Temperatures derived from Mg/Ca ratios of the calcite tests of planktonic foraminifera are distorted when samples are partially dissolved, and methods are required to quantify this source of inaccuracy. Here we compare a dissolution index (XDX), based on X-ray computed tomography scans, to Mg/Ca of four species of foraminifera (G. ruber (white), G. sacculifer (without sac), N. dutertrei, and P. obliquiloculata) from core top sediments from the tropical Pacific, Atlantic, and western Indian Ocean. Deepwater calcite saturation values (Δ[CO32−]) at the sites ranged from 55 to −23 μmol/kg. An estimate of ΔMg/Ca (reduction in Mg/Ca due to dissolution) was made for each sample. ΔMg/Ca decreased linearly from deepwater Δ[CO32−] values of between 10 (±4) and 15 (±5) μmol/kg. These values are minimum estimates of the threshold below which Mg/Ca is affected by dissolution, as they are limited by assumptions made in calculating ΔMg/Ca. Sensitivity of Mg/Ca to Δ[CO32−] was greatest for G. ruber, where Mg/Ca decreased by 0.102 (±0.036) mmol/mol per μmol/kg. Sensitivity was similar for G. sacculifer (0.047 ± 0.015 mmol/mol per μmol/kg), N. dutertrei (0.037 ± 0.010 mmol/mol per μmol/kg), and P. obliquiloculata (0.040 ± 0.008 mmol/mol per μmol/kg). Sensitivity was similar at all sites for each species, excepting an apparently greater response for N. dutertrei from the Caribbean compared to other sites. Calibrations between XDX and ΔMg/Ca provide a means to estimate dissolution bias on Mg/Ca. Poor correlation between XDX and δ18O suggests that, for the small sample size typical for analysis, variability in initial δ18O overwhelms dissolution effects.
 Mg/Ca content of foraminiferal tests is particularly sensitive to dissolution. This is a major drawback in the application of a proxy which has greatly increased the information available to paleoceanographers. The Mg/Ca paleotemperature proxy was developed during the 1990s [Nürnberg, 1995; Nürnberg et al., 1996; Rosenthal et al., 1997; Hastings et al., 1998; Mashiotta et al., 1999; Lea et al., 1999; Elderfield and Ganssen, 2000] and has come into increasingly routine use in the last few years. It is based on the recognition that Mg incorporation into foraminifera tests is temperature-dependent. Mg/Ca of test calcite therefore offers a means to isolate the temperature component of carbonate δ18O for foraminiferal samples [e.g., Mashiotta et al., 1999]. However, Mg incorporation to the test affects its solubility [Brown and Elderfield, 1996] with the consequence that partial solution of the test preferentially removes Mg-rich calcite, biasing temperatures toward lower, colder values. An assessment of test preservation is therefore vital to estimate the accuracy of temperatures derived using this proxy.
 There are several methods that aim to quantify the effect of dissolution on Mg/Ca of planktonic foraminifera. Evaluation of corrosivity of deepwater at a particular location, based on present-day core depth or calcite saturation [Dekens et al., 2002], offers a useful correction over the past few thousand years. However, calcite chemistry of deepwater varies through a glacial cycle. Thus, fluctuations in calcite preservation may confound temperature comparisons between interglacial and glacial periods. Change in preservation has been suggested to account for some of the disparity between sea surface temperature (SST) records based on different proxies during the last deglaciation [Mix, 2006]. On timescales greater than several thousand years, therefore, a direct method of assessing the effect of calcite dissolution on Mg/Ca is necessary.
 Examination of foraminifera tests during the early stages of dissolution reveals that material is first lost from the inside of the test, while the outer layer remains intact [Brown and Elderfield, 1996]. This means that test mass is reduced while test size remains relatively constant [Lohmann, 1995]. Test mass has been directly calibrated to reduction in Mg/Ca [Rosenthal and Lohmann, 2002]. This approach requires that initial test mass be temporally and spatially constant. In fact, there is evidence that environmental conditions, such as nutrient availability [de Villiers, 2004] or carbonate ion concentration [Barker and Elderfield, 2002], influence the thickness of the test walls and therefore the mass of the test.
 A method of assessing dissolution in the tests of planktonic foraminifera using X-ray computed tomography (CT) has been developed [Johnstone et al., 2010]. Details of inner test structure mapped by this nondestructive imaging technique give an insight into test preservation (Figure 1). Dissolution index XDX is based on dissolution features identifiable in CT scans and appears to be insensitive to initial test mass.
 In this present study the paleoceanographic proxies Mg/Ca and δ18O are first compared to deepwater calcite saturation (Δ[CO32−]) to assess sensitivity to dissolution. The two proxies are then compared directly to XDX to ascertain the potential of XDX to correct for dissolution bias.
2. Samples and Methods
 This study expands on the work of Johnstone et al. . In that work, foraminifera tests (300–355 μm size fraction) from core top sediments (Table 1) were scanned using CT and allocated values on dissolution index, XDX. In the current study, these previously scanned samples were analyzed for Mg/Ca and δ18O.
Table 1. Details of the Core Tops Used in This Study
 The core sites used here span the calcite saturation horizon (CSH), the depth where Δ[CO32−] is equal to zero. Δ[CO32−] of deepwater ranges from 55 to −23 μmol/kg. The main sample set is from the Ontong Java Plateau (OJP) in the Pacific Ocean. In order to discover if the sensitivity of Mg/Ca to Δ[CO32−] is similar for different ocean basins, additional core top samples from the Ceara Rise, the Caribbean, the Mid-Atlantic Ridge (MAR) and the western Indian Ocean were used (Table 1).
 The four species of foraminifera analyzed calcify at different depths in the water column [Hemleben et al., 1989], and so have different initial Mg/Ca and δ18O values. The mixed-layer dwelling species, Globigerinoides ruber (white) and Globigerinoides sacculifer (without a saclike final chamber), have high Mg/Ca, and low δ18O, compared to the thermocline dwellers, Neogloboquadrina dutertrei and Pulleniatina obliquiloculata.
2.2. CT Scanning and Dissolution Index XDX
 Samples were scanned using a Skyscan 1072 desktop X-ray micro-CT scanner [Sasov and Van Dyck, 1998; Van Dyck and Sasov, 1998] at the Department of Earth Sciences, University of Cambridge. The scanner uses an air-cooled point X-ray source to create a series of radiographs of a sample as it rotates. Tests were glued to the sample holder using water-soluble glue and scanned in batches of 8 to 10 tests. Anode voltage was set at 80 kV and the rotation step was 0.9°. A 0.5 mm Al filter was used to cut out the softest X-rays and so reduce beam hardening effects. Exposure time was 4.5 s. Cross-sectional slices (“tomographs”) were reconstructed using Skyscan’s own software which uses the Feldkamp cone-beam algorithm [Feldkamp et al., 1984]. A 10% beam hardening correction was applied to reduce edge effects.
 Reconstructed slices, with a resolution of 7 μm, reveal the preservation state of the inner calcite of foraminiferal tests. Based on a visual inspection of the CT scans, each test was assigned to one of the five categories that make up dissolution index XDX. The categories can be summarized as: 0, no dissolution; 1, dissolution detectable in the inner chambers; 2, severe dissolution of the inner chambers; 3, all inner calcite (the inner chambers and the inside of the outer wall) severely altered; and 4, extreme dissolution, only outer crust remains. (G. ruber, which lacks an outer crust, does not show this stage.) XDX values were averaged for each sample. Values, from Johnstone et al. , are given in the auxiliary material.
2.3. Mg/Ca Analysis
 After CT scanning, samples were recombined to provide samples of ∼30 tests for Mg/Ca analysis. Tests were gently crushed between two glass plates in order to open the chambers. Cleaning followed the protocol of Barker et al.  (“Mg cleaning”) outlined as follows. Samples were rinsed five times with distilled deionized water and twice with methanol to remove clays. Between rinses, samples were treated with one minute of ultrasonication. To remove organic matter samples were heated with 250 μL of an oxidizing solution (1% H2O2 buffered with NaOH) for 10 min. Every 2.5 min samples were removed from the water bath and tapped on the bench to dislodge gas bubbles. After 5 min they received a few seconds of ultrasonication in addition. This step was repeated after replacement of the oxidizing solution. Any remaining solution was rinsed off with distilled deionized water and samples were transferred to clean vials. A weak acid rinse (0.001 M HNO3) and two subsequent water rinses ended the cleaning process. Samples were dissolved for analysis in 0.075 M HNO3.
 Core tops from the OJP and western Indian Ocean were analyzed by Inductively Coupled Plasma-Mass Spectrometry [Yu et al., 2005] at the Department of Earth Sciences, University of Cambridge. Long-term reproducibility for Mg/Ca is better than 1.5%. Additional samples (Ceara Rise, Caribbean and Mid-Atlantic Ridge core tops) were analyzed by Inductively Coupled Plasma-Optical Emission Spectroscopy at University of Bremen. Element lines for Mg/Ca were Ca: 315.89 nm; Mg: 279.55 nm. Reproducibility was within 2%. Duplicate G. sacculifer and N. dutertrei samples from the OJP showed no systematic offset in Mg/Ca between the two analytical methods (see auxiliary material). Most importantly for this study, regressions between Δ[CO32−] and Mg/Ca were practically identical for the two analyses. All temperatures derived from Mg/Ca values were calculated using the species specific equations of Anand et al.  as follows:
where Mg/Ca ratios are in mmol/mol and T is in °C.
2.4. Oxygen Isotopes
 Samples for δ18O analysis were rinsed to remove the glue with which they were fixed for CT scanning. This glue gave no signal on the Mass Spectrometer; δ18O was analyzed at University of Bremen on a Finnigan MAT 251 with automated carbonate preparation device Carbo Kiel II. Reproducibility of the carbonate standard (a Solnhofen Limestone) was 0.07 ‰.
 Sensitivity of Mg/Ca to Δ[CO32−] for G. ruber, like G. sacculifer, was similar at the OJP and Ceara Rise sites. For N. dutertrei and P. obliquiloculata sensitivity was similar for OJP, Ceara Rise and Indian Ocean sample sets (slopes of regressions are given in Figure 2). Mg/Ca of G. ruber from the Ceara Rise was more sensitive to dissolution than G. sacculifer from the same site (no overlap on 95% confidence interval of slopes). Regenberg et al.  carried out a similar sensitivity study using Caribbean core tops. The response of Mg/Ca to Δ[CO32−] of G. ruber and G. sacculifer in that study are comparable to our results for these species (regressions shown in Figure 2), although the apparently higher sensitivity of G. ruber compared to G. sacculifer did not reach statistical significance. Mg/Ca of N. dutertrei from the Caribbean [Regenberg et al., 2006] appears to be more sensitive to Δ[CO32−] than that of N. dutertrei from the other core top samples used here suggesting that sensitivity can vary within a species.
 The disparity in Mg/Ca values of mixed layer dwelling (G. ruber and G. sacculifer) and thermocline dwelling (N. dutertrei and P. obliquiloculata) species indicates that, despite dissolution, Mg/Ca values retain part of the initial temperature signal. The wide scatter of G. ruber and G. sacculifer values reflects the range of SST at different sites. Annual average SST at the OJP is ∼29°C, while SSTs of Indian Ocean sites are between 17 and 21°C (Table 1).
3.1. Sensitivity of Dissolution-Induced Reduction in Mg/Ca (ΔMg/Ca) to Δ[CO32−]
 In order that the effect of dissolution on Mg/Ca can be compared for samples with different calcification temperatures, and hence different initial Mg/Ca, we estimate the reduction in Mg/Ca due to dissolution, ΔMg/Ca,
where Mg/CaINITIAL is Mg/Ca before any dissolution effect and Mg/CaMEASURED is analyzed Mg/Ca. The first stage of calculating ΔMg/Ca is to estimate values of Mg/CaINITIAL.
 There are two possible ways to estimate Mg/CaINITIAL. The first, (1), is to use Mg/Ca values of well-preserved samples where Mg/Ca is unaffected by dissolution. The threshold values of Δ[CO32−] above which dissolution effects are likely to be insignificant are discussed in section 5.3. This method is appropriate where there are samples within a set above these dissolution thresholds and all samples represent similar environmental conditions. Sample sets from the OJP, Ceara Rise, Mid-Atlantic Ridge and the Caribbean Sea each cover a small geographic area. There is less than 1°C difference in annual average temperature, and similar water column temperature profiles, for sites within each sample set [Locarnini et al., 2006]. For these sample sets, it is reasonable to assume that Mg/CaINITIAL values were comparable to each other for all samples within the set.
 The other method, (2), is to estimate calcification temperature based on observations of habitat depth and season for each species, and then back calculate Mg/Ca using established calibrations. This approach must be used for sample sets which contain no site representing dissolution unaffected Mg/Ca. This is the case for our G. ruber and G. sacculifer samples from the western Indian Ocean. Sample sites cover a wide geographic area, spanning a latitudinal range from 10 to 33°S, and SST (hence Mg/CaINITIAL) is not the same at all sample sites.
 The correlation between Δ[CO32−] and ΔMg/Ca is defined as
where m is the slope of the regression and Δ[CO32−]CRITICAL the intercept on the y axis below which Mg/Ca progressively declines. Sensitivity of Mg/Ca to dissolution is 1/m (mmol/mol reduction in Mg/Ca per 1 μmol/kg decrease in Δ[CO32−]).
3.1.1. Estimate of Mg/CaINITIAL (1) From Well-Preserved Samples
 Samples used to define Mg/CaINITIAL were as follows. For the Ceara Rise core top sample GeoB4415, where Δ[CO32−] was 22 μmol/kg, was used. Samples from sites where Δ[CO32−] was above 25 μmol/kg were averaged to estimate Mg/CaINITIAL for each species for the Mid-Atlantic Ridge (GeoB4420 and GeoB4421) and Caribbean Sea (M35014 and M35010) sample sets.
 Sites with the highest Δ[CO32−] values in the western Indian Ocean and OJP sample sets have Δ[CO32−] values of 18 μmol/kg (WIND20B) and 14 μmol/kg (1BC3), respectively. These Δ[CO32−] values are below the threshold where Regenberg et al.  identify dissolution effects on Mg/Ca (at 22–25 μmol/kg). This raises the question of whether these samples are free from dissolution bias (further discussion in section 5.3). However, an offset in ΔMg/Ca and Δ[CO32−] from pristine values does not alter the sensitivity calculated between the two parameters if the relationship is linear, as is assumed.
 Linear regressions between ΔMg/Ca and Δ[CO32−] (equation (6)) are similar for the Ontong Java and Ceara Rise sites for G. ruber and G. sacculifer. Regressions are similar for N. dutertrei and P. obliquiloculata from Ontong Java Plateau, Ceara Rise and western Indian Ocean sites (Table 2).
Table 2. Coefficients for Regressions Between Δ[CO32−] and ΔMg/Caa
Ontong Java Plateau
Sensitivity of Mg/Ca to Δ[CO32−] for All Samples Combinedb (mmol per μmol/kg)
Sensitivity of Mg/Ca to Δ[CO32−] (mmol per μmol/kg)
Sensitivity of Mg/Ca to Δ[CO32−] (mmol per μmol/kg)
Sensitivity of Mg/Ca to Δ[CO32−] (mmol per μmol/kg)
Δ[CO32−]CRITICAL is the intercept on the y axis, the value of Δ[CO32−] where ΔMg/Ca is zero. Numbers in parentheses are 95% confidence intervals.
Sensitivity (inverse slope) of ΔMg/Ca to Δ[CO32−] for the regressions shown in Figure 4.
Mg/Ca of Shallow Samples Used as Mg/CaINITIAL
G. ruber (white)
SST Used to Derive Mg/CaINITIAL
G. ruber (white)
 Correlation between ΔMg/Ca and Δ[CO32−] (r2 > 0.70) for N. dutertrei and P. obliquiloculata from the western Indian Ocean suggests that Mg/CaINITIAL, hence calcification temperature, is similar at all sites in this sample set despite the difference in water column structure between sites. Thermocline structure east of Madagascar varies with latitude, being deeper at 20°S, where the 18°C isotherm is at 250 m depth, than at 10 or 30°S where the 18°C isotherm is at 100 m [Locarnini et al., 2006]. N. dutertrei and P. obliquiloculata inhabit the thermocline and calcify over a range of depths and hence temperatures [Ravelo and Fairbanks, 1992]. Estimates of their depth habitat vary from place to place [Shackleton and Vincent, 1978; Anand et al., 2003] and probably reflect local thermocline structure and chlorophyll maximum rather than a fixed habitat depth. It may be that calcification is to some extent temperature controlled in these species. For instance, Hemleben et al.  suggested N. dutertrei crust growth is triggered below 15°C.
3.1.2. Estimate of Mg/CaINITIAL (2) From SST for G. ruber and G. sacculifer
G. ruber and G. sacculifer calcify predominantly in the mixed layer and the photic zone [Bé, 1980; Hemleben et al., 1989; Savin and Douglas, 1973] and Mg/Ca is thought to represent temperatures close to SST. For these species therefore, Mg/CaINITIAL could be calculated from SST.
 According to the calibration of Anand et al.  (equation (1)), Mg/Ca of G. ruber from sites where Δ[CO32−] > 10 μmol/kg in the western Indian Ocean represents temperatures close to warm season SST [Rayner et al., 2003] (Figure 3). Mg/Ca of G. ruber from OJP, Ceara Rise, Caribbean Sea and MAR sites where Δ[CO32−] > 10 μmol/kg, however, underestimates both warm season SST and annual average SST by at least 2°C (further discussion in section 5.3). Due to the disparity between SST and Mg/Ca of G. ruber, using SST to estimate Mg/CaINITIAL for G. ruber samples from the Ceara Rise and OJP gave Δ[CO32−]CRITICAL values, of 55 (±56) μmol/kg for the OJP and 38 (±8) μmol/kg for the Ceara Rise, which are much higher than estimates from any other sample set. Mg/Ca values of well-preserved samples from the shallowest site within each set (described in section 3.1.1) appear to offer a better estimate of Mg/CaINITIAL than SST for most sites for G. ruber.
 Mg/Ca of G. sacculifer from sites where Δ[CO32−] is above 10 μmol/kg gave temperatures (equation (2)) close to observed SST (Hadley Centre warm season average) (Figure 3). There is little difference in the regressions whether Mg/CaINITIAL is based on Mg/Ca of samples from the shallowest site within a sample set or on SST for this species (Table 2).
3.1.3. Unified Regression Between Δ[CO32−] and ΔMg/Ca
 The similarity of regressions between Δ[CO32−] and ΔMg/Ca for OJP, Ceara Rise and Indian Ocean sample sets (Table 2) supports the use of Δ[CO32−] proxies to correct for dissolution bias of Mg/Ca. Figure 4 shows one regression fitted for all samples for each species. ΔMg/Ca values are based on Mg/CaINITIAL estimated from SST for G. ruber and G. sacculifer from the Indian Ocean and on Mg/Ca of the shallowest sites within each transect (described in section 3.1.1) for all other sites. Correlation (r2) between Δ[CO32−] and ΔMg/Ca was above 0.60 for all four species (Figure 4). Sensitivity of ΔMg/Ca to dissolution was similar for G. sacculifer, N. dutertrei and P. obliquiloculata. Sensitivity was higher for G. ruber (slopes did not overlap at 95% confidence interval) than for the other three species.
4.1. Effect of Dissolution on Mg/Ca-Derived Temperatures
 Of the species examined, temperatures derived from the Mg/Ca of thermocline-dwelling foraminifera were the most disturbed by dissolution (Figure 7). The low Mg/Ca values of these species make the temperature inferred from them particularly sensitive to small Mg losses. In addition, tests of robust thermocline-dwelling species can sustain the loss of a large proportion of the test calcite before they finally disintegrate. Severely dissolved tests contain no inner calcite and consist only of outer crust (Figure 1d). Temperatures calculated from such remnants from the deepest sites on the OJP (where water depth is greater than 4000 m and Δ[CO32−] is around −22 μmol/kg), are lower by ∼8°C (13°C compared to 21°C) for N. dutertrei and by ∼6°C (17°C compared to 23°C) for P. obliquiloculata (using equations (2) and (3), respectively) compared to temperatures calculated from samples from the shallowest site (at 1614 m water depth).
 Sensitivity of Mg/Ca to Δ[CO32−] is similar for G. sacculifer and N. dutertrei, but the former is typically not preserved in samples at such low values of Δ[CO32−]. Severely dissolved G. sacculifer (XDX > 3) from the OJP yield temperatures 2–3°C lower than well-preserved G. sacculifer from the shallowest site. G. sacculifer from the deepest sites in the Indian Ocean sample set gave temperatures 4–6°C less than local SST. Such temperatures are found at water depths of ∼100 m for OJP and 110–135 m for the Indian Ocean. Rosenthal et al.  suggested that δ18O and Mg/Ca of progressively dissolved G. sacculifer reveals increasing depth of calcification. G. sacculifer continue to calcify in the thermocline [Duplessy et al., 1981; Rosenthal et al., 2000; Wilke et al., 2009] and temperatures calculated from partially dissolved samples thus plausibly represent the penultimate or final stages of calcification.
 The vertical attenuation of Mg/Ca in the water column tails off after a ∼0.6 mmol/mol decrease for G. ruber from the Ontong Java Plateau (Figure 2). Calculated temperatures are ∼1.5°C lower for the deepest OJP site where data exists than for the shallowest site (Figure 7). Sensitivity of Mg/Ca to Δ[CO32−] for G. ruber from our OJP sample set is similar to that found by Fehrenbacher et al.  for G. ruber from the Atlantic Ocean. In that study, the response of Mg/Ca of G. ruber to calcite undersaturation was similar to that of G. sacculifer and lower than that of N. dutertrei. However, despite this support for the low sensitivity of Mg/Ca of G. ruber to dissolution, considering the response of G. ruber from the OJP in isolation may be misleading. Temperatures calculated from neither δ18O (24.7°C using the species specific equation of Mulitza et al. ) nor Mg/Ca (25°C) of G. ruber from the shallowest (1616 m) sample from the OJP are in agreement with modern SST of 29°C (Figure 3). G. ruber can continue to calcify below the mixed layer [Duplessy et al., 1981] or favor the winter season [Wilke et al., 2009], which could explain the low SST estimated from its Mg/Ca in some locations. This is presumably not the case for OJP where a thick mixed layer exists and there is little seasonal change in temperature [Locarnini et al., 2006].
 The poor representation of SST by Mg/Ca of G. ruber from the OJP casts doubt on the sensitivity of Mg/Ca to Δ[CO32−] calculated for this sample set. G. ruber from the Ceara Rise, MAR, Caribbean Sea and western Indian Ocean all suggest greater sensitivity of Mg/Ca to Δ[CO32−] than does G. ruber from the OJP (Figure 2).This results in a larger offset in temperatures calculated from samples from deep compared to shallow sites for these sample sets than for the OJP. Temperatures derived from Mg/Ca of poorly preserved G. ruber from the western Indian Ocean (where Δ[CO32−] is −13 μmol/kg) are 4–5°C below local SST, consistent with calcification depths of 70–160 m. Atlantic samples are better preserved than those from Indian and Pacific sites. G. ruber from the deepest Atlantic sites (where Δ[CO32−] is −2.5 μmol/kg) give temperatures 2–3°C below SST, consistent with temperatures at 80–170 m water depth. These temperatures and depths are below the mixed layer and the photic zone.
 Unlike the other species analyzed here, G. ruber has no dissolution-resistant outer crust [Caron et al., 1990]. As dissolution proceeds, microporosity, indicated by light greyscale values in the CT scans, develops in the calcite of the entire test [Johnstone et al., 2010]. If this porosity is due to preferential leaching of Mg-rich calcite, it would explain why bulk Mg/Ca does not represent actual calcification temperatures of poorly preserved G. ruber.
 Leaching of the inner calcite occurs also in other species. CT scans show that the inner calcite of poorly preserved G. sacculifer, N. dutertrei (Figure 1c), and P. obliquiloculata is porous and dissolved. This leached porous calcite may have less effect on the Mg/Ca of species such as N. dutertrei and P. obliquiloculata where the outer crust dominates the signal, than on G. ruber.
 If temperatures calculated from severely dissolved N. dutertrei and P. obliquiloculata indicate accurate calcification temperatures for the outer, gametogenic calcite, these species continue to calcify deep within the thermocline. P. obliquiloculata, like N. dutertrei, inhabits the seasonal thermocline [Jones 1967; Curry et al., 1983; Faul et al., 2000] and both species are associated with the deep chlorophyll maximum [Ravelo and Fairbanks, 1992]. The deep chlorophyll maximum is at ∼110 m at on the OJP [Nathan and Leckie, 2009]. Temperatures from the most dissolved outer crust, of 12°C, are found well below this depth level, at ∼250 m, while 17°C represents water depths of ∼200 m. Calcification at such depths is supported by the depth integrated growth model of Wilke [2006, chap. 5]. They find calcification continues well below both the chlorophyll and temperature gradient maxima, to depths of 133 m for N. dutertrei and 344 m for P. obliquiloculata, at their Atlantic sites.
4.2. Two Systematic Biases in Mg/Ca–Derived Temperatures Over a Glacial Cycle
 The transfer of CO2 between ocean and atmosphere is associated with enhanced deep sea calcite preservation during deglacial transitions [Berger, 1977; Le and Shackleton, 1992; Hodell et al., 2001]. The associated shift in ocean Δ[CO32−] has been estimated at 25–30 μmol/kg for the deep Pacific during the last deglaciation [Marchitto et al., 2005]. Sensitivity of ΔMg/Ca to Δ[CO32−] (Figure 4) allows an estimate of dissolution bias in Mg/Ca derived temperatures connected with such a shift in deepwater calcite saturation. At sites where Δ[CO32−] decreased from 15 μmol/kg (where tests are well preserved) to −20 μmol/kg (tests are severely dissolved) temperatures of 21°C would be measured as 17°C by N. dutertrei, and P. obliquiloculata calcification temperatures of 23°C would be measured as 20°C.
 Theoretically, a decrease of 25 μmol/kg in Δ[CO32−] at the seafloor would result in typical tropical SST of 28°C [Locarnini et al., 2006] being underestimated by 4°C in G. ruber and 7°C in G. sacculifer. However, it may be that neither of these species would represent the entire 25 μmol/kg adjustment: tests of these species are sparse in our samples from sites where Δ[CO32−] is below −15 μmol/kg. A decrease in deepwater calcite saturation from 15 μmol/kg to zero μmol/kg would be recorded as an apparent cooling from 28°C to 26°C by G. ruber and from 28°C to 24°C for G. sacculifer. The absolute reduction in derived temperatures due to dissolution is greater at lower temperatures. SST of 25°C would be recorded as 21°C by G. ruber and 20°C by G. sacculifer for the same 15 μmol/kg decrease in Δ[CO32−].
 As suggested by Mix  such a preservation artifact in Mg/Ca–based SST could explain much of the dissimilarity in patterns of deglacial warming for the Pacific derived from different biotic carriers. SST derived from Mg/Ca of G. ruber for a site at 2830 m water depth the Pacific [Lea et al., 2006] diverges from that calculated from alkenones (which are not affected by calcite saturation state) at a nearby site [Prahl et al., 2006].
 The deglacial shift in Δ[CO32−] would increase Δ[CO32−] at the site from values close to zero to values above 15 μmol/kg, resulting in well-preserved tests during the transition. It is during this interval that the difference in SST recorded by the two proxies is greatest. During the deglaciation, Mg/Ca of G. ruber indicates temperatures of 25°C, 3°C above those recorded by alkenones. In the episode from 20 to 30 kyr before present, when calcite preservation was likely to be poor, SST derived from Mg/Ca was ∼22.5°C, only ∼1°C above alkenone-derived SST. Our calibrations (Figure 2) suggest that a reduction of 2°C in SST calculated from G. ruber can be explained by dissolution bias of Mg/Ca for such a change in calcite saturation. Variation in preservation could, therefore, explain the differing pattern of foraminiferal- and alkenone-based SST records.
5. Toward an Independent Correction for Dissolution Bias of Mg/Ca
5.1. Calibrations Between XDX and ΔMg/Ca
 The Δ[CO32−] values where Mg/Ca is first affected by dissolution (ΔMg/Ca = 0) of 11 (±4) to 15 (±5) μmol/kg) (Figure 4) are similar to the Δ[CO32−] values where dissolution is first detectable in CT (XDX = 0) of 12 (±4) to 14 (±5) μmol/kg) (Figure 5). For this reason regressions between XDX and ΔMg/Ca have been forced through the origin (Figure 6). Potential dissolution effects on Mg/Ca at Δ[CO32−] values above that where dissolution is detectable in CT are discussed in section 5.3.
 Correlation between XDX and ΔMg/Ca is strongest for the thermocline dwelling species (N. dutertrei and P. obliquiloculata). Tests of these species have a distinct outer crust which protects the inner calcite as it dissolves and the progress of dissolution through the test is clearly shown in the CT scans. Correlation between XDX and ΔMg/Ca is less well constrained for G. ruber. This species does not show the more advanced stages of dissolution as it has no outer crust. This may mean that XDX stages are more difficult to estimate. Another factor may be that G. ruber, which is fragile with thin walls, cannot sustain much dissolution and is lost more easily from the fossil assemblage. Therefore, the deviation in initial properties of samples from shallow and deep sites is expected to be larger for G. ruber than is the case for more robust species.
 Regressions plotted for each species can be used to estimate the reduction of Mg/Ca due to dissolution (ΔMg/Ca) according to
where q is the species specific gradient of the regression (Figure 6). ΔMg/Ca can then be combined with analyzed Mg/Ca (Mg/CaMEASURED) to provide an estimate of Mg/Ca free from dissolution bias, according to a rearrange of equation (5) to
5.2. Correcting Dissolution Bias of Mg/Ca Using XDX
 The effect of applying the corrections is to reduce scatter in temperatures derived from Mg/Ca for each sample set (Figure 7). For instance, the standard deviation on corrected temperature of the OJP sample set is on average ±0.6°C as compared to ±1.7°C of the uncorrected data set.
 Error bars shown in Figure 7 are calculated assuming a 5% error on the measured Mg/Ca portion. For the ΔMg/Ca portion the error is one standard deviation on the slope of the calibration between XDX and ΔMg/Ca. Errors for each portion are combined when the two values are summed. The propagated error on temperature is on average ±1.5°C for all samples. This is small compared to the systematic error caused by dissolution, which can be several degrees.
5.3. Is Mg/Ca Altered Before Dissolution is Detectable in CT?
 Although it is accepted that Mg/Ca is affected by dissolution in waters oversaturated with respect to calcite [Brown and Elderfield, 1996], literature values vary for where dissolution starts to distort Mg/Ca. Dekens et al.  suggest that Mg/Ca is affected at all depths below the surface and propose a linear correction based on water depth or Δ[CO32−]. Regenberg et al.  found threshold Δ[CO32−] values of between 22 and 25 μmol/kg for Caribbean G. ruber (white), G. sacculifer and N. dutertrei and assumed a linear decrease in Mg/Ca below these values.
 Our data do not clearly isolate the threshold of dissolution effects on Mg/Ca as we have few samples in the crucial interval where Δ[CO32−] is between 10 and 30 μmol/kg. Regressions between Δ[CO32−] and ΔMg/Ca give Δ[CO32−]CRITICAL values of between 11 (±4) and 15 (±5) μmol/kg (Figure 4). These Δ[CO32−]CRITICAL values are constrained for some sample sets as Mg/CaINITIAL was defined as Mg/Ca of the sample from the shallowest site within the set (section 3.1.1). If dissolution affects Mg/Ca at water depths above that of this sample, then our Δ[CO32−]CRITICAL values would be too low. Values of Δ[CO32−]CRITICAL (of between 11 (±4) and 15 (±5) μmol/kg) established in this study are lower than those of Regenberg et al.  (of between 22 and 25 μmol/kg), although, given the uncertainty on the intercepts of the regressions and on calculated Δ[CO32−] these differences may not be significant.
 Coincidence in Δ[CO32−] values where XDX and ΔMg/Ca are zero (Figures 4 and 5) suggests that decrease in sample Mg/Ca is related to observable signs of test breakdown. However, we must consider the possibility that our Δ[CO32−]CRITICAL values are underestimates. This would mean that although CT offers a direct estimate of sample preservation state, it does not capture early diagenetic changes in Mg/Ca.
Schiebel  and Schiebel et al.  estimate that significant dissolution of foraminiferal calcite occurs in the “twilight zone” of the top 1000 m of the ocean, yet CT shows tests to be well preserved from sites where Δ[CO32−] of deep water is above 15 μmol/kg (Figure 5). It may be that the first stages of dissolution selectively remove certain fractions of the sample population while leaving the remaining tests relatively unaffected. If these early stages of dissolution affect Mg/Ca, then apparently well-preserved samples could be biased by dissolution. For example, if Mg/Ca of N. dutertrei declines from the Δ[CO32−] value of 23 μmol/kg identified by Regenberg et al. , at sites where dissolution is first detectable in CT (where Δ[CO32−] is 13 μmol/kg) a sample with a calcification temperature of 21°C would record 19°C. In the case of G. ruber, if Mg/Ca declines from 25 μmol/kg [Regenberg et al., 2006], samples from sites where dissolution is first apparent in CT scans (at 14 μmol/kg) would underestimate calcification temperatures by 3°C where SST was 25°C and 2°C where SST was 29°C.
 Dissolution above the limit of dissolution detection in CT could therefore explain more than half of the offset between temperatures derived from Mg/Ca of G. ruber from the shallowest core top from the OJP (1BC3), of 26°C, and SST at the site, of 29°C [Locarnini et al., 2006]. However, this mechanism does not explain why Mg/Ca of G. ruber from sites in the Indian Ocean where, like OJP site 1BC3, Δ[CO32−] values are between 11 and 17 μmol/kg, do seem to represent SST. The reasons for the offset in Mg/Ca and SST of G. ruber from the OJP remain unclear.
 More work on the early effects of dissolution on Mg/Ca is required to identify dissolution thresholds more securely and to ascertain whether these threshold values are globally valid. Another uncertainty in the use of dissolution assessment to correct Mg/Ca arises from the possibility that the sensitivity of Mg/Ca to dissolution may not be constant. In our study, XDX offers a poor correction of dissolution bias for N. dutertrei from the Caribbean, and it appears that Mg/Ca of this species is more sensitive to dissolution at this site than at other locations. However, despite these potential limitations, XDX does offer a direct insight to preservation state of tests and, at the least, identifies where Mg/Ca is unreliable.
5.4. Mg/Ca and Dissolution Susceptibility of Foraminiferal Calcite
 Impurities in the crystal lattice increase dissolution susceptibility of calcite. Mg2+ is the major impurity in foraminiferal calcite and Mg content has been used to explain the dissolution susceptibility of biogenic calcite in waters oversaturated with respect to calcite [Walter and Morse, 1985; Brown and Elderfield, 1996; Yu et al., 2007].
 As described more fully by Johnstone et al. , the pattern of dissolution shown in CT scans is consistent with preferential dissolution of Mg-rich calcite. Figure 1 shows tomography scans of N. dutertrei tests illustrating progressive stages of dissolution. Compared to a well-preserved sample (Figure 1a), tests exhibiting the early effects of dissolution (Figure 1b) have less sharply defined inner chambers and show a contrast between light colored inner calcite, and dark colored calcite of the outer wall. This “inner calcite” is the area of the test richest in Mg. Around the depth of calcite saturation horizon all of the inner calcite is partially dissolved and porous (Figure 1c). In severely dissolved samples, only the low-Mg outer crust [Sadekov et al., 2005] remains (Figure 1d).
 Although test calcite rich in Mg dissolves preferentially, species with high Mg/Ca do not appear to be more sensitive to dissolution than those inherently low in Mg/Ca. Mg/Ca of G. ruber, the species where Mg/Ca is highest, may show the greatest reduction for a given change in calcite saturation (section 4.2 and Figure 4), however Mg/Ca of G. sacculifer, which is only slightly lower than that of G. ruber, is similar to P. obliquiloculata in its sensitivity to dissolution, despite the lower Mg/Ca of the latter species (Figure 2). Other factors which influence dissolution susceptibility, such as crystallinity, thickness of the test walls or the proportion of crust to inner calcite, presumably also influence dissolution sensitivity of Mg/Ca for each species.
 Although interspecies differences in Mg/Ca do not appear to control sensitivity of Mg/Ca to dissolution, variation within one species may have an effect. Sensitivity to dissolution of N. dutertrei Mg/Ca is highest for the Caribbean (Figure 2). Tests of N. dutertrei from this site have the highest Mg/Ca of the N. dutertrei samples in this study (Figure 2). Mg/Ca of well-preserved N. dutertrei from the Caribbean is 26% higher than it is for the Mid-Atlantic Ridge and 30% higher than for the Ontong Java Plateau. This intraspecies variation in Mg content may affect dissolution susceptibility.
5.5. Comparison of Two Methods of Correcting Mg/Ca: Test Mass and XDX
 Decrease in test mass has been used to correct for the effects of dissolution on Mg/Ca for planktonic foraminifera [Rosenthal and Lohmann, 2002]. In order to judge how well dissolution index XDX compares to the test mass method, we compare the correlation coefficients (r2) between (1) ΔMg/Ca and test mass and (2) ΔMg/Ca and XDX for each species. Test mass values for these samples are from Johnstone et al. .
 Correlation (r2) between ΔMg/Ca and XDX is 0.40 for G. ruber; 0.60 for G. sacculifer; 0.61 for N. dutertrei and 0.87 for P. obliquiloculata. These are higher than correlations between ΔMg/Ca and test mass of 0.31, 0.54, 0.37 and 0.64 for the above species. Test mass is controlled by environmental factors and differs between sites, particularly for thermocline dwelling species due to the variation in the thickness of the outer crust. XDX offers an assessment of dissolution which is largely independent of test mass [Johnstone et al., 2010]. In the early to mid stages, dissolution does not significantly alter test wall thickness but acts to increase porosity of the test calcite. The same stage of dissolution is identifiable in light and heavy tests. This makes XDX more widely applicable as an indicator of dissolution bias than test mass.
6. Effect of Dissolution on Foraminiferal δ18O
 Previously published δ18O for G. sacculifer and P. obliquiloculata from the OJP shows a large amount of scatter. Wu and Berger  found 90% of δ18O data fell in a range of ∼0.4 ‰ Our data show a similar range (Figure 8). As illustrated in Figure 1, the effect of dissolution is to selectively remove the inner parts of the calcite test. While it might be supposed that this would shift δ18O toward values of the deeper colder waters where the outer calcite is precipitated, in fact there is little correlation between δ18O and XDX (Figure 9). Although the highest values of δ18O generally came from the most dissolved tests (XDX > 3), well-preserved tests also cover a range of values. For example, well-preserved and poorly preserved N. dutertrei from the OJP have similar δ18O values (Figure 9). The samples used in this study were small, containing between five and nine tests. Probably such a small number is insufficient to average out the range of initial values of δ18O. The wide age span of the OJP samples [Barker et al., 2007] and the range of calcification depths, especially for thermocline dwellers, means that the initial variability could be very high.
 All four species of planktonic foraminifera used in this study show a strong linear decrease in Mg/Ca with decreasing Δ[CO32−] and this decrease started in waters oversaturated with respect to calcite. ΔMg/Ca (reduction in Mg/Ca due to dissolution) for each species decreased linearly from Δ[CO32−] values of between 10 (±4) (G. sacculifer) and 15 (±5) μmol/kg (N. dutertrei). These threshold values for dissolution effects on Mg/Ca are minimum estimates as they are limited by the assumptions made in calculating ΔMg/Ca. Sensitivity of Mg/Ca to dissolution was greatest for G. ruber and was similar for G. sacculifer, N. dutertrei and P. obliquiloculata.
 Dissolution can have a large effect on Mg/Ca–derived temperatures. Mg/Ca ratios of poorly preserved samples underestimate temperatures by 4–6°C for G. ruber and G. sacculifer; 9°C for N. dutertrei and 6°C for P. obliquiloculata.
 The CT-based deepwater Δ[CO32−] proxy, XDX, correlates positively with ΔMg/Ca. Calibrations provided here provide an independent method of estimating dissolution bias of Mg/Ca–derived temperatures.
 Attempts to correct Mg/Ca for dissolution would be confounded if the response of Mg/Ca to partial solution of the test were variable. N. dutertrei illustrates that this is a possibility. Mg/Ca of Caribbean N. dutertrei apparently shows a greater sensitivity to dissolution than those from other sites. N. dutertrei from the Caribbean has the highest Mg/Ca of all the N. dutertrei samples measured here and it is suggested that variation in Mg/Ca within one species may control dissolution susceptibility.
 Correlation (r2) was greater between ΔMg/Ca and XDX than it was between ΔMg/Ca and test mass. Unlike test mass, which can vary temporarily and spatially, XDX offers a direct measure of dissolution. The small sample size typically used for δ18O analysis is insufficient to isolate the effect of dissolution from initial variability in δ18O, due to age and calcification depth range, in these samples.
 This study was funded through DFG-Research Center/Cluster of Excellence “The Ocean in the Earth System.” Thanks go to Christina de la Rocha, whose comments improved an earlier version of this manuscript, and to three anonymous reviewers for useful comments and suggestions. Thanks to Karen Alexander for proofreading.