4.2.1. Approach 1: Assume Equal Partitioning of δ18Obenthic Into Temperature and δ18Oseawater Components
 Assuming that δ18Obenthic records comprise a 50% contribution from temperature (Table 1, scenario 1), the Site 1209 record would support variations of 0.30‰–0.65‰ in δ18Oseawater, equivalent to 36–78 m ΔASL. The lower range (0.30‰) of these estimates are similar to those reported by Edgar et al.  based on a δ18Obenthic record from the deep equatorial Atlantic (ODP Site 1260) and the same assumption of a 50% temperature contribution (Figure 2). Edgar et al.  argue that Δδ18Oseawater of this magnitude indicates variations in continental ice budgets that could be accommodated by ice storage solely on Antarctica. Although this calculation does not preclude the occurrence of ephemeral glaciations in the Northern Hemisphere as previously proposed by Tripati et al. .
 The higher range of the calculated Δδ18Oseawater at Site 1209 greatly exceeds that reported for Site 1260 (Figure 2 [Edgar et al., 2007]) and would support significantly larger ice budgets. Variations in δ18Obenthic ranging from 0.6‰ to 1.2‰ are also observed in middle Eocene records from ODP Site 1218 (deep Pacific) [Tripati et al., 2005] and at multiple sites in the Southern Ocean [Bohaty and Zachos, 2003]. Using this same framework to interpret δ18Obenthic, this amplitude of variability would also indicate changes in δ18Oseawater of 0.3‰–0.6‰ during the late middle Eocene, and also indicate substantially greater ice budgets than proposed by Edgar et al. .
4.2.2. Approach 2: Applying a δ18Obenthic–Sea Level Calibration
 A second approach to estimating ice budgets from δ18Obenthic is to use a published calibration based on contemporaneous measurements of apparent sea level and δ18Obenthic [e.g., Pekar et al., 2006] (scenario 2; Table 1). We use calibrations derived from Oligocene and Miocene records [Pekar et al., 2006], as these likely provide more appropriate climatic analogies than Pleistocene records. However, we note that the propagated error for the calibration is likely to be large, as it will reflect uncertainties associated with the δ18Obenthic records, the backstripped eustatic estimates and conversion of eustasy to apparent sea level. Based on end-member calibrations, the δ18Obenthic record for Site 1209 supports variations of 23–108 m in apparent sea level (Table 1). These wide-ranging estimates of changes in apparent sea level support the occurrence of glaciation during the late middle Eocene, but could be accommodated by different ice budget scenarios.
4.2.3. Approach 3: Using Coupled δ18Obenthic and Benthic Foraminiferal Mg/Ca-Paleotemperature Reconstructions to Independently Constrain δ18Oseawater
 Benthic foraminiferal Mg/Ca ratios and Mg/Ca derived estimates of bottom water temperatures from Site 1209 (Figures 3 and 5) are similar to values previously reported in a low-resolution study [Dutton et al., 2005]. Mean temperatures are estimated to be ∼9°C during the middle Eocene (based on a constant seawater Mg/Ca ratio of 2.25 mol/mol, see discussion below), with variations of up to 2°C. The Mg/Ca-based temperature reconstruction for Site 1209 likely reflects variations in Southern Ocean sea surface temperatures, as the neodymium isotopic composition of fish teeth at this site has been cited as evidence of a Southern Ocean source for intermediate bottom waters in the Pacific Ocean during the middle Eocene [Thomas, 2004]. These estimates are consistent with Mg/Ca-based temperatures reconstructed for several high-latitude sites, including in the Weddell Sea (ODP Site 689 [Lear et al., 2000; Billups and Schrag, 2003]) and Indian Ocean (ODP Site 757 [Billups and Schrag, 2003]), and to Arctic Ocean sea surface temperatures reconstructed for an overlapping interval from alkenones (44.5 Ma: ∼8°C–11°C) [Weller and Stein, 2008].
 At Site 1209, the long-term and short-term trends in benthic foraminiferal Mg/Ca are notably different to the benthic δ18O record (Figure 1). Taken at face value, this observation suggests that there may have been significant increases in δ18Obenthic that were accompanied by either no temperature change or an increase in temperature at Site 1209. As a result, the majority of the δ18Obenthic record for Site 1209 could be attributed to changes in δ18Oseawater (ice volume). Previous studies have also documented little or no temperature decrease at low-latitude open ocean sites associated with the large increase in δ18Obenthic observed across the Eocene-Oligocene boundary (Mg/Ca reconstruction of Lear et al. ; TEX86 and U37k′ reconstructions of Liu et al. ) and during the middle Miocene (Mg/Ca reconstruction of Holbourn et al. ). In contrast, organic proxy-based temperature reconstructions for the high latitudes support significant cooling across the Eocene-Oligocene boundary, which may indicate that the temperature change at this time was very heterogeneous, or may reflect inaccuracies in one or both temperature reconstructions [e.g., Sachs et al., 2000; Elderfield et al., 2006; Turich et al., 2007; Lipp et al., 2008].
 The inferred relationship between ice growth and Pacific Ocean warming at ∼2 km water depth (and Southern Ocean warming) is enigmatic, and contrasts with Quaternary records, which show ice volume, and high-latitude sea surface temperatures are positively correlated [e.g., Martin et al., 2002; Elderfield et al., 2010]. For the purpose of this discussion, we adopt a “straw man” approach to evaluating the fidelity of the Site 1209 Mg/Ca data set as a record of water temperature.
18.104.22.168. Testing End-Member Interpretations for the Benthic Mg/Ca Record at Site 1209
 In an exercise, we assume that benthic foraminiferal δ18O, Mg/Ca, and Mg/Ca-based temperatures should positively covary, as observed in Pleistocene records [e.g., Martin et al., 2002; Elderfield et al., 2010]. We therefore consider the additional parameters (e.g., seawater Mg/Ca, diagenesis, carbonate ion effect) that may influence foraminiferal Mg/Ca [Lear et al., 2000, 2004; Billups and Schrag, 2003; Elderfield et al., 2006] and explore the extent to which these parameters must have changed in order to reconcile the benthic δ18O and Mg/Ca records. We then evaluate whether these “competing” parameters can account for some or all of the observed variability observed in the Mg/Ca record from Site 1209 (Table 2).
Table 2. Estimates of the Potential Impact of Parameters Other Than Temperature on Mg/Ca Record at Site 1209
|Changing seawater Mg/Ca overprints short-term (<106 years) variations in benthic Mg/Ca record|| ||Long residence time for Mg and Ca in seawater (>106 years)|
|Changing seawater Mg/Ca overprints long-term (>106 years) variations in benthic Mg/Ca record||All reconstructions and modeled histories of seawater Mg/Ca show increase of up to 0.5 mol/mol; an increase of 0.2 mol/mol could overprint to 3°C of cooling|| |
|Bottom water Δ[CO32−] overprints short-term (<106 years) variations in benthic Mg/Ca record|| ||No significant variations observed in carbonate accumulation; potential Δ[CO32−] bias on Mg/Ca temperatures during CAE events acts to suppress rather than amplify the apparent negative covariation of benthic δ18O and temperature.|
|Bottom water Δ[CO32−] overprints long-term (>106 years) variations in benthic Mg/Ca record||Increasing planktonic foraminiferal fragmentation in late middle Eocene||Planktonic fragmentation not necessarily representative of bottom water Δ[CO32−]|
|Diagenetic alternation overprints variations in benthic Mg/Ca record||Benthic foraminifera appear “frosty” and increasingly fragmented in the late middle Eocene. The uncertainty is difficult to accurately quantify.||Intersample foraminifera preservation variable; the uncertainty is difficult to accurately quantify|
 We show below that changing seawater Mg/Ca probably cannot account for the observations at Site 1209, and that the discrepancies between the δ18Obenthic and foraminiferal Mg/Ca records cannot be reconciled with existing proxy estimates of carbonate dissolution at Site 1209. The effect of preservation on the Site 1209 bottom water temperature reconstruction is difficult to estimate, and although benthic foraminifera at Site 1209 do not exhibit the exquisite preservation found in continental shelf and slope environments, the record is less likely to contain dissolution-related artifacts than records from deeper sites.
22.214.171.124. Consideration of Nontemperature Effects on Foraminiferal Mg/Ca: Seawater Mg/Ca Ratios
 Absolute temperature estimates based on benthic foraminifera Mg/Ca are sensitive to the precise value of seawater Mg/Ca used in calculations [Tripati et al., 2003; Billups and Schrag, 2003; Sexton et al., 2006b], and there are several different reconstructions for the history of Cenozoic seawater Mg/Ca (Figure 4). It is unlikely that short-term variations (<1 Myr) in foraminiferal Mg/Ca records from the middle Eocene reflect changes in seawater Mg/Ca because of the long residence times of Mg and Ca in the oceans (107 and 106 years, respectively [Broecker and Peng, 1982]). However, as the total duration of the benthic foraminiferal Mg/Ca record for Site 1209 exceeds the residence time of these ions in seawater, some component of the long-term (>1 million years) trends may reflect variations in seawater Mg/Ca.
 We consider a range of seawater Mg/Ca histories and evaluate the effect on bottom water temperature estimates (Figure 4). In order to reconcile the benthic foraminiferal δ18O and Mg/Ca records (i.e., to have them positively covary), seawater Mg/Ca would have had to increase during the middle Eocene (Figure 4). If deep ocean temperatures at Site 1209 cooled by ∼3°C over the middle Eocene (45–36 Ma), to attain this magnitude of cooling from the Site 1209 Mg/Ca record, seawater Mg/Ca would have had to increase by ∼0.2 mol/mol over 9 million years (green model in Figure 4).
 Existing reconstructions suggest that seawater Mg/Ca may have varied by as much as 60% over the past 65 million years [Sandberg, 1983; Wilkinson and Algeo, 1989; Lowenstein et al., 2001; Dickson, 2002; Horita et al., 2002; Creech et al., 2010]. As discussed previously by Tripati et al. , Billups and Schrag , and Sexton et al. [2006b], over tens of millions of years there are notable differences between published seawater Mg/Ca reconstructions that are based on fluid inclusion concentrations in halite crystals [e.g., Horita et al., 2002] (closed squares in Figure 4a) and models based on midocean ridge spreading rates [e.g., Stanley and Hardie, 1998] and cation fluxes [Wilkinson and Algeo, 1989] (circles in Figure 4a). During the middle Eocene, interpolations from all proxy- and model-based reconstructions suggest a relatively small change (<0.5 mol/mol) in seawater Mg/Ca may have occurred, consistent with what is needed to reconcile the Site 1209 benthic Mg/Ca record with a long-term 3°C cooling.
 Within the considered range of seawater Mg/Ca histories, there are a number of different of scenarios that are compatible with both the estimated long-term cooling of 3°C and proxy/model-based data (green and light blue models in Figure 4). However, the different scenarios result in different absolute temperature estimates. The cation flux-based seawater Mg/Ca reconstruction of Wilkinson and Algeo  predicts average bottom water temperatures at Site 1209 of ∼9°C. We assume that the Wilkinson and Algeo  seawater Mg/Ca reconstruction is most appropriate for the Eocene (Figure 4c) as the absolute temperature estimates are most compatible with other independent temperature estimates from alkenones [Weller and Stein, 2008].
 Although it is possible to reconcile the long-term change in foraminiferal Mg/Ca–bottom water temperatures at Site 1209 with the decline in benthic foraminiferal δ18O by changing seawater Mg/Ca, there are still notable discrepancies on shorter timescales (<1 Myr). Given the residence time of these ions in seawater, it is unlikely that the negative covariation of Mg/Ca and benthic δ18O on relatively short timescales (<106 years) results from changing seawater Mg/Ca.
126.96.36.199. Bottom Water Carbonate Saturation
 The influence of bottom water carbonate saturation (Δ[CO32−]) on foraminiferal Mg/Ca ratios may introduce some bias in paleotemperature reconstructions, although the exact cause of this effect is not yet well understood and its magnitude is of debate [Elderfield et al., 2006; Rosenthal et al., 2006; Lear, 2007; Yu and Elderfield, 2008; Elderfield et al., 2010]. Recent work suggests that the sensitivity of foraminiferal Mg/Ca ratios to Δ[CO32−] may differ significantly between species [Elderfield et al., 2006; Yu and Elderfield, 2008], with infaunal species exhibiting much weaker sensitivities in comparison to epifaunal species [Elderfield et al., 2010]. The reason for the apparent differences in the species specific sensitivity of Mg/Ca ratios to Δ[CO32−] is unclear and likely complex, but Elderfield et al.  suggest that the weaker sensitivity of infaunal species may in part reflect the fact that they calcify in pore waters rather than bottom waters. Although pore water temperatures in the upper few centimeters of the sediment (where O. umbonatus resides [Corliss, 1985]) will be the same as bottom waters, pore water Δ[CO32−] tends away from bottom water values toward zero [Martin and Sayles, 1996]. As a result, the Mg/Ca ratio of infaunal species may be less sensitive to changes in Δ[CO32−]. In fact, there is some indication from Mg/Ca values of modern Oridorsalis umbonatus that this species may not be sensitive to changes in saturation state [Rathmann and Kuhnert, 2008]. This study was based on comparing estimates of pore water Δ[CO32−] to test Mg/Ca ratios in a limited number of samples (n = 6), and therefore additional study is necessary to test whether this conclusion is robust.
 Using the sensitivity of benthic foraminiferal Mg/Ca to bottom water Δ[CO32−] established for Cibicidoides wuellerstorfi [Elderfield et al., 2006] we include in our error propagation a conservative estimate of the error in bottom water temperature that would result from a large change in carbonate ion saturation of ∼20 μmol/kg [Tripati and Elderfield, 2005]. Based on a new core top calibration for Oridorsalis umbonatus (C. F. Dawber and A. K. Tripati, Relationships between bottom water carbonate saturation and element/Ca ratios in core top samples of the benthic foraminifera Oridorsalis umbonatus, submitted to Paleoceanography, 2011), we note that this sensitivity may overestimate the contribution of changes in carbonate saturation to the Mg/Ca record of Oridorsalis umbonatus.
 Previous studies have demonstrated that there were large oscillations in Pacific deep-water carbonate preservation during the middle Eocene, likely linked to variations in global ice storage as evidenced by contemporaneous increases in benthic foraminiferal and seawater δ18O [Tripati et al., 2005]. These carbonate accumulation events (CAE) support substantial changes in the carbonate compensation depth (CCD) of up to 1 km [Tripati et al., 2005], which may have introduced some bias into deep water Mg/Ca–bottom water temperature reconstructions [Tripati et al., 2005]. A criticism of the records from Site 1218 is that they are from sites that are near the CCD [Edgar et al., 2007]. Site 1209, however, had a paleodepth of ∼1.9–2.5 km during the middle Eocene [Bralower et al., 2003], which is ∼2 km above the estimated average depth of the CCD and ∼1 km above the estimated CCD at its shallowest point. It is unclear if such large changes in deep-water carbonate saturation would have propagated to intermediate-depth bottom waters. Carbonate accumulation data for Site 1209 does not support large changes in intermediate-depth bottom water carbonate saturation [Hancock and Dickens, 2005; Bohaty et al., 2009], with one possible exception at ∼40.1 Ma [Bohaty et al., 2009]. Other carbonate dissolution proxy data have been cited as evidence for a relatively shallow Pacific lysocline during the middle Eocene (Hancock and Dickens, see discussion below, section 188.8.131.52). We therefore consider whether the Mg/Ca bottom water temperature estimates at Site 1209 may be biased by changing carbonate saturation (Figure 5).
 At the peak of CAE-3 (∼41 Ma, the largest carbonate accumulation event), a transient, but substantial (0.7‰) decrease is observed in the Site 1209 benthic foraminiferal δ18O record (Figure 5), accompanied by a small (∼1°C) decrease in Mg/Ca-based bottom water temperatures. An increase in carbonate saturation at this time may have a positive bias on foraminiferal Mg/Ca values, resulting in an overestimation of the change in bottom water temperature. Thus, the observed temperature decrease should be considered as a minimum estimate. If changing carbonate saturation is causing a bias in foraminiferal Mg/Ca at Site 1209 during CAE-3, it is acting to suppress, and not amplify, the enigmatic relationship between benthic foraminiferal δ18O and Mg/Ca-based intermediate bottom water temperatures.
 During CAE-4, an increase of 0.6‰ is observed in the Site 1209 benthic δ18O record (Figure 5). The change in Mg/Ca-based temperature estimates across this event can be described by two trends. Between the onset of CAE-4 and peak accumulation, temperatures at Site 1209 are invariant, while between peak accumulation and the termination of CAE-4, temperatures decline. It may be possible that a decline in bottom water temperature between the onset of CAE-4 and peak accumulation is obscured by a positive bias resulting from an increase in carbonate saturation. If we assume the temperature decline should contribute approximately half of the benthic δ18O increase, the change in bottom water carbonate saturation needed to reconcile a ∼0.15 mmol/mol change in foraminiferal Mg/Ca is on the order of ∼+20 μmol/kg (based on the sensitivity established for C. wuellerstorfi [Elderfield et al., 2006]). This value is similar to the magnitude (although opposite in direction) to that calculated by Yu and Elderfield  for the change in Atlantic deepwater carbonate saturation between the Last Glacial Maximum and the Holocene, which was accompanied by an 800 m shoaling of the carbonate saturation depth. The amplitude of the CCD deepening across CAE-4 (∼700 m) may be consistent with a potential carbonate saturation bias on foraminiferal Mg/Ca only if the change in intermediate-depth bottom water carbonate saturation is similar in amplitude (and direction) to deep water. This hypothesis is difficult to test at present given the difficulty in interpreting changes in carbonate accumulation in carbonate-dominated sediment. Interestingly, two proxies commonly used to infer carbonate dissolution, planktonic foraminiferal fragmentation, and the relative abundance of benthic foraminifera, suggest increased dissolution at Site 1209 between 39.5 Ma and 38.5 Ma [Hancock and Dickens, 2005] (Figure 1, 151.5 – 148 rmcd). If these proxy reconstructions are accurate, they suggest that the lysocline and carbonate saturation depth may have been decoupled during CAE-4. As a result, decreased carbonate saturation at intermediate sites may result in a negative carbonate ion bias on foraminiferal Mg/Ca–bottom water temperatures, which again is acting to suppress the apparent negative covariation of benthic δ18O and bottom water temperatures.
 There likely were changes in Pacific intermediate and deep-water carbonate saturation during the middle Eocene. However, if our present understanding of the sensitivity of foraminiferal Mg/Ca to changing carbonate saturation is appropriate for the Eocene, and carbonate dissolution reconstructions at Site 1209 are accurate, then changes in intermediate water carbonate saturation cannot be solely responsible for the apparent negative correlation between records of benthic δ18O and Mg/Ca at Site 1209 on timescales of <106 years.
184.108.40.206. Dissolution and Diagenetic Alteration
 Diagenetic processes including dissolution and recrystallization may also influence the geochemistry of foraminiferal calcite, although the impacts are difficult to robustly determine. The influence of dissolution should be minimal given the relatively shallow water depth of Site 1209 relative to the carbonate compensation depth. Two proxies for dissolution have been cited as evidence for a shallower lysocline and enhanced dissolution at Site 1209 during parts of the middle and late Eocene [Hancock and Dickens, 2005]. We developed high-resolution records of these indices for this study (Figure 1); these are broadly consistent with the results from Hancock and Dickens  and exhibit cumulative increases during the late middle Eocene.
 The tests of planktonic foraminifera may be more susceptible (than benthics) to postmortem dissolution as they calcify in surface waters that are likely to be more saturated (with respect to Δ[CO32−]) than their depositional environment and they have highly porous tests. In addition, dissolution in planktonic foraminifera begins as they descend through the water column [e.g., Schiebel et al., 2007]. The test structure of Eocene planktonic foraminifera also appears to influence the susceptibility to dissolution [e.g., Petrizzo et al., 2008]. Unless planktonic fragments are identified to a genera level, records of this index will be sensitive to changes in the faunal assemblage [Petrizzo et al., 2008]. We note that planktonic fragmentation in early Eocene sediments from Site 1209 could not unequivocally be attributed to carbonate dissolution [Petrizzo et al., 2008]. As the benthic foraminiferal abundance index is quoted with respect to the number of whole planktonic foraminifera, it may also be biased by these same factors. Nevertheless, increased planktonic fragmentation and benthic abundance at Site 1209 correlates with darker core material (Figure 1), which may indicate a reduction in the carbonate:organic carbon ratio. Although a few studies have documented notable heterogeneity of Mg/Ca in deep-water benthic foraminifera [e.g., Rathmann et al., 2004], the sensitivity of the Mg/Ca thermometer to dissolution is still poorly constrained.
 Scanning electron microscope (SEM) images of benthic foraminifera indicate an increasing effect of dissolution in the upper part of the studied interval. However, within a single sample, the preservation of individual tests varies significantly (refer to auxiliary material). Care was taken to select the best preserved (i.e., intact and nonchalky) foraminiferal tests for analysis and specimens that were fragmented or had obvious holes were not selected. None of the benthic foraminiferal δ18O or Mg/Ca excursions discussed below corresponds to a major change in foraminiferal preservation (Figure 1).
 It is likely that any secondary diagenetic calcite would have formed in pore waters similar in temperature and seawater δ18O to bottom waters. An inability to quantify the amount of “neomorphosed” spar present in tests, and the lack of consensus over appropriate partition coefficients for Mg in diagenetic calcite [Tripati et al., 2003; Sexton et al., 2006b], means that recrystallization represents a source of uncertainty that is difficult to accurately quantify.