occur within the water column, at the water-sediment interface, and within the sediments and is observed when the in situ concentration of carbonate ion is less than the saturation concentration of carbonate ion (ΔCO32− = [CO32−]in situ − [CO32−]saturation) [Berger et al., 1982]. ΔCO32− in deep water results from the relative age of bottom waters and is the main control on calcite dissolution within the water column and at the sediment-water interface. Additional effects on ΔCO32− within the sediments result from the degradation of organic matter in the sediments, which releases metabolic acids that titrate CO32−, resulting in a smaller ΔCO32− [Emerson and Bender, 1981; Hales and Emerson, 1996, 1997]. Increased Mg/Ca in calcite increases the calcite dissolution susceptibility, and dissolution of Mg rich calcite may occur well above the lysocline [Brown and Elderfield, 1996].
 Although several researchers have noted the dissolution effect on Mg/Ca, little effort has been made to correct for dissolution in paleoceanographic application of the proxy. Instead, most researchers have chosen to apply Mg/Ca only at locations where cores well above the calcite lysocline are available. As demonstrated by Brown and Elderfield  and Lea et al. , however, dissolution can affect Mg/Ca at depths significantly shallower than the lysocline. It appears that the Mg rich parts of a shell are dissolved preferentially to those parts with lower Mg concentration [Rosenthal et al., 2000].
 Our data set includes four depth transects: the Ontong Java Plateau in the equatorial Pacific, the Ceara Rise and the Sierra Leone Rise in the equatorial Atlantic, and the Rio Grande Plateau in the subtropical South Atlantic (Figure 1, Table 1). These sites have different surface hydrography and climatology as well as contrasting calcite preservation in their sediments. The inclusion of the depth transects, with the exception of the Rio Grande Plateau (see section 3.3), allows us to develop calibration equations that include a preservation correction for Mg/Ca, parameterized as a depth correction and as a ΔCO32− correction. Our extensive data set has allowed us to develop a more precise calibration for SST between temperatures of 20° and 30°C. These calibrations, in addition to accounting for dissolution, render Mg/Ca of foraminiferal calcite a more robust paleothermometer for tropical settings.
Table 1. Core Top Locations, Depths, and Estimates of Age of Core Top
Water Depth, m
δ18Owater SMOW ‰
Depth in Core, cm
Approximate Age of Core Top
Reference for Age of Core Top
Methods for approximating the age of the core tops from previous studies include δ18O records.
The 14C dating.
The %CaCO3 records.
The 14C dates are from box cores from the same location. Core tops with no known previous estimate of age were determined to be recent if the δ18Ocalcite−δ18Owater of G. ruber was not greater than two standard residuals from the mean regression of δ18Ocalcite−δ18Owater and SST for all core tops.
 We analyzed Mg/Ca and δ18O in foraminifera shells from 24 core tops in the tropical and subtropical Atlantic and 18 core tops in the tropical Pacific (Figure 1, Table 1). Fifty to sixty shells of G. ruber (white) and G. sacculifer (without sac-like final chamber) were picked from the 250–350 μm fraction, and 40–50 shells of N. dutertrei were picked from the same size fraction. Of those shells, 10–15 were separated for isotopic analysis. The remaining shells (∼0.7–0.9 mg) were gently crushed between two glass plates and separated into two aliquots. The crushed shells were then cleaned of contaminating phases using a rigorous multistep process designed for trace metal analysis [Boyle and Keigwin, 1985/1986; Mashiotta et al., 1999]. Samples were handled under trace metal clean conditions. Splits of the sample were analyzed in separate runs, and the results were compared for reproducibility.
 The clean samples were simultaneously dissolved and spiked in 500 μL of a multielement spike. The sample was then injected into a Fisons/VG PlasmaQuad 2+ inductively coupled plasma-mass spectrometry (ICP-MS) coupled to a Cetac ultrasonic nebulizer. Mg/Ca is determined by simultaneous isotope dilution and internal standardization, using 25Mg and 45Sc. Long-term reproducibility is estimated as ±2.1% (1σ) based on 268 runs of the consistency standard CN3 (Mg/Ca = 1.56 ± 0.03 mmol/mol) over a 2 year period. The average reproducibility of sample splits in our data set was ±0.08 mmol/mol (±3.3%).
 For analysis of δ18O in foraminifera, 10–15 shells were sonicated for ∼5 s in methanol and roasted at 375°C for 30 min in vacuo. The samples were then reacted individually in 105% H3PO4 at 90°C using an Isocarb common acid bath autocarbonate device. The CO2 was analyzed using a Fisons Optima isotope ratio mass spectrometer. The analytical precision of the measurements is ±0.06%, determined by replicate analyses of NIST-19 and a second Carrara marble laboratory standard.
 There are three major trends in the Mg/Ca data: (1) there is a clear increase in Mg/Ca with increased mean annual SST in both the Pacific and Atlantic, although an offset between the basins is apparent (Figure 2); (2) there is a persistent offset between the three different species (Figures 2 and 3), indicating the importance of differences in water column depth habitat; and (3) shell Mg/Ca decreases with increased water depth (Figure 3). These trends demonstrate the importance of differences in species habitat depths, as well as the importance of dissolution on Mg/Ca as a temperature proxy. The offset in the δ18O of the three species generally confirms the pattern noted in the Mg/Ca data (Figures 4 and 5), and more positive δ18O with increased water depth indicates the influence of dissolution [Wu and Berger, 1989].
3.1. Relationship Between Mg/Ca and Temperature and δ18O and Temperature
 Regressions were performed using mean annual temperature data. Although any one individual foraminifera has a lifespan of ∼1 month [Bijma et al., 1990], the analysis of 40–60 shells from a core top yields an average that is broadly representative of annual conditions. Mean annual temperatures were also a better statistical fit to the data for all three species.
 The increase in Mg/Ca with higher SST for both G. ruber and G. sacculifer is most clearly demonstrated in the equatorial Pacific (Figure 2). The trend in the Pacific basin appears to be exponential, agreeing with previous calibration work on planktonic foraminifera [Nürnberg et al., 1996a; Lea et al., 1999; Mashiotta et al., 1999]. The range in Mg/Ca at 29°C in the Pacific is due to increased dissolution along the Ontong Java Plateau depth transect (Figures 3a and 7a). Data in the Atlantic also indicate an increase in Mg/Ca with increased SST, but the relationship is complicated by several factors. The data at the lower temperature range (∼22°C) are from the Rio Grande Rise, a subtropical site with a large seasonal SST signal (see section 3.3). The Atlantic data are also complicated by the wide range in Mg/Ca at SST ∼27.5°C. This range is due to increased dissolution with increased depth along the Ceara Rise and the Sierra Leone Rise, as well as an offset between these two sites (Figures 3 and 7) (see section 3.3). Despite these complications, however, the overall trend in Mg/Ca is clear and is statistically demonstrated to be exponential (see section 4.1).
 To compare the δ18O of foraminiferal calcite to SST, local salinity effects on the δ18O of water must first be subtracted from δ18Ocalcite. To do this we used previously established equations relating salinity to δ18Owater in the western and eastern Atlantic [Fairbanks et al., 1992], the Caribbean [Ruhlemann et al., 1999], and the eastern [Fairbanks et al., 1982] and western [Fairbanks et al., 1997] equatorial Pacific. These relationships were derived from representative locations of each of the respective basins and therefore likely represent a basin-wide trend. The calculated δ18Owater is also in good agreement with estimates from the Global Seawater Oxygen-18 Database (G. A. Schmidt, G. R. Bigg, and E. J. Rohlingn available at (http://www.giss.nasa.gov/data/o18data/) [Schmidt, 1999]. It is important to keep in mind, however, that local changes may also strongly influence the δ18Owater — salinity relationships. Additionally, the equations derived for the different basins are strictly a reflection of surface water. If particular species of foraminifera are living deeper in the water column, the calculations of δ18Owater may introduce an error.
 The trend toward more negative δ18Ocalcite−δ18Owater with increased temperature, as expected from thermodynamic fractionation [Kim and O'Neil, 1997; Bemis et al., 1998], is apparent for both G. ruber and G. sacculifer (Figure 4). The wide range of δ18O at 29°C in the Pacific and at 27.4°C in the Atlantic can be attributed to more positive δ18O with increased depth on the Ontong Java Plateau (OJP) and the Ceara Rise and Sierra Leone Rise, respectively (Figure 5) (section 3.3).
3.2. Differences Between Species
 Shell Mg/Ca in G. sacculifer is 5–15% lower than in G. ruber; N. dutertrei Mg/Ca is 49–55% lower than in G. ruber (Figures 2 and 3), in agreement with previous studies [Bender et al., 1975; Rosenthal and Boyle, 1993]. Statistical analysis indicates that this difference is significant (see section 4.1). The offset between G. ruber and G. sacculifer is likely related to two factors. First, G. sacculifer has been shown to add more gametogenic calcite (∼28%) [Bé, 1980] than G. ruber, which shows no evidence of gametogenic calcite [Caron et al., 1990]. Because gametogenic calcite is added at depth [Bé, 1980], it records colder temperatures, therefore decreasing the average Mg/Ca of the shell of G. sacculifer. Second, δ18O studies [Fairbanks et al., 1982; Faul et al., 2000] indicate that G. sacculifer lives at greater depths than G. ruber, thereby generally recording cooler temperatures. N. dutertrei is a thermocline dwelling species [Bé, 1977; Fairbanks et al., 1982; Faul et al., 2000] and has been shown to inhabit a wide depth range [Fairbanks et al., 1980, 1982; Sautter et al., 1991]. N. dutertrei consequently records colder temperatures, and the calibration equation for this species must therefore be developed using temperatures at depth rather than SST (see section 4.1.3).
 The δ18O of N. dutertrei is consistently 1–1.5‰ more positive than that of G. ruber (Figure 5), a trend which is consistent with the offset seen in shell Mg/Ca and with our current understanding of the depth habitat of N. dutertrei. The offset in the δ18O between G. sacculifer and G. ruber is not as large as is expected from the Mg/Ca data, particularly along the OJP (see section 4.3).
3.3. Transect Data
 Data from the depth transects illustrate both the offset between species and the increasing dissolution effect with water depth on shell Mg/Ca (Figure 3). The depth transects are from different parts of the ocean and therefore exhibit both common trends and regional differences. The Ontong Java Plateau, the site with the most intense dissolution as demonstrated by ΔCO32− values (Figure 7), shows a linear decrease in Mg/Ca with depth for all three species (Figure 3). The mean annual SST at this location does not vary significantly between the different core locations (29.1°–29.3°C); therefore, assuming a similar age for the core tops (Table 1), the decrease in Mg/Ca must be due to dissolution rather than temperature. Evidence of dissolution is observed even in the shallowest cores, which are almost 2000 m above the lysocline in this region (3400 m [Berger et al., 1982; McCorkle et al., 1995]), where ΔCO32− is ∼20 μmol/kg. G. ruber appears to be more susceptible to dissolution at this site than at any other location, showing a decrease in Mg/Ca of ∼14%/km along the Ontong Java Plateau, compared to 5%/km and 7%/km at the Ceara Rise and the Sierra Leone Rise, respectively (percentage change calculated as an exponential). Mg/Ca of G. sacculifer decreases by 5%/km, and N. dutertrei Mg/Ca decreases by 20%/km along the Ontong Java Plateau.
 The δ18O for all three species also shows evidence of dissolution along the Ontong Java Plateau. The δ18O of G. ruber increases by 0.2‰/km, G. sacculifer δ18O increases by 0.2‰/km, and N. dutertrei δ18O increases by 0.3‰/km (Figure 5). Wu and Berger  demonstrated this trend previously and noted two clusters of data points for G. sacculifer, one group shallower than 2.9 km with δ18O values of −2.1‰ and one group deeper than 2.9 km with δ18O of −1.7‰ [Wu and Berger, 1989]. Our data clearly indicate that dissolution affects the δ18O of all species well above the lysocline and that the increase in δ18O is gradual rather than a dramatic shift at any particular depth. This trend toward more positive δ18O with increased core depth can lead to a low temperature bias with increased depth, thus influencing temperature estimates in the same direction as the dissolution influence on Mg/Ca. The bias in δ18O is potentially as large as 4°C for G. sacculifer between the shallowest and deepest core along the Ontong Java Plateau.
 The dissolution trend observed along the Ontong Java Plateau can also be seen in the Ceara Rise and the Sierra Leone Rise (Figures 3 and 7). Unlike the Pacific data, however, the transects in the equatorial Atlantic do not show a clear decrease in shell Mg/Ca in cores shallower than ∼4 km. The dissolution trend toward more positive δ18O is also not seen shallower than 4 km in the Atlantic, although the transition at 4 km is not as clear as it is in Mg/Ca. This difference in the pattern of dissolution between basins is consistent with the difference in calcite saturation. The profile of ΔCO32− with depth at a representative Pacific site (the Ontong Java Plateau) and a representative Atlantic site (the Ceara Rise) indicates a 2.5 km difference in the depth of the calcite saturation horizon between the different basins (Figure 6). It is interesting to note that ΔCO32− is ∼20 μmol at the depths where we first infer dissolution (from Mg/Ca and δ18O) along the Ceara Rise (∼4 km) and the Ontong Java Plateau (∼1.5 km) (Figures 6 and 7).
 The mean annual SST at both the Ceara Rise and the Sierra Leone Rise is 27.4°C [Levitus and Boyer, 1994]. However, Mg/Ca is greater at the Ceara Rise (CR) than at the Sierra Leone Rise (SLR) for all three species at all depths (Figure 3). Because the mean annual SST is the same at both sites and seasonal variations of SST is not great at either site (∼0.6°C at the CR and ∼1.0°C at the SLR [Levitus and Boyer, 1994]), the offset between the two is inferred to be due to better preservation at the Ceara Rise. The δ18O data at these two sites support this hypothesis because δ18O is more positive at the Sierra Leone Rise, also suggesting better preservation at the Ceara Rise. Both Mg/Ca and δ18O data bias the temperature estimates of the Sierra Leone Rise toward lower temperatures. Preservation differences between these two sites should be reflected in the measurement of ΔCO32−. However, Mg/Ca at the Ceara Rise are consistently greater than those at the Sierra Leone Rise where ΔCO32− values are the same (Figure 7). The degree of bottom water calcite saturation may not be able to account for the offset between the two sites because (1) ΔCO32− in bottom waters does not account for changes in ΔCO32− within the sediment owing to the degradation of organic matter in sediments, which causes dissolution [Hales and Emerson, 1997] and/or (2) the World Ocean Circulation Experiment (WOCE) sites from which data was used to calculate ΔCO32− may not be proximal enough to the core locations to accurately record the ΔCO32− of the bottom waters at these sites (Table 2).
 No evidence of dissolution is apparent in trends of Mg/Ca data from the subtropical Rio Grande Plateau where all cores are located well above the lysocline in this region (∼4 km [Jones et al., 1984]). The offset between the three different species is apparent in the Mg/Ca data at this site. The δ18O offsets between the species are not as clear, however, and G. ruber has a somewhat more positive δ18O than G. sacculifer (Figure 5). This would seem to imply that G. ruber is living at deeper depths than G. sacculifer, which is not in agreement with other sites in this study or previous studies [Fairbanks et al., 1982; Faul et al., 2000]. Additionally, the Mg/Ca of all three species is higher than would be expected for this location, where the mean annual SST ranges between 21.2° and 22.4°C. The seasonal SST range at this site is quite large (18°–24°C [Levitus and Boyer, 1994]), so tropical species probably thrive only during the warmer season. If this were the case, the δ18O data should record more negative values, but this does not appear to be the case.
Table 2. Latitude and Longitude of Core Locations and WOCE Sites Used to Calculate ΔCO32−a
WOCE Cruise Number
Latitude of Station
Longitude of Station
The ΔCO32− and [CO32−]sat were calculated using the Program Developed for CO2 System Calculations by E. Lewis and D. W. R. Wallace. Constants K1 and K2 from Roy et al. , constant KSO4 from Dickson  and Ksp for calcite from Mucci .
OCE 173-4 “G”
 An additional complication of the Rio Grande Rise data is the trend toward more negative δ18O values with increased depth. This trend goes in the opposite direction from the dissolution trend demonstrated in the other transects (Figure 5) as well as previously published work [Wu and Berger, 1989]. The very tops of the cores were not available at this location, and because the sedimentation rates at the Rio Grande Rise are low (≤1 cm/1000 years), our samples (between 1 and 7 cm; Table 1) are probably not representative of modern SST regimes and likely sample different ages along the transect. Additionally, previous work indicates that bioturbation has caused some mixing of glacial material with the tops of the cores [Lohmann, 1995]. Because of these complications, we determined that data from the Rio Grande Rise may not be accurately recording modern conditions, and therefore it is not included in the statistical analysis developed in section 4.
4.1. Mg/Ca Calibrations Using Core Depth as a Dissolution Correction
 The current major limitation of the Mg/Ca paleotemperature proxy is the known dissolution effect, which decreases Mg/Ca of foraminifer shells, thereby biasing the temperature estimate to colder values. The goal of this section is to establish a calibration relating shell Mg/Ca to temperature while taking into account the effect of increased dissolution with increased water depth, as well as basin-basin differences in bottom water calcite saturation. Although calcite dissolution is driven by carbonate chemistry, the relationship between Mg/Ca and depth is very apparent, and therefore the equations we developed in this section use depth as a dissolution correction (see section 4.2 for equations using ΔCO32−). Depth is in fact a good proxy for ΔCO32− below ∼1500 m owing to the pressure effect on calcite dissolution [Ingle, 1975]. We used a statistical approach that allowed development of equations with two variables to predict temperature. All statistical analyses include the entire data set except the Rio Grande Plateau data (see section 3.3). An initial backward elimination regression analysis to predict temperature, including indicator variables for different species, demonstrated that the offset between the species is statistically significant. Thus calibrations were developed separately for each species.
4.1.1. Mg/Ca in G. ruber (white)
 The highest Mg/Ca of the three species in this study is observed in G. ruber, indicating that it inhabits the shallowest habitat and therefore records the warmest temperatures at the surface. This has previously been shown in studies using δ18O to determine depth habitat [Faul et al., 2000]. Dissolution of G. ruber appears to be strongest along the Ontong Java Plateau, where Mg/Ca decreases by ∼14%/ km and completely disappears from sediments below ∼2500 m. Mg/Ca loss in G. ruber shells occurs at 7%/km and 5%/km at the Sierra Leone Rise and Ceara Rise, respectively.
 We used multiple linear regression to develop a calibration equation that includes the dominant temperature effect, as well as a depth parameter to account for the dissolution effect, on shell Mg/Ca. Initial statistics demonstrated that an exponential relationship rather than a linear relationship is a better fit to the data, in agreement with previous calibration studies of both cultured [Nürnberg et al., 1996a; Mashiotta et al., 1999; Lea et al., 1999] and core top samples [Elderfield and Gannssen, 2000]. We noted that the relationship between depth and Mg/Ca appears to be different between the Atlantic and the Pacific, with the effect of dissolution apparent in the Pacific at ∼1.5 km but not apparent until ∼4 km in the Atlantic (Figure 3). We tested if the difference between the basins is statistically significant by running a backward regression analysis using the following model:
where Y is SST [Levitus and Boyer, 1994], X1 is ln(Mg/Ca), X2 is water depth of sediment (km), and X3 is 0 if the core is in the Atlantic or 1 of the core is in the Pacific.
 This model allows both the slope (β3X2X3) and the intercept (β4X3) of the depth Mg/Ca relationship to vary between the two basins. Backward elimination indicated the slope term was not statistically significant, resulting in the following equation:
Therefore, for cores in the equatorial Atlantic, the equation is
In the Pacific the equation is
 This indicates that the relationship between dissolution and water depth (the slope) is not significantly different between the basins but rather that there is an offset in the dissolution between the Atlantic and the Pacific (schematically illustrated in Figure 8). One problem with extrapolating the deep trends to shallow waters is that no offset is expected for surface waters because surface water chemistry is similar between basins. Therefore we have graphically depicted two alternate hypotheses of the relationship between Mg/Ca and depth in shallow waters (Figures 8b and 8c). One hypothesis is that no dissolution occurs shallower than ∼4 km in the equatorial Atlantic, while dissolution continues to the surface in the equatorial Pacific (Figure 8b). An alternate hypothesis is that dissolution occurs at shallow depths in both basins due to the rapid change in carbonate ion concentration at shallow depths (Figure 8c). Our data set does not include enough data points at shallower depths to differentiate between these two hypotheses with any statistical significance. However, plankton tow data indicate that Mg/Ca of G. ruber at 27°C is ∼4.5 mmol/mol (P. von Langen, University of California, Santa Barabara, unpublished data, 2001), which favors the third scenario (Figure 8c). Because our data set does not allow us to differentiate between these two hypothesis, the depth corrections derived in this study should only be applied to sediment deeper than ∼2.8 km in the equatorial Atlantic and ∼1.6 km in the equatorial Pacific.
 The importance of the described statistical approach is that it allows for a different depth correction for the Atlantic and the Pacific, while maintaining the same relationship between SST and Mg/Ca. If equations are derived separately for each basin, they will differ in the preexponential constant and the exponential constant, which implies a difference in the fundamental relationship between temperature and Mg/Ca in G. ruber. Because there is no biological or oceanographic reason such a difference should exist, we chose to derive one equation that maintains the same Mg/Ca temperature relationship between basins. The resulting equation is
for G. ruber, Atlantic and
for G. ruber, Pacific, where R2 = 0.70, Standard error of estimate is 1.2°C, 95% confidence interval for the preexponential constant is ±0.05, and 95% confidence interval for the exponential constant is ±0.015.
 The depth correction for G. ruber is essentially a measure of how depth biases the temperature estimate of sedimentary specimens. The regression indicates that the temperature bias is 0.6 ± 0.3°C/km. The exponential constant, which quantifies the increase of Mg/Ca per °C, agrees with estimates of this relationship from previous calibration studies that range from 0.08 to 0.10, depending on species (Table 3). The preexponential constant is also consistent with previous work. The offset between the Atlantic and the Pacific data is entirely accounted for by the basin offset of 1.6 ± 0.5°C. Again, this offset is only applicable to the depth range of the data included in this study and likely decreases at shallower depths, where carbonate chemistry between the two basins converges near surface waters.
Table 3. Calibration Equations for Various Species of Planktonic Foraminifera
G. sacculifer consistently has lower Mg/Ca than G. ruber by 5–15% (Figure 3, Table 4). This indicates that, assuming similar uptake coefficients, either G. sacculifer adds gametogenic calcite at depth and/or generally lives at deeper depths than G. ruber. To predict the depth at which G. sacculifer calcifies, we used the same approach described in section 4.1.1 using SST as well as water temperatures from different depths from the available Levitus climatology [Levitus and Boyer, 1994]. Backward regression analysis of the data using the model shown in equation (1) resulted in the removal of the slope term for the Mg/Ca depth relationship, therefore again confirming the applicability of equations (3) and (4) for the equatorial Atlantic and Pacific, respectively. Although the equations using temperatures at different depths vary little in their R2 (range from 0.61 to 0.74) or standard errors (all are ±1.4°C), the equation at 20 m most closely represents equations for G. sacculifer previously established by culturing (Table 3) [Nürnberg et al., 1996a, 1996b]. The preexponential constant is within error of 0.39 calculated from Nürnberg et al.'s [1996a, 1996b] culturing work, as is the exponential constant. The G. sacculifer equation for temperatures at 20 m is
for G. sacculifer, Atlantic and
for G. sacculifer, Pacific, where R2 = 0.67, standard error of estimate is 1.4°C, 95% confidence interval for the preexponential constant is ±0.03, and 95% confidence interval for the exponential constant is ±0.013.
 Although the basin offset term is somewhat larger for G. sacculifer (2.0° ± 0.5°C) than for G. ruber (1.6° ± 0.5°C), the terms agree within error, indicating that the dissolution offset between the basins is consistent for the two spinose species. The temperature bias due to increased dissolution with depth is 0.36°C ± 0.24°C/km. This is again within the error of the slope term for G. ruber, which is 0.61°C ± 0.26°C/km. There is some indication, however, that G. sacculifer Mg/Ca is somewhat less sensitive to dissolution, in agreement with previous studies of selective dissolution susceptibility [Bé, 1977].
4.1.3. Mg/Ca in N. dutertrei
N. dutertrei consistently has Mg/Ca ratios 49–55% lower than G. ruber. Previous studies have indicated that although this species is usually associated with the thermocline [Bé, 1977; Fairbanks et al., 1982; Faul et al., 2000], it has a very wide range in depth habitat. Oxygen isotope data from the eastern equatorial Pacific indicate that N. dutertrei lives along the 15°C isotherm [Faul et al., 2000], while other studies have shown that it is more associated with the location of the chlorophyll maximum [Fairbanks et al., 1982]. Both sediment trap data in the San Pedro Basin [Sautter et al., 1991] and plankton tow data from the Sargasso Sea [Fairbanks et al., 1980] and the Panama Basin [Fairbanks et al., 1982] indicate that N. dutertrei has a large depth habitat range, which fluctuates depending on hydrographic conditions.
 We used the basic model described in equation (1) to develop equations at water column depths ranging from 0 to 100 m. Statistical analysis indicates that the slope of the water depth — Mg/Ca relationship does not differ between the basins, confirming the trend noted for both G. ruber and G. sacculifer. The best fits, as judged by how closely the preexponential constant and exponential constants resemble previous calibrations, are to temperatures at 30 and 50 m. The N. dutertrei equation for temperatures at 50 m is
for N. dutertrei, Atlantic and
for N. dutertrei, Pacific, where R2 = 0.59, Standard error of estimate is 1.6°C, 95% confidence interval for the preexponential constant is ±0.10, and 95% confidence interval for the exponential constant is ±0.010.
 The preexponential constant for the equation at 30 m is much lower than would be expected (0.16), while that at 50 m is somewhat higher than would be expected. The exponential constant for the equation derived from temperatures at 30 m is higher than is expected (0.12), while that for the 50 m equation is within the range of previous values. It seems reasonable to infer that the average calcification for N. dutertrei is somewhere between 30 and 50 m, with 50 m being the closest fit. Note that the dissolution corrections are more than twice as large for N. dutertrei than the two spinose species. In general, N. dutertrei is thought to be quite resistant to dissolution [Bé, 1977], so our results might reflect the preferential loss of the dissolution prone, Mg-rich part of the shell formed in shallow waters, as previously suggested by Brown and Elderfield  for Globorotalia tumida.
 At this time we are unable to resolve a more accurate calibration for N. dutertrei. Because this species varies its depth habitat as a function of local hydrographic conditions, a comprehensive calibration would need to take into account the depth of the thermocline, the depth of the chlorophyll maximum, and the temperatures at those depths, data that are not available for all of our sites. An alternative approach is to develop calibration equations from culturing.
4.2. Mg/Ca Calibrations Using ΔCO32− as a Dissolution Correction
 Dissolution of calcite occurs when bottom waters are undersaturated with respect to calcite [Berger et al., 1982]. We calculated ΔCO32− as a measure of dissolution at all sites and developed alternative calibration equations using ΔCO32− in place of water depth. Input data (temperature, salinity, pressure, concentration of silicate and phosphate, alkalinity, and total carbon at the depth of the core) were taken from the WOCE data set [Schlitzer, 2000], and ΔCO32− was calculated using the Program Developed for CO2 System Calculations by E. Lewis and D. W. R. Wallace.
 We used the calculated ΔCO32− to perform a multilinear regression using both Mg/Ca and ΔCO32− to predict SST. The model equation takes the following form:
where Y is SST, X1 = ln(Mg/Ca), and X2 = ΔCO32−.
 This equation differs from the model used to develop equations using core depth (equation (1)) in that there is no variable that differentiates between the Atlantic and the Pacific. We found that there is no statistical difference between the Atlantic and the Pacific when using ΔCO32−, indicating that ΔCO32− effectively accounts for the difference in calcite saturation between the two basins.
 The uncertainties with this approach are two fold. First, although ΔCO32− accounts for the corrosivity of the bottom waters, it does not account for changes in ΔCO32− within the sediments due to the input of metabolic CO2 to sediments [Hales and Emerson., 1996, 1997]. This may in part explain the offset noted earlier between the Ceara Rise and the Sierra Leone Rise (section 3.3). The second uncertainty is due to the availability of the data required to calculate ΔCO32−. Data from WOCE are the most extensive available, but gaps remain, and on average, the WOCE site used to calculate ΔCO32− was only within ∼1° latitude and ∼5° longitude of the core sites (Table 2).
4.2.1. Mg/Ca for G. ruber using ΔCO32−
 We used SST and ΔCO32− to develop the following equation:
where R2 = 0.71, standard error of estimate is 1.2°C, 95% Confidence Interval for the preexponential constant is ±0.002, and 95% Confidence Interval for the exponential constant is ±0.014.
 Taking into consideration the stated uncertainties for estimating ΔCO32−, this equation is remarkably similar to equations (5) and (6), which were derived using SST and core depth. The preexponential constant in equation (14a) is 0.33 when ΔCO32− is zero (i.e., saturated bottom water). Incorporating the depth where ΔCO32− is zero in the western tropical Atlantic (4.5 km, Figure 6) into equation (5) yields a preexponential constant of 0.30, in agreement with the value from equation (14a) within the confidence intervals.
 It is possible to compare equations (5) and (14a) by setting the depth correction portion of equation (5) equal to the ΔCO32− correction in equation (14a). Setting the two terms equal and taking the natural log of both sides
and then solving for depth yields
Equation (14c) is the relationship between depth and ΔCO32− as derived independently based on their influence on Mg/Ca. A general rule-of-thumb is that for each 1 km increase in ocean depth, calcite saturation increases by 15μmol/kg [Broecker and Peng, 1982]. Plugging this value into equation (14c) yields a depth of 1 km, confirming the interchangeability of equations (5) and (14a).
4.2.2. Mg/Ca for G. sacculifer using ΔCO3−2
 Equations for G. sacculifer were developed using temperatures from different depths. The equation which most closely agrees with previously established calibrations for G. sacculifer arises from using temperatures at 20 m water depth, confirming that shell Mg/Ca is recording temperatures slightly below the surface. The equation for G. sacculifer at 20 m is
R2 = 0.65, standard error of estimate is 1.4°C, 95% confidence interval for the preexponential constant is ±0.006, and 95% confidence interval for the exponential constant is ±0.014.
 The preexponential constant is again somewhat lower than in equations (7) and (8), but if an estimate of ΔCO32− at 4 km in the Atlantic is included in the preexponential constant, it increases to 0.34, compared to 0.33 computed from equation (7) at 4 km. The exponential constant in the equation is 0.084 ± 0.014, which is within error of 0.09 in equations (7) and (8) and previous calibrations for this species (Table 3) [Nürnberg et al., 1996a, 1996b]. The preexponential constant in equation (15) is 0.31 when ΔCO32− is zero (i.e., saturated bottom water). Incorporating the depth where ΔCO32− is zero in the western tropical Atlantic (4.5 km, Figure 6) into equation (7) yields a preexponential constant of 0.32, in agreement with the value from equation (15) within the confidence intervals.
 The equation for N. dutertrei using temperatures at 50 m is
where R2 = 0.77, standard error of estimate is 2.0°C, 95% confidence interval for the preexponential constant is ±0.04, and 95% confidence interval for the exponential constant is ±0.008. The equation for N. dutertrei using temperatures at 75 m is
where R2 = 0.78, standard error of estimate is 2.4°C, 95% confidence interval for the preexponential constant is ±0.01, and 95% confidence interval for the exponential constant is ±0.014.
 Using ΔCO32− to account for dissolution suggests a depth habitat for N. dutertrei between 50 and 75 m; using core depth as a dissolution parameter suggests a depth habitat between 30 and 50 m. Taken together, the two approaches suggest 50 m as the most likely depth for N. dutertrei, in good agreement with previous studies based on δ18O [Fairbanks et al., 1982; Faul et al., 2000].
4.3. Comparison of Mg/Ca and δ18O Calibrations
 Oxygen isotopes have a long history as a temperature proxy, but direct interpretation is complicated because δ18Ocalcite records both temperature and the isotopic composition of the water [Emiliani, 1955; Shackleton and Opdyke, 1973]. In this section we first compare the offset between the different species in the δ18O data to the offset noted in the Mg/Ca data. We next use δ18O to calculate the temperature at which the foraminifera calcified using a calibration equation developed from culturing [Bemis et al., 1998]. We then discuss using these calcification temperatures to develop calibration equations for Mg/Ca. Finally, we compare the predictive power of the derived Mg/Ca equations in section 4.1 to the predictive power of the δ18O — temperature relationship for G. ruber and G. sacculifer.
4.3.1. The δ18O and Mg/Ca offsets between G. ruber and G. sacculifer
 The offset in δ18O between G. ruber and G. sacculifer is not as large as is expected from the Mg/Ca data, particularly along the Ontong Java Plateau (Figure 5), where the largest offset in shell Mg/Ca is observed (Figure 3). This observation is unexpected given previous observations of a significant δ18O offset between these two species in plankton tow samples [Duplessy et al., 1981] and in sediments [Stott et al., 1996; Mulitza et al., 1998; Arz et al., 1999; Faul et al., 2000]. Downcore data for an Ontong Java Plateau site also reveal a consistent offset between G. ruber and G. sacculifer δ18O [Patrick et al., 1997]. In addition, downcore data from Ocean Drilling Program (ODP) Hole 806B (OJP) and TR163-19 (eastern equatorial Pacific) indicate that there is a consistent offset between G. ruber and G. sacculifer in both δ18O and Mg/Ca [Lea et al., 2000] (Figures 9a and 10a). Although the offset between the species in our data set is larger for shell Mg/Ca, clearly the offset also exists for shell δ18O. To explain both the large offset in shell Mg/Ca and a small offset in the δ18O between the two species, G. sacculifer would need to live in colder temperature water with lower salinity and δ18O values. Although there are seasonal differences in the salinity profile in the waters overlying the Ontong Java Plateau, there is currently no evidence that either species would prefer a lower or higher salinity environment without an associated temperature difference (changes in seasonal SST at this location are <1°C).
 At this time we are unable to explain the discrepancy between the two proxies. Culturing studies are making important advances to quantify species-specific offsets in the disequilibrium precipitation of calcite, which points to a δ18O correction for G. sacculifer of +0.35‰ relative to G. ruber (at 25°C) (H.J. Spero et al., unpublished data, 2001). Given that other studies have demonstrated consistently more negative δ18O for G. ruber, the implication of culturing studies on the species-specific offset in δ18O, and given the consistency of the same offset in the down core record from ODP Hole 806B and TR163-19, we suggest that the minimal OJP core top δ18O offset is not representative and does not conflict with the inferences based on Mg/Ca offsets.
4.3.2. Comparing temperatures calculated using δ18O and Mg/Ca equations
 Measurements of both Mg/Ca and δ18O in three species of tropical foraminifera allows us to directly compare the utility of both of these proxies to predict temperatures. Additionally, we have used these two measurements to apply the Elderfield and Ganssen  approach, which uses calcification temperatures calculated from δ18O to calibrate Mg/Ca. This approach was developed as a method to account for differences in depth habitat between different species. However, applying this method to our data set resulted in what we infer to be unrealistic calibrations for Mg/Ca, as judged by comparing the preexponential and exponential constants to previous calibrations. It is important to note that cores used by Elderfield and Ganssen  were from the North Atlantic and uniformly displayed good preservation. This differs from our data set, which includes cores that have undergone varying degrees of dissolution, including deep sites and Pacific sites, most of which exhibit poorer preservation than sites in the North Atlantic.
 As a way to evaluate the utility of Mg/Ca and δ18O as temperature proxies, we compared both the predicted temperature using Mg/Ca and the predicted temperature using δ18O to SST. This comparison should be done using an independent data set, but at this time no other extensive core top data set exists in the tropics. Therefore results derived from this comparison cannot be used to validate our relationship. Predicted temperatures for G. ruber using Mg/Ca (equations (5) and (6)) are in good agreement with mean annual SST (standard error = ±1.2°C) (Figure 11a). The range in predicted temperatures at 27°C results from the offset between the Ceara Rise and Sierra Leone Rise previously discussed (section 3.3). All cores along the Ontong Java Plateau have predicted temperatures very close to 29°C, confirming the efficacy of the depth correction in equations (5) and (6).
 To predict temperatures using δ18O we used the Orbulina universa low light calibration equation, which was developed from culturing work and appears to provide good temperature estimates for the white variety of G. ruber based on plankton tow and sediment samples [Bemis et al., 1998]. Over the tropical temperature range, it is quite similar to the Kim and O'Neil  inorganic calcite calibration. A comparison of these temperatures predictions to mean annual SST for G. ruber indicates a significantly larger range of predicted temperatures for any given SST (standard error = ±2.2°C) (Figure 11c). The range of predicted temperatures at a mean annual SST of 29°C, based on the Ontong Java Plateau depth transect, demonstrates the temperature bias of dissolution on δ18O. The predicted temperature range at 27°C represents the Ceara Rise and the Sierra Leone Rise transects. Note that predicted temperatures for these sites using δ18O results in temperatures that are much higher than the mean annual SST. Because the O. universa low light equation was developed from culturing [Bemis et al., 1998] it does not include a correction for dissolution, resulting in a range of temperatures predicted along a depth transect, as well as an offset between the Atlantic and the Pacific presumably due to differences in preservation between the basins. The standard error derived using other established δ18O-temperature relationships results varies little from that derived using the O. universa low light equation but still demonstrates the variation of temperature estimates along depth transects due to a dissolution effect on δ18O. The predicted temperature using the derived Mg/Ca equations does not show this offset because a basin offset parameter is part of the calibrations.
 We also compared the predicted temperature using Mg/Ca and the predicted temperature using δ18O to temperatures at different water depths for G. sacculifer. A comparison of the predicted temperature using Mg/Ca (equations (7) and (8)) demonstrates good agreement between the predicted temperature and temperatures at 20 m in the water column, as expected (standard error = ±1.4°C) (Figure 11b). A comparison of predicted temperatures using δ18O to temperatures at different water depths showed the standard error was smallest when comparing predicted temperatures to temperatures at 30 m (±2.4°C) and 20 m (±2.6°C) (Figure 11d shows comparison to temperature at 20 m). The increased errors at lower temperatures may be due to increased hydrographic complexity and shallow mixed layer in the eastern equatorial Pacific, which represents most of the data points in the lower temperature range. The smaller overall standard error at 20 and 30 m, however, is in close agreement with the Mg/Ca data we used to infer a depth habitat of 20–30 m for G. sacculifer. The comparisons for both G. ruber and G. sacculifer demonstrate that the δ18O data requires a correction for dissolution.
 To better understand the habitat depth of N. dutertrei, we compared the standard errors of the δ18O temperature prediction to temperatures at different depths in the water column. We found that the lowest standard error occurs when comparing predicted temperatures to temperatures at 50 m (±2.1°C). This is in agreement with the results from Mg/Ca, which indicate a best-fit calibration equation for temperatures at 50 m.
5. Paleoceanographic Implications
 Our results indicate that there is close agreement between equations using ΔCO32− as a dissolution correction and those using core depth. This demonstrates that core depth does a good job at estimating dissolution at the depth ranges included in our data set and that the basin offset can account for dissolution differences between the two basins. Although dissolution of calcite is controlled by the carbonate chemistry of the water, depth is a good proxy for dissolution depths greater than ∼1500 m due to the pressure effect on dissolution.
 There are two main reasons why using the calibrations with a depth correction may be more appropriate to use for paleoceanographic work than equations with ΔCO32− as the dissolution parameter. First, there is essentially no uncertainty in the measure of depth, whereas there are significant uncertainties in the estimates of ΔCO32−. These uncertainties are introduced because the data required to estimate ΔCO32− are only sparsely distributed. In addition, ΔCO32− of deep water does not account for increased dissolution within sediments due to decreased ΔCO32− resulting from the production of metabolic CO2. Second, and perhaps most important, estimates of changes in lysocline depth through time have been published [Farrell and Prell, 1989, 1991], while there are no such estimates of ΔCO32−. Thus the equations using depth as a dissolution parameter are currently more applicable to paleorecords, although future estimates of ΔCO32− through time can be used with our ΔCO32− calibration equations.
 To demonstrate the utility of our calibrations, we have applied our equations for G. ruber and G. sacculifer using depth as a dissolution correction to previously published downcore records in the western and eastern equatorial Pacific (Figures 9 and 10, respectively) [Lea et al., 2000]. We used published changes in the lysocline depth in the Pacific to adjust our depth correction [Farrell and Prell, 1989, 1991]. Also shown are the downcore estimates of temperature using Mg/Ca of G. ruber and the Lea et al.  equation (Mg/Ca = 0.30e0.09(SST)). Note that in both cores the estimates of modern SST are more closely approximated using the equation presented in this paper for G. ruber compared to the Lea et al.  equation (0.2°C versus 0.5°C error for both the western equatorial Pacific (WEP) and eastern equatorial Pacific (EEP)), indicating the improved accuracy of a large-scale calibration that includes a dissolution correcting.
 The temperature difference between modern SST and the SST in the last glacial maximum (LGM) based on G. ruber is 2.9°C in the WEP and 2.8°C in the EEP (Figures 9b and 10b). These SST differences between the LGM and the Holocene are somewhat larger in magnitude than those reported by Lea et al.  (2.8°C in WEP and 2.6°C in the EEP), due to the incorporation of changes in lysocline depth in our calibrations. Lea et al.  estimated that downcore preservation difference could bias temperatures estimates by ±0.5°C, based on preservation along the Ontong Java Plateau. Calibrations developed in this paper, however, indicate that this is an overestimate of the potential dissolution bias. If dissolution correction is not adjusted to account for changes in lysocline depth, the predicted SST using G. ruber Mg/Ca is on average 0.3°C greater than estimates which do include changes in lysocline depth in the EEP (0.1°C greater in the WEP). This small difference is within the error of the calibration and therefore cannot be statistically distinguished. This implies that preservation difference on glacial-interglacial timescales introduce a very small bias in Mg/Ca temperature estimates.
 Temperature estimates from G. sacculifer Mg/Ca are lower than those for G. ruber because G. sacculifer is recording subsurface temperatures (section 4.1.2 and 4.2.2). The temperature difference between the modern temperature and that during the LGM is ∼2°C for both the WEP and EEP (note that data for G. sacculifer is of lower resolution than that for G. ruber). The largest difference between recorded temperatures by G. ruber and G. sacculifer is during the Holocene, while the smallest difference is recorded during the LGM. This could potentially reflect changes in water column structure.
 We analyzed shell Mg/Ca and δ18O in three species of tropical foraminifera for a series of core tops which includes four depth transects: the Ontong Java Plateau in the equatorial Pacific, the Ceara Rise and the Sierra Leone Rise in the equatorial Atlantic, and the Rio Grande Rise in the subtropical Atlantic (data from the Rio Grande Rise were not used in statistical analysis due to complications with these core tops, see section 3.3). These data indicate that temperature is the dominant effect on shell Mg/Ca but that dissolution with increasing water depth and changing calcite saturation are also a major influence. We developed calibration equations for each species that account for the influence of both temperature and dissolution.
 Because there are clear offsets between Mg/Ca for different species, we develop individual calibration equations for each species. The equations were developed using water temperatures ranging from 0 to 100 m [Levitus and Boyer, 1994] so that we could estimate the average depth habitat of each of the species. We found that G. ruber is the most accurate recorder of surface temperature. G. sacculifer is inferred to live at ∼20 m. The depth habitat of N. dutertrei is not as clear, owing to the wide range of depths inhabited by this species depending on hydrographic conditions, but both the Mg/Ca and δ18O calibrations suggest an average depth for this species of ∼50 m. Further constraint of the depth habitat for N. dutertrei will require taking into account other variables, such as the depth and temperature of the thermocline and chlorophyll maximum.
 Three approaches were used to develop calibration equations for each species: (1) water depth was used as a measure of dissolution, with an additional factor to account for the offset in bottom water carbonate ion concentration between the Atlantic and Pacific; (2) ΔCO32− was used as a measure of dissolution, which eliminates the need for a basin offset; and (3) calcification temperatures were calculated from δ18O and Mg/Ca was calibrated against these temperatures rather than against water column temperatures. The use of calcification temperatures based on paired δ18O data to calibrate the Mg/Ca data appears to yield the least reliable calibrations, presumably because our data set includes many deep cores, as well as cores from both the Atlantic and the Pacific, with different levels of preservation. Although dissolution biases temperature estimates for both Mg/Ca and δ18O to lower values [Rosenthal et al., 2000], the magnitude of the effect is not always the same for both proxies, thereby introducing artifacts into the Mg calibration.
N. dutertrei appears to be the most susceptible to dissolution, with shell Mg/Ca decreasing by up to 23%/km. G. sacculifer is the most resistant to dissolution (5–9% decrease in shell Mg/Ca per km), while G. ruber shell Mg/Ca decreases by 5–14%/km. Although G. ruber is more susceptible to dissolution than G. sacculifer, we believe that this species is the most promising recorder of SST. The equations we developed for G. ruber, which include a correction for dissolution, have a standard error of only ±1.2°C.
 Downcore applications of our calibrations for G. ruber and G. sacculifer demonstrate the utility of these new equations where paleolysocline estimates exist. It is clear that our calibrations provide a more accurate estimate of modern SST owing to the large data set and the inclusion of a dissolution correction. Downcore application indicates that the error introduced due to changes in preservation is small on glacial-interglacial timescales. Our equations provide the dual benefit of a more global calibration yielding more accurate SST estimates as well as the ability to estimate errors due to changes in preservation history.
 Taken as a whole, our results suggest that Mg/Ca in G. ruber and G. sacculifer can be used effectively for paleo-SST analysis. Calibrated over a broad scale of water depths, basins and preservation, the standard error of temperature measurements is between ±1.2 and 1.4°C.
 We thank Dan McCorkle for providing samples from the Ontong Java Plateau and Jim Kennett for samples from TR163 cores. We also thank J. Kennett for review of this manuscript. G. Paradis was responsible for the analysis of samples, and H. Berg provided technical support in the PQ laboratory at UCSB. J. Michaelson provided input regarding the statistical analysis. A. Dave and A. Schilla assisted with cleaning of the samples. This material is based upon work supported by the National Science Foundation under Grant OCE-9819254 (D.W.L.) and OCE-9903632 (H.J.S).