Influence of volcanic shards on foraminiferal Mg/Ca in a core from the Galápagos region



[1] Variability in the Mg, Fe, Al, Mn, and U to Ca composition of the planktonic foraminifera G. ruber in core TR163-22, an 8.8 m long piston core from just northwest of the Galapagos Islands, reveals distinct patterns of downcore behavior. Mg/Ca data follow a general climatic trend, with lower values during cold intervals (i.e., marine isotope stages (MIS) 2–4) and higher values during warm intervals (i.e., MIS 1 and 5). Fe/Ca and Al/Ca data reveal distinct intervals in which concentrations are elevated tenfold or more over baseline values, with highly variable elemental ratios in these intervals. The Mg/Ca data are highly correlated with the Fe/Ca and Al/Ca data in these intervals, with slopes of 1.13 ± 0.03 and 0.66 ± 0.07 mole/mole, respectively. Examination of the sediments from these intervals reveals that they are rich in volcanic debris that is likely derived from Galápagos volcanic rocks. The presence of small amounts of this volcanic debris can bias shell Mg/Ca to elevated values. Attempts to use enhanced cleaning to remove the volcanic particles were largely unsuccessful. The most successful strategy is to use the Fe/Ca and Al/Ca from the shell digests along with the Fe to Mg ratio of the volcanic debris to correct the Mg/Ca data for Al-silicate contamination, in which case the corrected data more clearly reveal the primary climatological patterns. Mn/Ca and U/Ca data are characterized by cycles of 20–200 cm length, with Mn/Ca having a distinct trend toward higher values between the core top and 200 cm. These trends appear to be related to diagenetic coatings on the foraminifera shells, most likely in the form of Mn-carbonates.

1. Introduction

[2] Since the development of the Mg/Ca paleothermometry technique in the mid-to-late 1990s, there has been a growing appreciation of the potential influence of secondary, postdepositional sedimentary phases such as Mn-oxides, Mn-carbonates and Al-silicates on shell Mg [Barker et al., 2003; Benway et al., 2003]. The secondary addition of Mg is of sufficient magnitude to compromise the accuracy and precision of Mg paleothermometry. This effect varies widely between basins, within depositional environments (i.e., oxic versus suboxic, carbonate ooze versus detrital-dominated, volcanic-rich sediments) and as a function of downcore depth within individual cores. For this reason, sites must be evaluated on a site-by-site basis to determine the potential level of influence of secondary factors. Furthermore, the effectiveness of any cleaning protocol is likely to vary from site to site.

[3] Improvements in analytical instrumentation and methodology, which enable researchers to analyze a much broader range of elemental abundances in foraminifera than previously possible, provide a clear step forward in assessing the influence of secondary phases. Analysis of elements associated with detrital phases (i.e., Fe and Al) and those associated with secondary coatings (i.e., Mn, Fe and U) make it possible to pinpoint which factors are affecting Mg/Ca signals [Barker et al., 2003]. At UCSB, this additional analytical capability was not realized until late 2000, when we installed a Finnigan Element2 sector ICP-MS. The multiple resolution capability of the Element2 enables simultaneous analysis of elements such as Fe and Al, which require high resolution to separate them from adjacent plasma peaks, along with Mg, Sr and other metals (see section 3).

[4] Our climate reconstructions in the eastern tropical Pacific have focused on a series of piston cores that were raised from bathymetric highs in the Galapagos region on cruise Trident 163 in 1975 (J. Kennett, PI). Previously, we have published planktonic Mg/Ca determinations from core TR163-19 on the Cocos Ridge just north of the Galapagos Islands [Lea et al., 2000; Dekens et al., 2002; Lea, 2004]. As part of a broader paleoclimate study (D. W. Lea et al., Paleoclimate history of Galápagos surface waters over the last 135,000 years, submitted to Quaternary Science Reviews, 2005) (hereinafter referred to as Lea et al., submitted manuscript, 2005), we have generated high resolution Globigerinoides ruber Mg/Ca and associated metal records from core TR163-22, which was recovered from just northwest of the Galapagos Islands. The vast majority of analyses include Fe and Al determinations. In this study we focus interpretation on the associated metal records in TR163-22, and in particular evaluate how well Fe, Al and Mn data can be used to assess secondary influence on Mg/Ca. We also develop a correction protocol for Mg/Ca in core TR163-22, whereby measured Al and Fe are used to subtract the Mg contribution suspected to come from volcanic ash.

2. Core Locations, Setting, and Sampling

[5] Core TR163-22 (92°23.9′W, 0°30.9′N, 2830 m water depth) lies on the distal south flank of the Galápagos Spreading Center, northwest of the Galápagos Islands (Figure 1). The calcite lysocline in the Panama Basin, to the east of our study site, is estimated to occur at ∼2900 m [Thunell et al., 1981]. Generally, however, dissolution is less intense in the open Pacific to the west of the Panama Basin [Kowsmann, 1973; Moore et al., 1973]. Dissolution and breakup of planktonic foraminifer shells starts to be evident in eastern equatorial Pacific sediments at ∼2500–3000 m water depth [Kowsmann, 1973]. Foraminifer shells from the core top of TR163-22 do show signs of dissolution and fragmentation, but there are sufficient intact individuals available for analysis. Deeper in the core and especially in the glacial sections, shell preservation is noticeably superior to the core top. The weight of Globigerinoides ruber specimens from the 250–350 μm fraction in TR163-22 readily exceeds 10 μg throughout the length of the core (see below), above the informal cutoff that UCSB uses to indicate when shell specimens are heavily dissolved. Total sedimentary CaCO3% ranges between 30 and 70%, with a mean of 47% (T. Herbert, Brown University, unpublished data, 1999).

Figure 1.

Bathymetric map of the Galapagos Islands region with the location of cores TR163-19 and TR163-22, the focus of this study, shown. Data from Global Bathymetry Model v. 7.2 (courtesy D. Wilson, UCSB). The locations of the Galapagos Spreading Center, Cocos Ridge, and Carnegie Ridge are indicated.

[6] Because of the abundant volcanoes on the Central American Margin, ash layers are a common feature of deep-sea sediments from the eastern equatorial Pacific [Kowsmann, 1973; Ninkovich and Shackleton, 1975; Drexler et al., 1980]. The description of core TR163-22 contains many references to suspected ash layers, but mapping of distinct layers like ash layer L suggest that ash from Central American volcanoes does not travel west of ∼87°W, implying that such ash would not reach the site of TR163-22 in any significant quantities [Ninkovich and Shackleton, 1975]. In a survey of core-top sediments of the eastern equatorial Pacific, volcanic glass shards of a distinct brown color were observed just west of the Galapagos Islands, close to the location of core TR163-22 [Kowsmann, 1973]. Examination of sediments from intervals in TR163-22 identified as enriched in volcanic debris (see below) indicates the presence of distinct brown volcanic shards, as well as colorless glass flakes and even large (∼0.5 mm) fragments of what look like basaltic volcanic rock. The brown shards, some of which are >150 μm and of a blocky character, appear to correspond to the description by Kowsmann [1973], who attributes their presence to sub–sea surface eruptions (i.e., phreatomagmatic explosion), which result in rapid quenching and dispersal of volcanic glass.

[7] Core TR163-22 was sampled at continuous 2 cm resolution throughout its 8.8 m length. Stable isotopes and radiocarbon determinations indicate that core TR163-22 spans ∼130,000 years, with its base corresponding to Termination II (Lea et al., submitted manuscript, 2005). Sedimentation rates vary from ∼9 cm/kyr in the top 6 m to ∼4 cm/kyr in the lower 3 m.

3. Methods for Metal Analysis

[8] For TR163-22, we picked 50–60 individuals of G. ruber-white variety from the 250–350 μm fraction for two metal analyses. The average weight of G. ruber shells was 11.2 ± 1.1 μg (errors are quoted as 1 σ unless otherwise indicated). There are subtle variations in shell weight throughout the core, from a minimum of ∼10 μg at 330 and 800 cm to a maximum of ∼13 μg at 160 cm. All of the shells from each interval were gently crushed between wide glass slides and separated into two approximately equal aliquots. These two aliquots were cleaned and analyzed separately to assess reproducibility.

[9] Shell samples were cleaned using the UCSB standard foraminifera cleaning procedure, without the DTPA (diethylene triamine pentaacetatic acid) step [Lea and Boyle, 1993; Lea et al., 2000; Martin and Lea, 2002]. Dissolved samples were analyzed by the isotope dilution/internal standard method [Lea and Martin, 1996; Lea et al., 2000, 2003] using a Thermo Finnigan Element2 sector inductively coupled plasma mass spectrometer (ICP-MS). Analytical reproducibility, assessed by analyzing consistency standards matched in concentration and Mg/Ca ratio to dissolved foraminifera solutions and analyzed over the course of the entire study (∼30 months), is estimated at ±0.9% (1 σ). The pooled standard deviation of replicate Mg/Ca analyses from TR163-22 was ±0.11 mmol/mol, or ±4.5% (1 SD, df = 428), significantly worse than the ∼3% value typical for tropical cores [Lea et al., 2000]. Less than 0.5% of the individual analyses (4 out of 854) were rejected as outliers.

[10] Elemental ratios of Mn/Ca, Fe/Ca, Al/Ca and U/Ca were analyzed at the same time as Mg/Ca to assess cleaning efficacy. Fe, Al, and U were only determined in some of the samples from the top 2 m of TR163-22 because the analysis of these samples occurred before implementation of the analytical protocol for these elements. U was determined by isotope dilution using 235U as the spike, with an estimated error on U/Ca based on matched consistency standards of 3.3%. Elemental ratios of Mn, Al and Fe (the latter two in medium resolution) were determined by direct reference to the two isotopes used for quantification of calcium, 43Ca and 48Ca. Analytical error for Mn/Ca based on matched consistency standards is estimated at 1.4%. Errors for Al/Ca and Fe/Ca determination are estimated at <5%.

[11] The analytical protocol we use for calcium is what led to the discovery of a cleaning problem in core TR163-22. During the analysis of samples from the 221–281 cm interval, we encountered 48/43Ca ratios significantly higher than we had previously measured. After extensive experimentation we narrowed the cause of these higher ratios to elevated Ti in the shell digests, which boosts the measured 48/43Ca ratio by contributing 48Ti. The presence of Ti, which unambiguously points to Al-silicate contamination, led us to add Fe and Al as part of our standard protocol for analyzing shell digests.

[12] Previous published analyses using quadrupole ICP-MS from core TR163-19 [Lea et al., 2000; Dekens et al., 2002], which lies north of the Galapagos Islands on the Cocos Ridge (Figure 1), did not include Fe or Al determinations. The 76 additional analyses from TR163-19 included by Lea [2004], however, were made by sector ICP-MS and include Al, Fe and U, which allowed for an evaluation of potential contamination problems at this site (see below).

4. Results

[13] Average elemental ratios to Ca of U, Mn, Al, Fe and Mg in cleaned G. ruber shells are shown as a function of core depth in Figure 2. The approximate intervals corresponding to marine isotope stages (MIS) 1–5, based on oxygen isotope stratigraphy, are indicated (Lea et al., submitted manuscript, 2005). The trends in elemental ratios show clear but contrasting patterns, which fall into three broad groups of behavior: (1) Mg/Ca, dominated by a long-term (∼100 kyr) periodicity with low values during glacial intervals (i.e., MIS 2 through 4, between ∼200–600 cm) and higher values during interglacial intervals (i.e., MIS 1, 0–100 cm, and MIS 5, 600–850 cm); (2) Fe/Al and Al/Ca, which correspond to each other very strongly and which are characterized by elevated intervals, with high variability, at 200–300 cm, 500–600 cm and ∼800 cm; and (3) Mn/Ca and U/Ca, both showing clear, but not always corresponding, cyclicity with wavelengths ranging between ∼0.2 and 2 m.

Figure 2.

Downcore data for Mg, Al, Fe, U, and Mn to Ca ratios determined from G. ruber shells (250–350 μm, white variety) in core TR163-22 (92°23.9′W, 0°30.9′N, 2830 m water depth). Samples span the core continuously at 2 cm spacing. Each point represents the average of 2 or more splits, separately prepared and analyzed. The estimated 1σ analytical errors are ±0.9% for Mg/Ca, ±1.4% for Mn/Ca, ±3.3% for U/Ca, and <±5% for Al/Ca and Fe/Ca. The approximate position of marine isotope stages 1–5 are indicated, as determined from the benthic oxygen isotope stratigraphy of the core (Lea et al., submitted manuscript, 2005).

[14] Simple correlations, determined by regression analysis between individual (as opposed to average) analysis, reveal that the strongest correlation in the metal data set is between Fe/Ca and Al/Ca, with an r2 value of 0.86 (df = 650, P ≪ 0.001) (Figure 3). This correlation suggests that both Fe and Al in the shell digests reflect silicate contamination, because there is no other likely source for the Al. The constancy of the molar Fe/Al value, 0.556 ± 0.018 (95% CI), suggests that the composition of the Al-silicate contaminant is quite uniform throughout the core. Because the site of TR163-22 is far from mainland continental weathering sources, it is likely that the Al-silicate contaminant is volcanic shards, ash and its weathered products, derived from Galápagos volcanoes.

Figure 3.

Comparison of measured Fe/Ca and Al/Ca in individual analyses (splits) of G. ruber shell digests. The strong correlation indicates that both elements are derived from Al-silicate debris, most likely derived from Galápagos volcanic rocks. The positive intercept for Fe/Ca, 29 ± 3 μmol/mol, suggests that a small residual of the Fe is not associated with Al-silicate debris.

[15] The U/Ca data show a correlation to Mn/Ca, with an r2 of 0.24 (df = 650, P ≪ 0.001). The slope of this relationship is 0.14 ± 0.02 mmol/mol U/Ca per mol/mol Mn/Ca (95% CI). The intercept is 1.5–8.3 nmol/mol U/Ca (95% CI), consistent with values of U/Ca in cultured planktonic foraminifera [Russell et al., 2004]. This correspondence can also be seen in the depth domain (Figure 2). The correlation between Mn and U is strongest between ∼200 and 300 cm, where r2 reaches 0.78, but there is a relatively consistent correspondence throughout the core length of TR163-22.

[16] Over the entire data set, there are weak but significant (P ≪ 0.001) correlations between Mg and the elements Fe, Al and U, but with very low r2 values ranging between 0.04 and 0.05, suggesting a very small (<5%) fraction of common variance. There are two intervals of the core, corresponding to the intervals of highly variable Fe/Ca and Al/Ca, where the correlations between Mg and both Al and Fe are greatly enhanced: 220–282 cm and 518–618 cm. Throughout both of these intervals, and for both element pairs, the correlations have r2 values between 0.5–0.6 (i.e., 50–60% shared variance). On the basis of the 220–282 cm interval, the slopes of the relationships are 1.16 ± 0.03 for Mg/Ca versus Fe/Ca and 0.73 ± 0.15 for Mg/Ca versus Al/Ca (Figure 4). Both correlations have the same intercept within 95% CIs: Mg/Ca = 2.09 ± 0.06 mmol/mol. The statistics for the 518–618 cm interval are very similar, except for the slope of the Mg/Ca versus Al/Ca relationship, which differs by ∼20% between the two intervals (Figure 4). The similarity of these two pairs of correlations in two different intervals of the core suggests that they arise from a common process or contaminant. The uniform slopes and intercepts of the correlations are consistent with a contaminant characterized by near-constant ratios of Al:Fe:Mg, such as volcanic Al-silicate debris.

Figure 4.

(a) Correlation between Mg/Ca and Fe/Ca (both in mmol/mol) from G. ruber shell digests in the 220–282 cm and 518–618 cm intervals of core TR163-22. Volcanic shards are concentrated in both of these intervals. The slope indicates that there are ∼1.1 moles of Mg associated with each mole of Fe in the volcanic material. The intercept gives the average “Fe-free” Mg/Ca value in these intervals, 2.1 mmol/mol. (b) Correlation between Mg/Ca and Al/Ca (both in mmol/mol) in the 220–282 cm and 518–618 cm intervals. The slopes indicate that there are ∼0.6–0.7 moles of Mg associated with each mole of Al in the volcanic material. The intercept for both intervals is the same within error as found for the Mg/Ca versus Fe/Ca plot.

5. Discussion

5.1. Origin of Fe, Al, and Mg in Shell Digests

[17] It has recently been recognized that even with rigorous purification methods, at certain sites Al-silicate contamination can influence shell Mg/Ca ratios [Barker et al., 2003]. An important advance in this regard is the ability to monitor Al/Ca and Fe/Ca in shell digests [Barker et al., 2003; Skinner et al., 2003]. Aluminum is an especially unambiguous indicator for the presence of detrital grains [Emiliani, 1955]. Iron is more ambiguous because it can be elevated due to other processes, such as oxide formation or secondary phases such as pyrite (D. Pak, UCSB, unpublished data, 2002). The potential for Mg contamination does not appear to depend so much on the presence of detrital grains but rather on their nature. For example, our determinations from the nearshore Cariaco Basin, which has high detrital abundances, indicate negligible levels (typically <0.01 mmol/mol) of Fe/Ca and Al/Ca on shells throughout a 10 m sequence [Lea et al., 2003]. Samples reanalyzed from core TR163-19 on the Cocos Ridge [Lea et al., 2000; Lea, 2004], which lies in a depositional environment in some ways similar to TR163-22 (although much farther from volcanic sources on the Galapagos Islands), also have negligible levels of Al and Fe. Thus the potential for Mg contamination by Al-silicate grains probably depends on the type of grains (i.e., their maturity, origin, size and shape) as well as the diagenetic history of the site.

[18] The results from core TR163-22 indicate that there is a very strong correlation of Fe/Ca and Al/Ca throughout the core (Figure 3). The constancy of this relationship suggests that the source of both Fe and Al in the shell digests is dominantly from Al-silicate grains (the positive Fe/Ca intercept of 29 ± 3 μmol/mol suggest that a small fraction of the Fe might be associated with another phase). The slope of the Fe-Al relationship, 0.56 ± 0.01, is equivalent to an FeO-Al2O3 weight ratio of ∼1:2, consistent with the composition of Galápagos volcanic rocks over a wide range of Mg contents (B. M. Gunn, Geochemistry of igneous rocks, 2005, available at

[19] The available data suggest that the intervals with elevated Al/Ca and Fe/Ca (most markedly 220–280 cm and 520–620 cm, but also 750–825 cm) are rich in volcanic shards. The original core sample descriptions for TR163-22 indicate suspected ash layers at 584, 782–787, 787–801, and 820–822 cm (unpublished report, Sedimentological Laboratory, Graduate School of Oceanography, University of Rhode Island). All of these intervals have elevated Al and Fe (Figure 2). The presence of volcanic shards is well documented in the Galápagos region [Kowsmann, 1973]. Galápagos volcanic rocks are typically tholeiitic basalts, with average (range) Mg:Fe:Al molar ratios of 1.2 (0.1–2) to 1.0 (0.5–1.4) to 1.0 (0.7–1.2) (B. M. Gunn, Geochemistry of igneous rocks, 2005, available at Peak values of Al/Ca in the shell digests from TR163-22 are ∼600 μmol/mol, which, given the relatively uniform Al2O3 weight percentage of ∼15 ± 3% in Galapagos volcanic, corresponds to a weight percentage of volcanic material in the shell digests of up to 0.1%, a level that is quite conceivable as a residual after the cleaning process [Barker et al., 2003].

[20] In the most conspicuously elevated intervals of TR163-22 (220–282 cm and 518–618 cm), there is a clear peak-to-peak correlation in the depth domain of Mg/Ca with both Al/Ca and Fe/Ca (Figure 5). The correlation of Mg to Fe and Al in the shell digests strongly suggest that some of the observed Mg variability is not associated with primary shell Mg, but rather reflects postdepositional contamination by volcanic phases. Support for this hypothesis comes from our observation that when dissolved shell archives from early analyses of TR163-22 samples were reanalyzed after two years, Al/Ca and Fe/Ca levels were up to 10 times higher and Mg/Ca levels up to 50% higher than the levels for freshly dissolved samples from the same interval. We interpret this observation as indicating that fine-grained residual insoluble material in the shell digests slowly dissolves in archived samples. We also observed that many of the samples from the elevated intervals had visible insoluble grains present after dissolution. For this reason, we centrifuged all samples immediately after dissolution and adjusted the sample probe to not reach the bottom of the vial to minimize any contribution of residual insoluble material (this is our standard procedure).

Figure 5.

Mg/Ca, Fe/Ca, and Al/Ca data from G. ruber shell digests in core TR163-22, between 100 and 700 cm. Two intervals, between ∼180–280 cm and ∼520–630 cm, have consistently elevated Al and Fe values indicating the presence of volcanic shards associated with the foraminifer shells. In both of these intervals, Mg/Ca shows a point-to-point correlation with Fe/Ca and Al/Ca. All three elemental ratios are given in mmol/mol to aid comparison. On the basis of the correlations in the shard-rich intervals, a 0.5 mmol/mol Fe/Ca level is associated with a Mg/Ca contribution of ∼0.6 mmol/mol, and a 0.5 mmol/mol Al/Ca level is associated with a Mg/Ca contribution of ∼0.3 mmol/mol. Note that the millennial-scale climatological cycles in Mg/Ca, typically 25–50 cm in length, are obscured in the volcanic shard-rich intervals by the excess Mg derived from Al-silicate contamination. This is especially clear in the 520–620 cm interval.

[21] The presence of Fe in association with plankton shells raises the question of whether enhanced Fe concentrations in the waters west of the Galápagos Islands, which support higher productivity there compared to the surrounding high nutrient low chlorophyll areas of the eastern equatorial Pacific [Gordon et al., 1998; Lindley and Barber, 1998], might have their origin in volcanic inputs. If this is true, foraminiferal Fe concentrations or shard abundance in cores like TR163-22 might serve as a proxy of regional Fe fluxes to surface waters. An interesting test of this hypothesis would be to compare downcore productivity indicators to the trends in Fe concentration (M. Kienast, Dalhousie University, personal communication, 2005).

5.2. Strategies to Determine True Calcite-Bound Mg/Ca

[22] The excess, non-calcite-bound Mg measured in shell digests from TR163-22 will bias SSTs calculated from Mg/Ca to higher values. We tried several different strategies to obtain Mg/Ca values most representative of the original shell material: (1) enhanced cleaning, by lengthening the ultrasonication time and adding extra cleaning steps; (2) enhanced cleaning, by picking out foreign grains [Barker et al., 2003]; and (3) correction of the original Mg/Ca data by subtracting a constant Mg/Fe ratio using the observed Fe/Ca of the sample [Barker et al., 2003]. We found that the last strategy yielded the most uniform improvements and most consistent final data.

[23] For enhanced cleaning, approach (1), using 30 samples ranging from 161 to 522 cm, we first tried lengthening the ultrasonication steps associated with the “physical cleaning” from 30 seconds to 1 minute and adding one additional water and one additional methanol ultrasonication step. These changes had no measurable effect on shell Mg/Ca, Al/Ca and Fe/Ca.

[24] For enhanced cleaning, approach (2), using 20 samples from between 221 and 281 cm, we picked out all visible noncarbonate grains [Barker et al., 2003] after the full purification procedure but prior to the transfer to clean vials. Comparing the results to our normal cleaning procedure reveals a 69 μmol/mol decrease in Fe/Ca (t-test for paired two sample means yields P < 0.02), a 100 μmol/mol decrease in Mg/Ca (P < 0.04), and a 30 μmol/mol decrease in Al/Ca (not significant)) (Figure 6). The ratio of Mg to Fe removal is ∼1.4, close to the inferred Mg to Fe ratio of the shards, which is 1.1 (i.e., slope in Figure 4a). Even with enhanced cleaning, however, the correlation between Fe/Ca and Al/Ca (P < 0.006) and between Mg/Ca and Fe/Ca (P < 0.06) are still present. In fact, the line fits of the correlations in the data from this experiment are identical, indicating that the volcanic shards cannot be completely removed by visual identification, probably because it is it not possible to see all of the volcanic particles. The extra cleaning step, which lengthens the cleaning procedure by several hours, also did not significantly change the trends in the data in the depth domain (Figure 6).

Figure 6.

Comparison of Mg/Ca and Fe/Ca in G. ruber shell digests from TR163-22 samples cleaned using the standard UCSB procedure versus samples cleaned with the addition of a step that removes non-calcite particles by visual identification [Barker et al., 2003]. The additional step significantly lowers Fe/Ca in the digests but has only a small influence on Mg/Ca. The correlation between Mg/Ca and Fe/Ca is the same in both sets of samples, indicating that volcanic shards cannot be completely removed by visual identification.

[25] The final strategy we tried, and ultimately adopted, was to establish a Mg/Ca to Fe/Ca ratio for the contaminant that could be used in a correction scheme [Barker et al., 2003]. This approach assumes that all of the Fe in the samples, and the accompanying Mg, is not shell bound and therefore should be removed by subtraction. The advantage of this approach is that it can applied to all the data that had been collected using our normal cleaning procedure. A potential weakness of this approach is that it assumes a constant Mg/Fe ratio of the contaminant throughout the core. This is probably a safe assumption at this site because the volcanic shards probably come from a single island or source, but it does not provide a general solution where Al-silicate contaminants might have multiple sources with different compositions.

[26] We used the intervals 220–280 cm and 518–618 cm, which exhibit the most systematic relationship between Mg/Ca, Fe/Ca and Al/Ca, to derive an algorithm for Al-silicate correction (Figure 4). The correlations between Mg/Ca and both Fe/Ca and Al/Ca are very clear in both of these intervals. Because mean Mg/Ca is relatively constant over these intervals (Figure 2), the intercept of the correlations (Figure 4), which is identical within error, likely represents the predepositional (i.e., uncontaminated) value of foraminiferal Mg/Ca. We tried two schemes for correcting the data. In the first, we simply used the molar Mg/Ca to Al/Ca ratio, 0.66 ± 0.07, derived from the slope of the data (Figure 4b), to correct each point by subtracting the measured Al/Ca of the sample multiplied by 0.66. In the second, we used the molar Mg/Ca to Fe/Ca ratio, 1.13 ± 0.03, derived from the slope of the data (Figure 4a), to correct each point by subtracting the measured Fe/Ca of the sample multiplied by 1.13. Because there is a small positive intercept of 30 μmol/mol in the regression of Fe/Ca versus Al/Ca (Figure 3), we only applied the correction to samples with Fe/Ca exceeding that value, and we subtracted 30 μmol/mol from all Fe/Ca values to which the correction was applied. About ∼0.5% of the data had high Fe values data with relatively low Al, suggesting the Fe in those samples was unrelated to detrital contamination (due to, for example, nonsilicate contamination such as pyrite). For this reason, we also only apply the correction if the Al/Fe ratio for that analysis exceeds 0.7, which is the minimum Al/Fe molar ratio in Galapagos volcanic rocks (B. M. Gunn, Geochemistry of igneous rocks, 2005, available at The average Al/Fe ratio of all of the analyses in TR163-22 is 1.8, equivalent to the slope of the Fe/Ca to Al/Ca regression (Figure 3).

[27] The two correction schemes produce very similar results and are essentially equally effective. The second scheme, which relies on both Fe/Ca and Al/Ca, appears to be slightly more effective in TR163-22, as judged by the improvement in the reproducibility of sample splits (see below). There is also less uncertainty in the correction using Fe/Ca because the slopes in the two intervals were so similar (Figure 4a). The slightly greater effectiveness of the Fe/Ca-based correction scheme might also be related to the threefold greater sensitivity of Fe versus Al determination by sector ICP-MS. In sediments in which Fe/Ca is present on foraminifer shells due to other processes, such as pyrite formation, the use of Al/Ca alone as a correction for Al-silicate might yield more reliable results.

[28] The corrected Mg/Ca data (Figure 7) is not dramatically different than the uncorrected data, and as expected the biggest changes occur in the intervals with markedly enhanced Fe and Al. A t-test (paired two sample for means) indicates that the corrected Mg/Ca data is significantly (P ≪ 0.001, df = 648) lower (0.07 mmol/mol) than the uncorrected data. The maximum correction is for an analysis from 589 cm, for which the correction is −0.38 mmol/mol relative to a measured value of 2.40 mmol/mol. The estimated error of the Mg/Ca correction itself is ∼±20%, dominated by the uncertainty in the Mg:Fe ratio of the shards, as determined from the 95% confidence interval of the slope (i.e., Figure 4a). Thus the estimated error for samples with large corrections is <±0.1 mmol/mol.

Figure 7.

Comparison of Mg/Ca in G. ruber shell digests from TR163-22 shell digests as measured and as calculated after correction using the Fe/Ca and Al/Ca as an indicator of extraneous volcanic contribution. A ratio of 1.13 mole/mole of Mg/Ca was subtracted for each mole/mole of Fe/Ca present. Only samples with Fe/Ca > 0.03 mmol/mol and Al/Fe > 0.7 were corrected. Although the basic patterns of the data are very similar, the corrected data more clearly reveal the climatological patterns in the data.

[29] A comparison of pooled standard deviations of replicate analyses (cleaned and analyzed separately) for uncorrected and corrected Mg/Ca data suggests that the correction scheme is effective: the pooled SD decreased to ±0.08 mmol/mol (3.5%) for the corrected data versus ±0.11 mmol/mol (4.5%) for the uncorrected data. This comparison provides the strongest evidence that the correction scheme improves the data, because it rests only on how well the correction scheme accounts for differences between samples prepared and analyzed separately from the same intervals. Additional support for the efficacy of the Fe-based correction scheme comes from the significantly improved correlation of Mg/Ca with G. ruber δ18O, which share a common surface temperature signal (Lea et al., submitted manuscript, 2005).

5.3. Calculation of SST From Measured and (Al, Fe)-Corrected Mg/Ca

[30] We use the water-depth corrected equation for G. ruber presented by Dekens et al. [2002], which was calibrated using mean annual SST from Levitus and Boyer [1994], to calculate SST:

equation image

The core-top Mg/Ca for G. ruber in TR163-22, 2.5 ± 0.1 mmol/mol (average of top 10 cm, 1 SD, n = 8), yields an SST of 24.3 ± 0.4°C using equation (1). The water depth correction term (0.61*2.83 = 1.7°C) and basin correction term (1.6°C) contribute equally to compensate for a ∼3.3°C offset due to dissolution loss of Mg. Modern mean annual SST at this site is 24.5°C [Levitus and Boyer, 1994] and 24.3°C [Conkright et al., 2002], so the calculated SST from Mg/Ca using equation (1) is in good agreement with modern observations. The simpler G. ruber calibration developed for equatorial Pacific cores between 1600–3200 m [Lea et al., 2000] yields an SST of 23.6°C, 0.5–1°C lower than modern SST. It should be noted that although the use of the water-depth corrected equation [Dekens et al., 2002] improves the agreement of the core top calculated SST with modern observations, the equation will not compensate for changes in carbonate preservation downcore which might have affected Mg/Ca.

[31] A comparison of downcore SSTs based on uncorrected and corrected Mg/Ca data indicates that the offsets are relatively subtle, although clearly apparent in the intervals of elevated Al and Fe (Figure 8). The largest SST correction is −2.3°C (589 cm), but the average correction for all of the data that includes Fe/Ca analyses is only −0.3°C. The estimated error of the correction (section 5.2) translates to an uncertainty of ±0.4°C for the largest correction, but for the average correction the uncertainty is less than 0.1°C.

Figure 8.

Comparison of SST, calculated using the dissolution-corrected equation of Dekens et al. [2002], based on measured G. ruber Mg/Ca and Fe, Al-corrected Mg/Ca (see Figure 7) in core TR163-22. The correction removes spurious peaks in SST unrelated to climatological variations, but it does not alter the basic patterns of the data.

[32] As expected, the correction makes the biggest change in the Al and Fe (volcanic shard) enriched intervals of the core (i.e., 220–282 and 518–618 cm). In these intervals the corrected SST data reveals trends (cycles) that are obscured in the uncorrected data. It is important to note, however, that the overall patterns of the data, with low Mg/Ca values between 165–565 cm (MIS 2, 3 and 4) and higher values between 565 cm and the base of the core (MIS 5) are similar in both the uncorrected and corrected data. In fact, interpretation of the SST data with and without the correction for Al-silicate contamination would lead to similar broad-scale paleoclimatological inferences.

5.4. Interpretation of Downcore Trends in U and Mn in Core TR163-22

[33] The distinct downcore trends in foraminiferal U/Ca and Mn/Ca indicate that variability in these elements is not related to Al-silicate contamination (Figure 2). In addition, the levels of Mn/Ca consistently exceed 100 μmol/mol below 100 cm, and levels of U/Ca exceed 10 nmol/mol throughout almost the entire core length. Mn/Ca and U/Ca in foraminifera recovered from the water column are very low, typically ∼5 μmol/mol for Mn/Ca [Pak et al., 2004] and ∼10 nmol/mol for U/Ca [Russell et al., 1994, 2004; Pak et al., 2004]. Concentrations exceeding these levels are thought to largely reflect postdepositional addition of Mn and U via diagenetic coatings on the surface of the shells [Boyle, 1983; Russell et al., 1996]. The data from core TR163-22 suggests Mn has been added to foraminifera shells, presumably in the form of Mn-carbonate, and that U is also precipitated on the shells in association with Mn.

[34] The Mg/Ca and Mn/Ca data show no correlation throughout the core length, although there are restricted intervals, chiefly between ∼200 and 600 cm, in which there are weak positive correlations between Mg/Ca and Mn/Ca with r2 values of up to ∼0.09 (P < 0.0002). The slope of this correlation is 0.89 ± 0.45 (95% CI) mol/mol, which is much higher than the inferred ∼0.1 mol/mol Mg:Mn ratio of Mn carbonates [Pedersen and Price, 1982; Pena et al., 2005]. For this reason the weak correlations in restricted intervals of the core are unlikely to reflect a primary association between Mg and Mn via Mn-carbonate precipitation. On the basis of data from this core and many others, we think that the use of a single “cut-off” value for Mn/Ca (i.e., 100 μmol/mol) as a threshold for acceptable trace element data is inadvisable because the composition, and therefore the impact, of Mn-coatings appears to vary between sites.

[35] What caused the observed downcore variability in Mn/Ca and U/Ca in core TR163-22? The rise in Mn/Ca that occurs in the top ∼200 cm is typical of cores from suboxic cores and presumably represents the gradual addition of Mn-carbonate as the foraminifera shells become buried and exposed to deeper, more reducing pore waters enriched in both Mn2+ and carbonate ion [Boyle, 1983]. A similar process might operate for U, although the reduced form U4+, is actually less soluble [McManus et al., 2005]. We do not have U/Ca data for the absolute top of the core, but there is much less of a clear secular signal in the U/Ca data available from the top ∼200 cm of the core, where the secular trend in Mn/Ca is clearly apparent. This difference might reflect the contrasting solubility behavior of the two ions.

[36] Both U/Ca and Mn/Ca are characterized by weak cyclicity of ∼20–200 cm, equivalent to ∼3–30 kyr. These cycles are most apparent in the U/Ca data. In the interval between 500–800 cm, these cycles are also apparent in rare earth element (La, Nd, Ce) data for the foraminifera shells (not shown). These cycles are not clearly apparent either in the Mg/Ca (SST) data or in planktonic and benthic oxygen isotope data (Lea et al., submitted manuscript, 2005). Hypotheses to explain the U and Mn cycles would include shifts in diagenetic (oxygenation) conditions on millennial-to-precessional timescales in response to changes in bottom water oxygenation or changes in the productivity of overlying surface waters that alter the delivery of reduced organic carbon to this site. Another possibility is that specific changes in diagenetic conditions promote the formation of Mn-rich coatings on foraminifera shells. Varying delivery of Mn from hydrothermal sites on the Galapagos Spreading Center could also play a role (Figure 1). These hypotheses could be tested by comparing bulk productivity and diagenesis indicators from core TR163-22 to the trends in shell chemistry. Another useful test would be to compare benthic shell chemistry with the planktonic trends, because the benthics would have been further subjected to the influence of changing pore water chemistry on their primary shell composition.

6. Conclusions

[37] Downcore determinations of Mg, Fe, Al, Mn and U to Ca ratios in purified shell digests of the surface dwelling planktonic foraminifera G. ruber from piston core TR163-22, lying just NW of the Galapagos Island, reveal that these different elements fall into three broad groups of behavior, with some overlapping controls between them. The trends in Mg/Ca are dominated by climatological changes, with lower values during cold intervals (i.e., MIS 2–4) and higher values during warm intervals (i.e., MIS 1, 5). The trends in Al/Ca and Fe/Ca are dominated by Al-silicate inputs in discrete, volcanic shard-rich intervals of the core. These shards leave their chemical mark on foraminifera shells because up to 0.1% by weight of the volcanic material is present in the shells even after intensive cleaning. Downcore trends in Mn/Ca and U/Ca appear to be dominated by common diagenetic processes associated with redox changes and the addition of Mn-carbonate coatings to the foraminifera shells. Shell Mn/Ca has a strong secular trend of increasing concentration with depth over the top 200 cm of the core, whereas U/Ca varies with little secular trend but with clearly defined cycles of ∼20–200 cm length.

[38] Because there is Mg in Galapagos volcanic rocks, the presence of volcanic debris in the shell digests can contribute up to 0.4 ± 0.1 mmol/mol Mg/Ca to the measured value. Changes to the cleaning procedure, such as additional ultrasonication steps or removal of dark grains [Barker et al., 2003] does not markedly reduce the volcanic contribution. We therefore propose correcting the measured Mg/Ca using the associated Fe/Ca and Al/Ca and the slope of the Mg:Fe relationship, as determined from the most Al-silicate rich intervals. This correction lowers the SST values calculated from Mg/Ca by up to 2°C in the most Al-silicate-rich intervals of the core, but more typically the correction is ∼−0.3°C. The reproducibility of Mg/Ca sample splits improves from ±4.5% to ±3.5% when the correction is applied on an individual analysis basis, suggesting that the correction procedure is effective in reducing the volcanic Mg contribution. Millennial scale trends in the SST data are also more clearly revealed in the corrected data.


[39] We thank J. Horton, A. Jienpradit (now Engen), R. Polcyn, C. Belanger, H. Berg, T. Herbert, P. Martin, and H. Spero for laboratory help, unpublished data, and comments. Reviews from I. Cacho and an anonymous reviewer and editorial advice from P. Martin and L. Labeyrie improved the paper. This research was supported by U.S. NSF grants OCE0117886 and OCE0317611.