Planktonic foraminiferal area density as a proxy for carbonate ion concentration: A calibration study using the Cariaco Basin ocean time series

Authors


Abstract

[1] Biweekly sediment trap samples and concurrent hydrographic measurements collected between March 2005 and October 2008 from the Cariaco Basin, Venezuela, are used to assess the relationship between [CO32−] and the area densities (ρA) of two species of planktonic foraminifera (Globigerinoides ruber (pink) and Globigerinoides sacculifer). Calcification temperatures were calculated for each sample using species-appropriate oxygen isotope (δ18O) temperature equations that were then compared to monthly temperature profiles taken at the study site in order to determine calcification depth. Ambient [CO32−] was determined for these calcification depths using alkalinity, pH, temperature, salinity, and nutrient concentration measurements taken during monthly hydrographic cruises. The ρA, which is representative of calcification efficiency, is determined by dividing individual foraminiferal shell weights (±0.43 µg) by their associated silhouette areas and taking the sample average. The results of this study show a strong correlation between ρA and ambient [CO32−] for both G. ruber and G. sacculifer (R2 = 0.89 and 0.86, respectively), confirming that [CO32−] has a pronounced effect on the calcification of these species. Though the ρA for both species reveal a highly significant (p < 0.001) relationship with ambient [CO32−], linear regression reveals that the extent to which [CO32−] influences foraminiferal calcification is species specific. Hierarchical regression analyses indicate that other environmental parameters (temperature and [PO43−]) do not confound the use of G. ruber and G. sacculifer ρA as a predictor for [CO32−]. This study suggests that G. ruber and G. sacculifer ρA can be used as reliable proxies for past surface ocean [CO32−].

1 Introduction

[2] Changes in the carbon dioxide (pCO2aq) concentration of the oceans alter surface ocean pH and in turn seawater carbonate ion concentrations [CO32−]. Over the last two centuries, anthropogenic input of carbon into the atmosphere and oceans has resulted in an unprecedented rapid decline in surface ocean pH, a process referred to as ocean acidification (OA) [Caldiera and Wickett, 2003; Honisch et al., 2012; Zeebe, 2012]. Though the uptake of ~30% of the anthropogenic CO2 by the ocean has mitigated modern pCO2atm rise [Sabine et al., 2004], the resulting decline in calcite and aragonite saturation states has had an adverse effect on several key marine organisms and ecosystems [Riebesell et al., 2000; Caldiera and Wickett, 2003; Orr et al., 2005; Hoegh-Guldberg et al., 2007, Moy et al., 2009]. With some exceptions [Iglesias-Rodriguez et al., 2008; Ries et al., 2009], marine calcifiers—including several species of planktonic foraminifera—have exhibited reduced rates of calcification when grown in low pH and low [CO32−] waters (Table 1) [Spero et al., 1997; Bijma et al., 1999, 2002; Wolf-Gladrow et al., 1999; Riebesell et al., 2000; Russell et al., 2004; Lombard et al., 2010; Manno et al., 2012]. In addition to having an adverse effect on the calcification efficiency of these organisms, a reduction in calcification rates of marine plankton will likely have a significant impact on the marine carbon cycle, as calcification is a process that increases aqueous CO2 [Wolf-Gladrow et al., 1999; Zeebe and Wolf-Gladrow, 2001]. Thus, on short time scales, a decrease in calcification will result in an increase in the ocean's ability to take up atmospheric CO2 [Feely et al., 2004], but would ultimately reduce the total flux of calcite and organic carbon to the deep ocean over longer time scales, which serves as a significant long-term carbon sink [Armstrong et al., 2001; Klaas and Archer, 2002; Ridgwell and Zeebe, 2005; Zeebe, 2012].

Table 1. Compilation of Previous Studies Assessing the Relationship Between Planktonic Foraminiferal Calculation and [CO32−]
Species/StudySample Type

Size Fraction

(µm)

Size-Normalizing Method

Range in [CO32−]

(µm/kg)

Correlation With [CO32−]%Δ[CO32−]200–300
  1. a

    This study uses data from Russell et al. [2004] and unpublished data from R. da Rocha, A. Kuroyanagi, G. J. Reichart, and J. Bijma.

  2. b

    Ambient [CO32−].

  3. c

    Sea surface [CO32−].

  4. d

    Percent change in mass of O. universa at high [CO32−] (over 600 µmol/kg) relative to the mass at ambient [CO32−].

  5. e

    Percent change in SBW from preindustrial (core top) G. bulloides to present (sediment trap) G. bulloides.

  6. f

    Percent change represents the difference between the SBW from thin (younger) and thick (older) shelled populations with 14C-derived age differences of 35–140 years.

  7. g

    Percent change in SBW from 25,000 years B.P. to 1000 years B.P.

 
Orbulina universa
Spero et al. [1997]culture75–774 (699)positive37d
Bijma et al. [1999]culture500–600 (100)SBW40–150 (110)bpositive, R2 = 0.5524.3
 culture500–600 (100)SBW180–640 (460)bpositive, R2 = 0.6712.8
Russell et al. [2004]culture427–653 (226)shell thickness76–468 (392)bpositive, R2 = 0.9715.1
 culture427–667 (240)SBW77–480 (404)bpositive, R2 = 0.6518.3
Lombard et al. [2010]aculture427–667 (240)calcification rate77–480 (404)bpositive, R2 = 0.047.7–14.7
 
Globigerina bulloides
Barker and Elderfield [2002]core top300–355 (55)MBW206–257 (51)cpositive, R2 = 0.67155.5
 core top300–355 (55)SBW199–264 (65)cpositive, R2 = 0.3163.9
Gonzalez-Mora et al. [2008]core250–300 (50)SBW150–250 (100)cpositive
Moy et al. [2009]sediment trap/core top300–355 (55)SBW153–189 (36)cpositive35e
 sediment trap/core top355–425 (70)SBW153–189 (36)cpositive30e
Beer et al. [2010a]plankton net200–250 (50)MBW166–276 (110)cpositive, R2 = 0.3816.6
Aldridge et al. [2012]plankton net150–200 (50)MBW148–181 (40)cpositive, R2 = 0.3562.6
 plankton net200–250 (50)MBW148–181 (40)cpositive, R2 = 0.5647.5
 
Pulleniatina obliquiloculata
Naik and Naidu [2007]core top350–420 (70)MBWpositive
Mekik and Raterink [2008]core top420–520 (100)MBW145–225 (80)bpositive, R2 = 0.67200.0
 
Neogloboquadrina dutertrei
Naik and Naidu [2007]core top350–420 (70)MBWpositive
Mekik and Raterink [2008]core top355–415 (60)MBW120–225 (105)bpositive, R2 = 0.6476.8
 
Globorotalia truncatulinoides
Barker and Elderfield [2002]core300–355(55)SBW216–264 (48)cpositive, R2 = 0.4438.2
 
Neogloboquadrina pachyderma
Barker and Elderfield [2002]core250–300 (50)MBW199–264 (65)positive, R2 = 0.6553.3
Gonzalez-Mora et al. [2008]core250–300 (50)SBWno response
Manno et al. [2012]culture100–200 (100)calcification rate60–120 (60)cpositive21–30
 
Globorotalia inflata
Barker and Elderfield [2002]core300–355(55)SBW200–268 (68)cpositive, R2 = 0.7770.7
Globogerinoides ruber
Gonzalez-Mora et al. [2008]core250–300 (50)SBWpositive
de Moel et al. [2009]core250–300 (50)SBW(6.5; 18)positive25.0f
 core300–355 (55)SBW(6.5; 18)positive25.0f
 core250–500 (250)shell thickness(6.5; 18)positive35.0f
Beer et al. [2010a]plankton net200–250 (50)MBW251–284 (33)cnegative, R2 = 0.78−88.6
 plankton net250–315 (65)MBW261–285 (24)cnegative, R2 = 0.64−74.2
 plankton net315–355 (40)MBW262–284 (22)cnegative, R2 = 0.50−71.5
This study (pink)sediment trap355–650 (295)ρA215–270 (55)bpositive, R2 = 0.8944.1
 sediment trap355–500 (145)ρA215–270 (55)bpositive, R2 = 0.7349.6
 sediment trap500–650 (150)ρA215–270 (55)bpositive, R2 = 0.8243.8
 
Globogerinoides sacculifer
Bijma et al. [2002]culture493–575 (82)SBW100–620 (520)bpositive, R2 = 0.397.7
 culture582–663 (81)SBW100–620 (520)bpositive, R2 = 0.224.6
 culture762–845 (83)SBW100–620 (520)bpositive, R2 = 0.285.4
Naik and Naidu [2007]core top350–420 (70)MBW240–250 (10)bpositive157.7
Lombard et al. [2010]aculture372–446 (74)calcification rate72–566 (494)bpositive, R2 = 0.03–0.076.3–8.1
Naik et al. [2010]core503–699 (196)SBW61–106 (36)cpositive32.3g
This study (no sac)sediment trap425–800 (350)ρA165–240 (70)bpositive, R2 = 0.8627.1
 sediment trap425–650 (225)ρA165–240 (70)bpositive, R2 = 0.7929.1
 sediment trap650–850 (200)ρA165–240 (70)bpositive, R2 = 0.5519.5

[3] Since planktonic foraminifera are responsible for up to 80% of the total calcium carbonate (CaCO3) accumulated in surface sediments [Schiebel, 2002], it is important to understand the extent to which their calcification will be affected by decreasing [CO32−] associated with OA. The quantification of the relationship between [CO32−] and planktonic foraminiferal calcification provides a useful proxy for determining past changes in ocean carbonate chemistry, as well as estimates of past pCO2atm. It has been shown that a positive linear relationship exists between planktonic foraminiferal size-normalized shell weight (SNW) and ambient [CO32−] [Spero et al., 1997; Bijma et al., 1999; Barker and Elderfield, 2002]. Rates of calcification increase in conjunction with increasing seawater [CO32−], resulting in a thickening of the shell wall and an increase in mean shell weight [Spero et al., 1997; Bijma et al., 1999; Russell et al., 2004].

[4] Culture studies have shown a decline in calcification efficiency with decreasing [CO32−] for a wide variety of foraminiferal species commonly used in paleoclimatic and paleoceanographic reconstructions (Table 1). Most studies of sediment core and water column material (plankton tow and sediment trap) report a stronger influence of [CO32−] on foraminiferal calcification than what has been reported in culture studies (Table 1). It has been suggested that the shallower slopes reported in culture studies could have been a result of the foraminifera not having completed their entire life cycle in culture conditions [Bijma et al., 2002].

[5] Additional differences amongst studies investigating the influence of [CO32−] on foraminiferal calcification arise from the variety of methodologies used to estimate changes in calcification. While some studies estimate or measure calcification rate or shell thickness, most studies use SNW to estimate a change in shell thickness or density in response to changes in [CO32−] (Table 1). In order to isolate the contribution of shell thickness to weight measurements, the influence of size on the overall weight of the foraminiferal test must be taken into account. Prior studies have examined the relationship between [CO32−] and shell weight using two general methods for size normalization: sieve-based weight (SBW) [Broecker and Clark, 2001; Naik et al., 2010; de Villiers, 2004] and measurement-based weight (MBW) [Barker and Elderfield, 2002; Beer et al., 2010a; Aldridge et al., 2012]. The SBW is the simpler of the two methods, in which the mean bulk weights are determined from traditionally used narrow size fractions. The use of MBW, which is a more effective method for reducing the influence of size on weight measurements [Beer et al., 2010b], normalizes mean bulk weights taken from narrow size fractions using equation (1) below, where parameter refers to either silhouette area or diameter:

display math(1)

[6] Most SNW studies, regardless of the normalization method used, reveal a positive linear relationship between foraminiferal shell weights and [CO32−], although there are significant inter and intraspecies differences (Table 1). However, a number of studies have found contradictory results, reporting either a negative (Globigerinoides ruber (white)) [Beer et al., 2010a] or no relationship (Neogloboquadrina pachyderma) [de Villiers, 2004; Gonzalez-Mora et al., 2008] between shell weight and [CO32−]. In these cases, it was suggested that other environmental parameters, or more generally optimal growth conditions [de Villiers, 2004], govern calcification efficiency for these species.

[7] A temperature effect on foraminiferal calcification has been reported in a number of studies [Bé et al., 1973; Hecht, 1976; Hemleben et al., 1987; Schmidt et al., 2004; Lombard et al., 2009]. These studies indicate that the size of foraminiferal tests varies with temperature, giving an additional reason for size normalization when using shell weight as a proxy for [CO32−]. Likewise, several studies have reported a potential relationship between SNW and temperature [Barker and Elderfield, 2002; Beer et al., 2010a; Aldridge et al., 2012]. However, this observed relationship between SNW and temperature can also be explained by a [CO32−] effect as the two parameters covary in surface waters. Barker and Elderfield [2002] evaluated the influence of temperature on Globigerina bulloides SNW by comparing shell weights across the most recent glacial-interglacial transition. They found that average shell weights were higher during the last glacial period when SST was low and [CO32−] was high, concluding that [CO32−], not temperature, was the dominant factor influencing calcification rates. A recent culture study using N. pachyderma (sinistral) specimens showed that the calcification rates of both juvenile and adult specimens decreased by 30% and 21%, respectively, when grown in low [CO32−] waters, but were unaffected by an increase in ambient temperature while keeping [CO32−] constant [Manno et al., 2012].

[8] Other authors have suggested that nutrient concentrations ([PO43−] and [NO3−]) may affect foraminiferal calcification efficiency—either by enhancing calcification [Bijma et al., 1992; Barker and Elderfield, 2002] or by hindering it [Aldridge et al., 2012]. For example, based on a North Atlantic plankton tow study, Aldridge et al. [2012] found that the MBW of G. bulloides had a strong negative correlation with [PO43−]. However, Aldridge et al. [2012] do not consider the strong collinearity that exists between [PO43−] and [CO32−] (R2 = −0.85). Simply put, when [PO43−] is increasing in surface waters, [CO32−] is decreasing, making it difficult to determine which environmental parameter best explains the observed variability in G. bulloides MBW.

[9] The objective of the current study is to better quantify the relationship between foraminiferal calcification and [CO32−] by utilizing a more precise method of eliminating the contribution of shell size to shell weight through area density (ρA; µg/µm2) calculations. The method for deriving ρA presented in this study uses the weight and silhouette area of individual shells of Globigerinoides ruber (355–650 µm) and Globigerinoides sacculifer (425–850 µm), allowing for the use of a broad size fraction while investigating the relationship between calcification efficiency and ambient [CO32−]. The Cariaco Basin, Venezuela, is an ideal study area to investigate the relationship between seawater [CO32−] and planktonic foraminiferal ρA as this region is characterized by the seasonal upwelling of low pH, low temperature, and low [CO32−] waters. In this study, we use linear regression modeling to investigate the relationships amongst foraminiferal ρA, ambient [CO32−], temperature, and [PO43−] at the depth of calcification.

2 The Cariaco Basin

2.1 Regional Setting

[10] The Cariaco Basin is located on the continental shelf of northern Venezuela and is divided into two subbasins by a 900 m saddle (Figure 1). Climatological conditions in the basin are controlled by the seasonal migration of the Intertropical Convergence Zone (ITCZ) and the associated latitudinal position of the easterly trade winds. The ITCZ is in its most southerly position during the boreal winter and early spring (November–May). During this time, the easterlies are positioned over the basin and generate Ekman-induced upwelling, which results in minimum sea surface temperatures (~22°C), maximum salinity (>36.8), elevated nutrient concentrations, and high primary production [Thunell et al., 2000; Muller-Karger et al., 2001; Goñi et al., 2003]. During the summer and early fall (August–October), the ITCZ migrates to its most northerly position; trade winds decrease over the basin and upwelling ceases, allowing sea surface temperatures to reach their maximum (~28–29°C), while nutrient concentrations and primary production are mutually diminished. The northerly position of the ITCZ over the Cariaco Basin at this time also increases precipitation, resulting in lower salinities in the surface waters (<36.6).

Figure 1.

Bathymetric map of the Cariaco Basin showing the location of the sediment trap mooring (10°30′N and 65°31′W).

[11] The Carbon Retention in a Colored Ocean Project (CARIACO) oceanographic time series began in November 1995 with the goal of providing a link between surface processes and the sediment record [Muller-Karger et al., 2000, 2001; Thunell et al., 2000; Goñi et al., 2003]. A bottom-tethered mooring (10°30′N and 65°31′W) with automated sediment traps at five depths (150, 230, 410, 800, and 1200 m) continuously measures the flux of settling particles and provides biweekly samples that can be examined and compared to monthly hydrographic data. The samples used in this study are from the upper three sediment traps. The planktonic foraminifera collected from these samples display excellent preservation, with specimens frequently having intact spines. Since its inception, the program has collected a wide range of hydrographic data at discrete depths throughout the water column (0–1300 m) on a monthly basis. All hydrographic data for the Cariaco Time Series are archived at http://www.imars.usf.edu/CAR.

2.2 The Carbonate System in the Cariaco Basin

[12] The ocean carbonate system can be quantitatively defined by the following six parameters: total dissolved inorganic carbon, total alkalinity (AT), pH, [CO32−], total CO2 in seawater ([CO2] = [CO2(aq)] + H2CO3), and bicarbonate ([HCO3−]). One can use the combination of any two of these parameters, in combination with temperature, salinity, pressure, and nutrient concentrations, to calculate the entire carbonate system (see Zeebe and Wolf-Gladrow [2001] and Zeebe [2012] for a review of the carbonate system). For our study, pH and total alkalinity are the only two parameters directly measured during the monthly hydrographic cruises. The carbonate system in the Cariaco Basin is influenced by a number of water column biogeochemical processes including primary production and respiration, CaCO3 precipitation and dissolution, and the remineralization and consumption of organic matter [Astor et al., 2005]. Additionally, physical factors such as seasonal upwelling, changes in evaporation and precipitation ratios, advection of Caribbean waters into the basin, air-sea gas exchange, and riverine input also impact the carbonate system in the basin. These processes collectively yield surface water (1 m depth) pH values that range from 8.03 to 8.11 during the study period (March 2005 to September 2008; http://www.imars.usf.edu/CAR). The oxidation of organic matter in the basin, coupled with increased CO2 solubility associated with decreasing water temperatures, causes a rapid decline in pH values with increasing depth. These processes, in combination with a lower average alkalinity at intermediate depths (2407 µmol/kg at 100 m versus 2418 µmol/kg at the surface), yield an average pH value of 7.91 at 100 m depth throughout the course of the study period.

3 Materials and Methods

3.1 Foraminiferal Collection

[13] Biweekly sediment trap samples were collected between May 2005 and September 2008 in cups containing a buffered formalin solution, ensuring good preservation of the foraminiferal tests. Shells of planktonic foraminiferal species G. ruber and G. sacculifer were separated from the sediment trap samples using the settling method described by [1959]. The shells were washed, wet sieved (>125 µm), and examined under a stereo binocular microscope. After washing, microscopic observation of the foraminiferal tests revealed clean surfaces, free of surficial organic matter (OM). All G. ruber (pink) and G. sacculifer (sac-less) individuals were wet picked and allowed to dry (>1 week) prior to weighing in an environmentally controlled weighing room. Following a 45 min oxidative treatment (30% H2O2 with 0.1 M NH4OH) on a select number of samples (n = 4), it was determined that differences between the pretreatment and posttreatment shell weights were within the analytical error associated with the weight measurements (±0.43 µg, repeat weighing of individual Orbulina universa; n = 60). This indicates that the settling and washing techniques were efficient in removing surficial OM and that oxidative cleaning is unnecessary for the samples used in this study. The microscopic imaging program Macnification 2.0 (Orbicule, Mac OS X Leopard) was used to sort G. ruber (355–650 µm) and G. sacculifer (425–850 µm) into their corresponding size fractions based on Feret's diameter (the longest distance between two points on the test).

3.2 Area Density Analysis

[14] Individual foraminiferal shells from each sample population were weighed using a Metler Toledo microbalance and photographed with an inverted light microscope for size analysis. Macnification 2.0 uses an RGB filter to determine individual foraminiferal 2-D (silhouette) areas. Calibration for the silhouette area and Feret's diameter measurements was performed using a microscale image taken at the same magnification as the foraminiferal images (50×). Foraminiferal shells are positioned to capture the maximum silhouette area of each individual, corresponding to the umbilical or spiral sides for both G. ruber and G. sacculifer. The difference in average areas for the spiral and umbilical orientations was determined to be negligible based on analyses performed on sample populations of G. ruber (n = 15, area difference between orientations is 0.15%) and G. sacculifer (n = 12, area difference between orientations is 0.40%). The ρA (µg/µm2) is determined by dividing individual weights by their corresponding silhouette area and taking the sample average (n > 10, mean = 18).

3.3 Temperature Calculations and Calcification Depths

[15] Randomly selected individuals (n = 4 to 8) from each sample population were analyzed for oxygen isotope composition to determine calcification temperature. Oxygen isotope analyses were performed on a GV IsoPrime stable isotope ratio mass spectrometer (long-term standard reproducibility is ±0.07‰) and are reported relative to Vienna Pee Dee Belemnite (V-PDB). Calcification temperatures for each sample were determined using the following species-appropriate δ18O temperature equations:for G. ruber [Bemis et al., 1998],

display math(2)

and for G. sacculifer [Mulitza et al., 2003],

display math(3)

where δc is the δ18O of the foraminiferal calcite, and δw is the δ18O of the calcifying waters. Time-equivalent δ18Ow estimates were established using the δ18Ow salinity equations from McConnell et al. [2009] for the Cariaco Basin for both upwelling (equation (4)) and nonupwelling conditions (equation (5)):

display math(4)
display math(5)

[16] The δ18Ow values are scaled from SMOW to PDB by subtracting 0.27‰ [Bemis et al., 1998]. The δ18O-derived calcification temperatures were then compared to the measured water column temperatures to determine calcification depths and the associated instrumental temperature, salinity, nutrient, pH, and alkalinity needed for calculating [CO32−]. It should be noted that a [CO32−]-δc relationship has been observed in culture studies [Spero et al., 1997]. To our knowledge, no calibration of this relationship exists for either G. ruber or G. sacculifer making it difficult to model this effect on the samples used in this study. Using the Δδ18O-[CO32−] model presented in King and Howard [2005], where Δδ18O is the difference between the measured δc from the foraminiferal samples and the predicted δc based on the instrumental temperatures, we find that there is no correlation between [CO32−] and Δδ18O for the sediment trap samples used in this study, suggesting that [CO32−] is not a controlling factor for this offset.

3.4 Carbonate Parameter Calculations

[17] Monthly records of aqueous [CO32−] were generated for the study site using CO2SYS.xls [Pelletier et al. 2007] (version 16) and the constants of Lueker et al. [2000] and Dickson [1990]. [CO32−] values were calculated for the upper 130 m at discrete depth intervals (1, 7, 15, 25, 35, 55, 75, 100, and 130 m) using AT, pH, temperature, salinity, and nutrient concentration measurements taken during monthly hydrographic cruises, and accounting for the depth (i.e., pressure) of collection. The measurement error for [CO32−], calculated from the errors associated with each carbonate parameter used for its calculation, is less than ±1.3 µmol/kg for all the samples used in this study. A comprehensive description of the methodologies used to collect monthly hydrographic data in the Cariaco Basin can be found at http://www.imars.usf.edu/CAR.

[18] Assuming an average 3 week life span for both G. ruber and G. sacculifer [Bijma et al., 1990] and a 1 day settling period to reach the trap depths of 150 and 410 m (sinking speed = 300 m/day) [Takahashi and Bé, 1984], the foraminifera collected in the biweekly sediment traps calcified in waters 8 to 22 days prior to the time the trap opened for collection. Thus, in all possible cases, we used hydrographic data that fell close to or within this range of day difference to pair with the average foraminiferal ρA.

3.5 Regression Analyses

[19] Simple, multiple, and hierarchical regression analyses (IBM SPSS) [Miles and Shevlin, 2001] were used to quantify the relationships between the response variable (foraminiferal ρA) and the predictor variables ([CO32−], temperature, and [PO43−]). Simple linear regression analysis (SLR) was used to determine the bivariate relationship between individual response and predictor variables. Various types of multiple linear regression analyses (MLR) were performed to examine the relationships between ρA and the predictor variables, as well as to examine the covariance amongst the predictor variables themselves. Both SLR and MLR can yield unreliable statistical outcomes for cases of multiple covarying predictor variables—a condition called collinearity or multicollinearity. Collinearity can be an issue in upwelling systems such as the Cariaco Basin as it is difficult to decouple covarying environmental variables and determine the actual amount each variable contributes to changes in the response variable (e.g., ρA). To examine the collinearity amongst the three predictor variables, two types of MLR were performed. The first used each predictor variable in turn as the dependent variable and the other two predictor variables as the dependent variables (Table S2 in the supporting information). The coefficient of determination (R2) resulting from these multivariate regression analyses is indicative of the amount of variance shared by the dependent variable (one of the predictor variables) with the independent variables (the other two predictor variables), essentially quantifying the redundancy one predictor variable shares with the other predictor variables. Two other collinearity diagnostics, tolerance (1 − R2) and variance inflation factor (VIF; (1 − R2)−1), were determined using MLR with ρA as the dependent variable and [CO32−], temperature, and [PO43−] as the independent variables (Table S3). Collinearity in two or more predictor variables will inflate the variance and standard errors associated with a regression analysis; thus, a strong R2 is the result of redundant predictor variables as opposed to a set of good independent predictor variables. In general, tolerance values below 0.50 and VIF values above 2 are indicative of an issue with collinearity amongst the independent variables [Miles and Shevlin, 2001].

[20] One way the current study addresses the issue of collinearity is by leaving the values for one predictor variable (X1) unchanged, but removing its covariance with the other two predictor variables (X2, X3) by regressing them on X1 and generating their residuals. For example, the residuals of temperature and [PO43−] (TCres, [CO32−] and [PO43−]res, [CO32−]) were quantified using equations (11) and (12) for G. ruber and (17) and (18) for G. sacculifer from Table 2 in order to determine the predicted values for calcification temperature and phosphate concentrations based on their relationship with [CO32−]. The residuals were then calculated by subtracting the predicted values from the measured values. The residuals represent the variability in temperature and phosphate that is unrelated to their covariance with [CO32−]. By using the residuals as opposed to the original calcification temperature and phosphate concentrations, we are able to estimate what additional influence these parameters have on G. ruber and G. sacculifer ρA once [CO32−] has been considered. Hierarchical multiple regression analysis (HMR) was used to determine the relative predictive capabilities of each variable for G. ruber and G. sacculifer ρA by assessing the change in R2R2) and the significance of this change (p ΔR2) as each predictor variable is added sequentially to the regression model. In addition to the R2, ΔR2, and p ΔR2, the beta or standardized coefficient (β) is also reported as this is indicative of the percentage of a standard deviation (SD) that the response variable (ρA) would change for a 1 SD change in the predictor variable (Tables S4 and 3). Each model assumes that X1 is the dominant predictor variable and assesses the relative contributions of X2 and X3 while holding all previously added variable(s) constant (Table 3). When using HMR, each predictor variable is added to the regression equation in an order specified by the researcher based on prior observations or an established theory. For the first model, we use the results of previous studies [Barker and Elderfield, 2002; Naik et al., 2010; Manno et al., 2012] to establish the order of the variables, using [CO32−] as X1 and TCres, [CO32−] and [PO43−]res, [CO32−] as X2 and X3, respectively. Models 2–4 placed the residuals of [CO32−] either second or third during HMR to determine if it still contributed significantly to predicting ρA once the other variables had been considered.

Table 2. SLR Equations and Associated Correlation Statisticsa
SpeciesEquation NumberY = a + b(X)nRR2pStandard Error of Estimate
YXab
  1. a

    Reported error with slopes and intercepts are 95% confidence limits.

G. ruber (pink)7[CO32−] (µmol/kg)ρA (355–650 µm; 10−4 µg/µm2)2.390 (±52.66)201.663 (±44.50)140.940.89<0.001±6.34
 8ρA (355–650 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)0.119 (±0.24)0.00442 (±0.001)140.940.89<0.001±0.03
 9ρA (355–500 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)0.00801 (±0.47)0.00488 (±0.002)130.860.73<0.001±0.06
 10ρA (500–650 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)0.126 (±0.33)0.00441 (±0.001)130.910.82<0.001±0.04
 11TC (°C)[CO32−] (µmol/kg)7.516 (±5.72)0.075 (±0.02)140.890.80<0.001±0.72
 12[PO43−] (µmol/kg)[CO32−] (µmol/kg)0.392 (±0.29)−0.001 (±0.002)40.60−0.36<0.05±0.04
G. sacculifer13[CO32−] (µmol/kg)ρA (425–850 µm; 10−4 µg/µm2)−118.224 (±84.74)195.687 (±52.10)130.930.86<0.001±9.63
 14ρA (425–850 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)0.745 (±0.24)0.00440 (±0.001)130.930.86<0.001±0.05
 15ρA (425–650 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)0.669 (±0.36)0.00465 (±0.002)120.650.77<0.001±0.07
 16ρA (650–850 µm; 10−4 µg/µm2)[CO32−] (µmol/kg)1.022 (±0.53)0.00328 (±0.003)120.880.43<0.001±0.09
 17TC (°C)[CO32−] (µmol/kg)10.577 (±1.31)0.064 (±0.006)130.990.98<0.001±0.25
 18[PO43−] (µmol/kg)[CO32−] (µmol/kg)0.863 (±0.52)−0.003 (±0.004)130.48−0.23<0.1±0.15
Table 3. Hierarchical Regression Model and Statistical Output for the Predictors of ρA
Species ModelY = a + b1(X1) + b2(X2) + b3(X3)R2ΔR2p ΔR2Beta
YX1X2X3β1β2β3
G. ruber
1ρA (10−4 µg/µm2)[CO32−] (µmol/kg)0.890.89<0.0010.94
 ρA (10−4 µg/µm2)[CO32−] (µmol/kg)TCres, [CO32−] (°C)0.910.02ns0.940.13
 ρA (10−4 µg/µm2)[CO32−] (µmol/kg)TCres, [CO32−] (°C)[PO43−]res, [CO32−] (µmol/kg)0.920.01ns0.940.17−0.01
2ρA (10−4 µg/µm2)TC (°C)0.810.81<0.0010.90
 ρA (10−4 µg/µm2)TC (°C)[CO32−]res, TC (µmol/kg)0.900.10<0.010.900.31
 ρA (10−4 µg/µm2)TC (°C)[CO32−]res, TC (µmol/kg)[PO43−]res, TC (µmol/kg)0.910.01ns0.900.25−0.10
3ρA (10−4 µg/µm2)[PO43−] (µmol/kg)0.360.36<0.05−0.60
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)[CO32−]res, [PO43−] (µmol/kg)0.890.54<0.001−0.600.73
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)[CO32−]res, [PO43−] (µmol/kg)TCres, [PO43−] (°C)0.920.02ns−0.600.430.34
4ρA (10−4 µg/µm2)[PO43−] (µmol/kg)0.360.36<0.05−0.60
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)TCres, [PO43−] (°C)0.880.52<0.001−0.600.72
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)TCres, [PO43−] (°C)[CO32−]res, [PO43−] (µmol/kg)0.920.04<0.05−0.600.340.43
G. sacculifer
1ρA (10−4 µg/µm2)[CO32−] (µmol/kg)0.860.86<0.0010.93
 ρA (10−4 µg/µm2)[CO32−] (µmol/kg)TCres, [CO32−] (°C)0.870.00ns0.930.09
 ρA (10−4 µg/µm2)[CO32−] (µmol/kg)TCres, [CO32−] (°C)[PO43−]res, [CO32−] (µmol/kg)0.870.00ns0.930.080.04
2ρA (10−4 µg/µm2)TC (°C)0.860.86<0.0010.93
 ρA (10−4 µg/µm2)TC (°C)[CO32−]res, TC (µmol/kg)0.870.00ns0.930.06
 ρA (10−4 µg/µm2)TC (°C)[CO32−]res, TC (µmol/kg)[PO43−]res, TC (µmol/kg)0.870.00ns0.930.070.04
3ρA (10−4 µg/µm2)[PO43−] (µmol/kg)0.160.16ns−0.40
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)[CO32−]res, [PO43−] (µmol/kg)0.860.70<0.001−0.400.84
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)[CO32−]res, [PO43−] (µmol/kg)TCres, [PO43−] (°C)0.870.01ns−0.400.400.47
4ρA (10−4 µg/µm2)[PO43−] (µmol/kg)0.160.16ns−0.40
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)TCres, [PO43−] (°C)0.860.71<0.001−0.400.84
 ρA (10−4 µg/µm2)[PO43−] (µmol/kg)TCres, [PO43−] (°C)[CO32−]res, [PO43−] (µmol/kg)0.870.00ns−0.400.470.38

4 Results and Discussion

4.1 Size-Fraction Relevance and Utilization

[21] A potential limiting factor in traditional SNW studies is the requirement that samples be restricted to narrow size fractions in an attempt to eliminate the contribution of size to the measured weights. However, the use of narrow size fractions may limit the number of foraminiferal shells per sample, and a small sample size increases the error associated with size-normalized weight or ρA estimations as defined by the following error estimation equation [Beer et al., 2010a; Aldridge et al., 2012]:

display math(6)

[22] A large number of individuals per sample greatly decrease the chances of having biased weight or ρA estimations. In this study, shell weight and silhouette area estimates were made for individual foraminifera as opposed to groupings of individuals, allowing for the application of ρA analysis over a wider size fraction (425–850 µm for G. sacculifer and 355–650 µm for G. ruber). SLR comparing ρA to the mean silhouette area for each sample reveals no statistically significant relationship for either species (Figure S1; R2 = 0.02, p = ns for G. ruber; R2 = 0.00, p = ns for G. sacculifer). We therefore conclude that the area density method used in this study is very effective at removing the influence of G. ruber and G. sacculifer shell size on ρA. These results support the use of broader shell size fractions in ρA studies.

[23] A concern with using broader size fractions for ρA and SNW analysis is that the morphology and calcification of foraminifera can change throughout ontogeny [, 1980; Hemleben et al., 1989]. For example, during gametogenesis and following the formation of a sac-like final chamber, G. sacculifer secretes a calcite crust over its entire shell, increasing the thickness of the shell by an average of 9 µm [, 1980]. In this study, only sac-less G. sacculifer individuals were used for ρA analysis in the effort to eliminate the complication of gametogenic calcite formation in this species. Globigerinoides ruber does not precipitate gametogenic calcite [Caron et al., 1990], but it is possible that the influence of [CO32−] on calcification could vary throughout ontogeny for both G. ruber and G. sacculifer. Both field and culture studies have shown that ontogeny, and by extension size-fraction utilization, does not have a significant influence on G. ruber (white) and O. universa SNW-[CO32−] calibrations [Beer et al., 2010a; Bijma et al., 2002]. To test whether the use of broader size fractions has an impact on the calibration equations derived for ρA and [CO32−], we examined the relationship between these two variables for three different size fractions of both G. ruber (355–500, 500–650, and 355–650 µm) and G. sacculifer (425–650, 650–850, and 425–850 µm; Figure 2). We found that the percent change in ρA that occurred with a change in [CO32−] from 200 to 300 µmol/kg (%Δ[CO32−]200–300) ranged from 44 to 50% for the three G. ruber size fractions (Table 2; equations (8)–(10)) and from 20 to 29% for the various G. sacculifer size fractions (Table 2; equations (14)–(16)). The correlation coefficients are lower and the ρA error higher for the narrower size fractions due to the smaller number of individuals per sample in these size fractions (n > 2; Figure 2). We speculate that the small difference in ρA change between the different size fractions is likely due to the difference in the number of individuals per sample, resulting in significant errors associated with the average ρA calculations for the narrower size fractions. Based on these observations, we conclude that the small range in %Δ[CO32−]200–300 for ρA exhibited by the different size fractions illustrates that ontogeny does not significantly affect the relationship between ρA and [CO32−]. The broader size fractions (355–650 µm for G. ruber, 425–850 µm for G. sacculifer) used in this study yield the largest number of individuals per sample and smallest errors and are therefore optimal for generating the calibration equations. Thus, only the data for the broader size fractions will be considered from here on.

Figure 2.

[CO32−]-ρA relationships for both G. ruber and G. sacculifer for size fractions (a) 355–650 and 425–850 µm, (b) 355–500 and 425–650 µm, and (c) 500–650 and 650–850 µm. Error bars represent the ρA multiplied by the reciprocal of the number of individuals in each sample populations. Samples with n > 2 were included for Figures 2b and 2c in order to compare to the broader size-fraction samples presented in Figure 2a (n ≥ 10).

4.2 Calcification Depth and Temperature Estimates

[24] Calcification temperatures derived from the δ18O analyses for each sample were paired with the closest time-equivalent measured water column temperatures. The instrumental temperatures rather than the δ18O-derived temperatures are used for the carbonate calculations and statistical analyses in order to maintain consistency with the rest of the hydrographic data used in this study. Average temperatures were comparable to previously published optimum temperatures for both G. ruber (26°C for this study versus 27°C from Mulitza et al. [1998]) and G. sacculifer (23°C for this study versus 22°C from Mulitza et al. [1998]). The instrumental temperature values for the upper 130 m for the 3 year study period, along with the estimated calcification depths for G. ruber (black circles) and G. sacculifer (blue diamonds) from each sediment trap sample are shown in Figure 3a. The mean calcification depth for G. ruber is 16 (±19) m, which is consistent with this species living in the surface mixed layer in the Cariaco Basin [Miro, 1971; Tedesco et al., 2007] and falls within the previously observed depth range of 0–50 m [Hemleben et al., 1989; Farmer et al., 2007].

Figure 3.

Contour plots of (a) temperature and (b) [CO32−] from March 2005 to October 2008 in the Cariaco Basin for the upper 160 m of the water column. Calcification depths, estimated from δ18O-derived calcification temperatures, are shown for G. ruber (black circles) and G. sacculifer (blue diamonds). Also shown are the estimated depth ranges for each sample estimated from the instrumental and δ18O-derived calcification temperatures (see the supporting information for more details). Optimal growth temperatures for both G. ruber (black line) and G. sacculifer (blue line) are also plotted [Mulitza et al., 1998] to compare to the estimated calcification depths for both species.

[25] The estimated calcification depths for G. sacculifer range from 15 to 100 m, with the mean being ~50 (±28) m. These results are in line with previous depth estimates for the species from the Cariaco Basin [Wejnert, 2011]. For both species, calcification depth changes seasonally in response to shifts between upwelling and nonupwelling regimes in the basin (Figure 3a). Changes in calcification depth are likely a response to changes in ambient water density at depth due to transitions between upwelling and nonupwelling regimes and to a certain extent species-specific preferences to live at a depth characterized by an optimal temperature, salinity, light, and/or chlorophyll and nutrient regimes [Hemleben et al., 1989; Sautter and Thunell, 1991; Tedesco et al., 2007].

4.3 Carbonate System Calculations

[26] The [CO32−] record for the upper 120 m over the course of the study period is shown in Figure 3b, along with the calcification depths determined for each sample population for both species. The [CO32−] at the calcification depths for G. ruber ranged between 215 and 270 µmol/kg (mean = 240 µmol/kg) throughout the study period, coinciding with calcite saturation states (Ωcalc) ranging from 5.0 to 6.5 (mean = 5.7). [CO32−] and Ωcalc for G. sacculifer were on average lower than those for G. ruber (165–240 µmol/kg, mean [CO32−] = 200 µmol/kg, mean Ωcalc = 4.7), in agreement with G. sacculifer's deeper depth habitat.

4.4 Carbonate, Temperature, and Phosphate Controls on Planktonic Foraminiferal ρA

4.4.1 Results from Simple and Multiple Linear Regression Analysis

[27] SLR revealed that the ρA for both species has a highly significant (p < 0.001) relationship with ambient [CO32−] (Table 2 and Figure 2a). SLR performed using temperature as the predictor variable for ρA also revealed a significant positive linear relationship (Figures 4a and 4b). In comparison, SLR using [PO43−] as the predictor variable revealed a less significant negative linear relationship with G. ruber ρA (p < 0.05), with no significant relationship between G. sacculifer ρA and [PO43−] (Figures 4c and 4d). Additionally, SLR using [CO32−] as the predictor variable and calcification temperature and [PO43−] concentrations individually as the response variables yielded significant bivariate correlation amongst these variables, with the exception of G. sacculifer [PO43−] and [CO32−] (Table 2 and Figure 4). Table S2 shows the results of the MLR using each predictor variable interchangeably as the dependent variable, with the other predictor variables serving as the independent variables. For both species, [CO32−] and temperature are highly redundant with the other environmental parameters (R2 = 0.86 and 0.82 and R2 = 0.98 and 0.98 for G. ruber and G. sacculifer, respectively). MLR using [PO43−] as the dependent variable and temperature and [CO32−] as the independent variables reveal that [PO43−] is moderately redundant with temperature and [CO32−] for both G. ruber and G. sacculifer (R2 = 0.44 and 0.26, respectively). By substituting in a variable we know does not share a statistically significant relationship with the predictor variables (i.e., mean area, R2 = 0.00, 0.01, and 0.04 and R2 = 0.01, 0.01, and 0.05 for G. ruber and G. sacculifer [CO32−], temperature, and [PO43−], respectively), we can quantify the specific redundancy each predictor variable shares with another (Table S2). This test revealed that G. ruber temperature and [PO43−] are 81 and 42% redundant with [CO32−] and that G. sacculifer temperature and [PO43−] are 98 and 30% redundant with [CO32−].

Figure 4.

Temperature-ρA and temperature-[CO32−] relationships for (a) G. ruber and (b) G. sacculifer. Also shown are [PO43−]-ρA and [PO43−]-[CO32−] relationships for (c) G. ruber and (d) G. sacculifer.

[28] The tolerance and VIF statistics were less than 0.2 and greater than 5, respectively, for both [CO32−] and temperature for both species, revealing a strong case for collinearity for these variables (Table S3). Substituting in the residuals of temperature (TCres, [CO32−]) and [PO43−] ([PO43−]res, [CO32−]) revealed no collinearity with [CO32−]. For both species, the tolerance and VIF diagnostics for [PO43−] did not indicate a strong case for collinearity with the other predictor variables. However, due its redundancy with the other predictor variables (Table S2), [PO43−] is treated for possible collinearity with [CO32−] or temperature in the subsequent HMR analyses. These results suggest that [CO32−] and temperature, particularly with the G. sacculifer data, are nearly indistinguishable from one another in regression analyses.

[29] Based on the results of these statistical analyses, we hypothesize that the relationships exhibited between one predictor variable and ρA could be due to its strong collinearity with another predictor variable that serves as the dominant predictor for ρA. This hypothesis is graphically represented in Figure 4. Figures 4a and 4b show the relationships between temperature and ρA and temperature and [CO32−] for G. ruber (left) and G. sacculifer (right), while Figures 4c and 4d illustrate the same relationships for [PO43−]. The slopes of the best fit lines illustrated in each graph cannot be directly compared as they are on different scales. However, the correlation coefficients (R) for both species are similar in all cases, suggesting that the relationship between ρA and temperature or [PO43−] could be due the collinearity of ρA with [CO32−].

4.4.2 Results from Hierarchical Multiple Regression Analyses

[30] Four HMR models were run in order to determine the relative contributions of each predictor variable to ρA. The means, standard deviations, and number of data points (n) for each variable used in the HMR models are included in Table S4. The R2, the R2 change (ΔR2), the significance of the ΔR2 (p ΔR2), and the standardized coefficient (β) for each model are listed in Table 3. The first model uses results from prior studies examining the dominant control variable on foraminiferal calcification [Barker and Elderfield, 2002; Naik et al., 2010; Manno et al., 2012] to determine the ordering of predictor variables for HMR, with [CO32−], and the residuals of calcification temperature and [PO43−] based on their relationship with [CO32−] (TCres, [CO32−] and [PO43−]res, [CO32−]) serving as X1, X2, and X3, respectively. The R2 values from model 1 indicate that [CO32−] accounts for 89 and 86% of the variability seen in ρA for G. ruber and G. sacculifer, respectively. For both G. ruber and G. sacculifer, ΔR2 for the additions of TCres, [CO32−] and [PO43−]res, [CO32−] were insignificant, with each accounting for ~0–2% of the variability in ρA once [CO32−] was controlled for in the model. Significantly larger β values for [CO32−] relative to those for TCres, [CO32−] and [PO43−]res, [CO32−] indicate a strong dominance of [CO32−] for predicting variability in ρA.

[31] The other three models were performed on both species to test if [CO32−] still played a significant role in predicting ρA when placed second or third in the ordering of predictor variables. For the G. ruber data, the addition of [CO32−] as X2 and X3 in models 2 and 4 following the addition of temperature generated a significant contribution to the R2 of the model, while the addition of temperature as X2 and X3 following the addition of [CO32−] in models 1 and 3 did not contribute significantly to the R2 of the model. However, the ordering of temperature and [CO32−] in models 1–4 for G. sacculifer did not make a significant difference in the model output. This is likely a result of the strong—nearly perfect—collinearity that exists between G. sacculifer [CO32−] and temperature (Tables 2 and 3), thus making them indistinguishable during HMR. The predictor variables for G. ruber are slightly less collinear and thus provide us with better estimates of the relative contributions of [CO32−] and temperature to the variability in ρA. Taken together with the knowledge that G. ruber temperature and [CO32−] are 81% redundant, these results indicate that [CO32−] is the dominant factor controlling ρA and that model 1 most accurately reflects the relative contributions of each predictor variable. Based on this model, we cannot say with any confidence that either calcification temperature or [PO43−] has an impact on the variability in G. ruber or G. sacculifer ρA, or by extension calcification efficiency. We conclude that [CO32−] alone acts as an excellent predictor for both G. ruber and G. sacculifer ρA and the SLR equations reported in Table 2 (equations (7), (8), (13), and (14)) serve as reliable calibration equations.

4.4.3 Globigerinoides ruber and Globigerinoides sacculifer Area Density as a Proxy for [CO32−]

[32] The ρA of G. ruber (pink; 355–650 µm) ranged from 1.04 to 1.31 × 10−4 µg/µm2 over the 3 year study period with an average of 1.18 × 10−4 µg/µm2, yielding a strong positive linear relationship with [CO32−] (R2 = 0.89, p < 0.001, Figure 2a). Globigerinoides sacculifer (425–850 µm) ρA ranged from 1.45 to 1.81 × 10−4 µg/µm2 with an average of 1.62 × 10−4 µg/µm2 and also correlated strongly with ambient [CO32−] (R2 = 0.86, p < 0.001, Figure 2a). The relationships between foraminiferal ρA and [CO32−] reported in Table 2 are in line with the results of prior studies reporting an adverse effect of reduced [CO32−] associated with ocean acidification on the calcification of planktonic foraminifera [Spero et al., 1997; Bijma et al., 1999; Barker and Elderfield, 2002; Russell et al., 2004; Mekik and Raterink, 2008; Moy et al. 2009; Lombard et al., 2010; Manno et al., 2012].

[33] The slopes and y intercepts for G. ruber reported for this study (Table 2) reveal a more sensitive relationship with [CO32−] than those for G. sacculifer, with a 200 to 300 µm/kg change in [CO32−] (%Δ[CO32−]200–300) resulting in a change in ρA of 44% for G. ruber and 27% for G. sacculifer (Table 1). It has been well documented that different species of planktonic foraminifera undergo varying degrees of isotopic fractionation and elemental incorporation during the calcification process due to various vital effects associated with calcification [Erez, 1978; Spero, 1992; Wolf-Gladrow et al., 1999; Zeebe et al., 2008; Henehan et al., 2013]. Though the ρA for G. ruber and G. sacculifer reflects changes in ambient seawater [CO32−], the [CO32−] at the site of calcification could vary amongst species due to differences in both foraminiferal and symbiont vital effects (calcification, photosynthesis, respiration) [Jorgensen et al., 1985; Rink et al., 1998; Wolf-Gladrow et al., 1999; Bentov et al., 2009]. Thus, when using ρA or any other measure of shell weight as a proxy for past surface-ocean [CO32−], it is necessary to use species-specific equations like those provided in this study.

[34] An additional consideration for the use of foraminiferal ρA as a proxy for [CO32−] is the extent to which foraminifera are preserved in marine sediment samples [Barker, 2004; Gibbs et al., 2010]. The dissolution of planktonic foraminiferal calcite due to a low [CO32−] at depth would result in lower foraminiferal ρA and complicate the use of ρA as a proxy for surface-ocean [CO32−]. Conversely, the addition of secondary calcite would increase the shell thickness and hence the ρA of foraminifera. Thus, foraminiferal specimens should be collected from well above the lysocline for the study region and thoroughly examined for signs of dissolution and/or the precipitation of secondary calcite prior to being used for ρA-[CO32−] reconstructions.

4.5 Comparison to Previous Studies

[35] Foraminiferal ρA cannot be directly compared to previous studies that investigated the relationship between foraminiferal calcification efficiency using SNW, shell thickness measurements or calculations, or calcification rates as the units are not the same (Table 1). Additionally, this study differs from most other field studies in that we use [CO32−] at the predicted depth of calcification rather than surface water [CO32−] to generate our calibration equations. We can compare the various shell weight proxies more directly by examining the change in each proxy resulting from a 200 to 300 µm/kg change in [CO32−]. These changes in calcification were determined using regression equations reported in the respective studies or regression equations derived from digitized figures (Table 1). For studies that did not include [CO32−] values, the %Δ reported in Table 1 represent either a reported %Δ in the study [Spero et al., 1997] or a change in SBW observed over the course of a geologic period characterized by significant changes in surface water [CO32−] [Moy et al., 2009; de Moel et al., 2009; Naik et al., 2010]. The %Δ[CO32−]200–300 varies widely depending on the species studied and the proxies used. The average %Δ[CO32−]200–300 reported for G. sacculifer is 32% (Table 1), which is close to the %Δ[CO32−]200–300 reported for this species in this study (27%), though this percentage was derived from both culture and core studies that vary widely in the range in [CO32−] and methods for determining calcification. The %Δ[CO32−]200–300 for G. sacculifer reported in this study is most similar to the percent change in G. sacculifer SBW reported in a core study spanning 25,000 years B.P. to 1000 years B.P. [Naik et al., 2010].

[36] The results from Beer et al. [2010a] are highly inconsistent with the results for G. ruber (pink) presented here, with Beer et al. [2010a] reporting a negative correlation between G. ruber (white) MBW and [CO32−] collected from the Arabian Sea. Recent studies have distinguished between five different genetic types for the white variety of G. ruber [Aurahs et al., 2011], characterized broadly by two distinct morphotypes: sensu stricto (s.s.) and sensu lato (s.l.) [Wang, 2000]. These morphotypes have different depth habitats and temperature preferences, and thus paleoceanographic and paleoclimatic studies should distinguish between them [Hecht and Savin, 1972; Wang, 2000; Kuroyanagi et al., 2008; Numberger et al., 2009; Aurahs et al., 2011]. This difference in calcification habitat, as well as the evident difference in shell geometry (s.l. is more heavily calcified than s.s. (J. Durrant and M. Henehan, 2013, unpublished data)), would likely result in a significant differences in the MBW for the two G. ruber (white) morphotypes. The differences between the results from Beer et al. [2010a] and the results from other SNW studies also using G. ruber (white) collected from the Arabian Sea [de Moel et al., 2009; Naik et al., 2010] may be due to Beer et al. [2010a] not distinguishing between the two morphotypes which we know to be present in this region [de Moel et al., 2009; Aurahs et al., 2011] and whose relative abundances may have changed as sampling traversed upwelling and nonupwelling waters. The use of G. ruber (pink) in the current study and G. ruber (white) of an undetermined morphotype by Beer et al. [2010a] makes comparisons difficult between the two studies. Our results are closest to those reported in de Moel et al. [2009], who used the SBW of G. ruber (white; both morphotypes in equal distributions amongst samples), but had a much narrower range in [CO32−] compared to the current study. As this is the first study reporting on the effect of [CO32−] on the calcification in the pink variety of G. ruber, we can only state that the %Δ[CO32−]200–300 reported here (44%) falls within the range of most percent changes reported in studies using other species (5 to 155%, with the majority between 5 and 50%). Thus, our results for the magnitude of change in ρA per unit change in [CO32−] for both G. ruber and G. sacculifer are comparable to those changes previously reported in foraminiferal calcification-[CO32−] studies.

5 Conclusions

[37] The results of this study suggest that surface [CO32−] is responsible for 89 and 86% of the variability in ρA for both G. ruber (pink) and G. sacculifer, respectively, and by extension calcification efficiency, with no significant evidence that temperature or [PO43−] contributes to ρA in these species. Thus, the ρA of G. ruber and G. sacculifer should serve as a reliable proxy for past [CO32−] using the species-specific equations reported in Table 2. It is recommended that only well-preserved foraminiferal shells with an absence of secondary calcite be used in down-core ρA reconstructions of past [CO32−]. The ρA technique described in this study should be particularly useful for down-core studies where foraminiferal shell numbers are limited and the use of a broad size range is required.

Acknowledgments

[38] We would like to thank Eric Tappa for his extensive analytical contributions, including the processing of samples for stable isotopes and the overseeing of the collection of the sediment trap samples used in this study. This research was supported by NSF awards 1039503 and 0752037.

Ancillary