Mg/Ca analyses were performed on G. bulloides retrieved from well-preserved core tops spanning the western Mediterranean Sea (Boussetta et al., 2011) and along the core MD99-2346 (42°02.61′N, 4°09.04′E, 2100 m of water depth, 10.61 m long; Fig. 1) collected in the Gulf of Lion. The MD99-2346 core, which is 10.63 m long, is well dated using accelerator mass spectrometry (AMS) ¹⁴C dating. The age model of core MD99-2346, for the last 28 kyr, has been established using AMS ¹⁴C dates from 13 planktonic foraminifera levels (Melki et al., 2009). For this core, in addition to micropaleontological analysis, which have been used for reconstructions of past sea surface temperatures (SST), the oxygen stable isotope analyses measured on the G. bulloides (δ¹⁸O) are available (Melki et al., 2009).

Oxygen isotopic (δ¹⁸O) measurements were performed on G. bulloides specimens along the core MD99-2346 and on the available western Mediterranean core tops. About six foraminifer tests (60 μg) per sample were picked from the 200–250 μm size fraction for each oxygen isotopic measurement. Shells were ultrasonically cleaned in a methanol bath to remove clays and other impurities. They were roasted under vacuum at 380 °C during 45 min to eliminate organic matter. Samples were analyzed with a Finnigan Δ+ mass spectrometer. All results are expressed as δ¹⁸O in % versus PDB via the calibration with respect to the international standards NBS-19 and NBS-18. The analytical reproducibility as determined from replicate measurements of an internal standard is ± 0.05% (1σ).

About 20–30 specimens of G. bulloides per sample were used from the 200–250 μm size fraction for each Mg/Ca measurement. These shells were gently crushed to open the chambers. Prior to analysis, samples were cleaned following the procedure of Barker et al. (2003), which includes several ultrasonic treatments in water then ethanol to remove adhesive clays, and then a reaction with hydrogen peroxide treatment 1% at 100 °C to remove any organic matter. A very slight acid leaching with 0.001 M nitric acid was finally applied to eliminate contaminants adsorbed on foraminiferal tests. Analyses were performed on a Varian Vista Pro Inductively Coupled Plasma Atomic Emission Spectrometer (ICP-AES) at the Laboratoire des sciences du climat et de l’environnement (LSCE), a member of an interlaboratory comparison study of calibration standards for foraminiferal Mg/Ca thermometry (Greaves et al., 2008), following the method of de Villiers et al. (2002). The mean external reproducibility of a standard solution of Mg/Ca = 5.23 mmol.mol⁻¹ ±0.5% (RSD).

On average, clay minerals contain between 1 and 10 weight % Mg (Deer et al., 1992). Clay contamination may cause, therefore, a significant bias in foraminifer Mg/Ca ratios. Barker et al. (2003) showed that the covariances of Fe/Ca and Mg/Ca, as well as Fe/Mg values exceeding 0.1 mol.mol⁻¹ are indicators of clay contamination. All our cleaned samples showed a Fe/Mg ratio smaller than 0.1 mol.mol⁻¹, and neither Fe/Ca nor Al/Ca correlate with Mg/Ca.

This clearly shows the efficiency of the cleaning procedure in removing silicate contaminants.

Sea surface temperature reconstructions were performed using foraminiferal assemblages along the core MD99-2346 based on the modern analogue technique (Hutson, 1980; Overpeck et al., 1985; Prell, 1985; Prentice, 1980), Fig. 1. This technique is used to compare fossil planktonic foraminiferal samples to a modern reference database and to identify, for each fossil association, the ten best modern analogues. The mean degree of similarity between each fossil assemblage and the corresponding ten best modern analogues is measured by a dissimilarity coefficient. The SST estimates are considered reliable when this coefficient is lower than 0.25 (Prell, 1985). Above this threshold value, SST estimates have to be considered with caution. The reference database used for the present study has been first developed by Kallel et al. (1997a) and then improved by Hayes et al. (2005) and corresponds to 274 core-top sediments, 145 from the Mediterranean Sea and 129 from the North Atlantic Ocean. In the Mediterranean Sea, it has been found that modern analogues of foraminiferal fossil assemblages are found in the Mediterranean Sea itself during warm periods of the Upper Quaternary. However, during glacial periods, the best modern analogues originate mainly from the North Atlantic (Essallami et al., 2007; Kallel et al., 1997b, 2000; Melki et al., 2009).

By combining SST estimates derived from foraminifera using modern analogue technique and the foraminiferal δ¹⁸O values, we reconstructed the Gulf of Lion surface salinity record (Fig. 1) following the method introduced by Duplessy et al. (1991) and Kallel et al. (1997a). Taking into account the SST estimates, the δ¹⁸O surface water variations were calculated by solving the paleotemperature equation (Shackleton, 1974).

In the Mediterranean Sea, surface water δ¹⁸O (δ_w) changes reflect both changes in the δ¹⁸O value of the incoming Atlantic water and variations of the local fresh-water budget (Precipitation plus continental Runoff minus Evaporation, P + R − E) of the Mediterranean basin (Kallel et al., 1997b). The effect of continental ice volume variations on global seawater δ¹⁸O is assumed to be equal to the sea-level curve of Lambeck and Chapell (2001) multiplied by a constant coefficient of 1.1‰/130 m from Waelbroeck et al. (2002). As the Mediterranean Sea is linked to the North Atlantic Ocean through the Straits of Gibraltar, we suppose that the effect of continental ice volume variations was recorded in a similar manner in the two areas. We therefore estimated the local Mediterranean Sea surface δ¹⁸O_sw variations with respect to the modern situation as the difference between calculated and modern seawater δ¹⁸O_sw, corrected for the global effect:δ¹⁸O_anomaly = calculated δ¹⁸O_sw − [modern δ¹⁸O_sw + global δ¹⁸Osignal].

To convert the δ¹⁸O anomaly to a salinity anomaly, we used the modern seawater δ¹⁸O_sw-salinity relationship for the western Mediterranean Sea (1‰ change in local salinity will result in 0.41‰ change in local δ¹⁸O_sw; Kallel et al., 1997a).

Core MD99-2346 exhibits four strong negative salinity anomalies (Fig. 3) during H2, H1, the Younger Dryas and the Middle Holocene period between about 10,000 and 4,000 yrs B.P. (3.5, 3.3, 1.7 and 1‰ respectively). The Gulf of Lion salinity decrease events are similar to those observed in other Mediterranean cores recovered in the Alboran Sea (Cacho et al., 1999), in the Tyrrhenian Sea (Kallel et al., 1997b, Paterne et al., 1999) and in the Sicilian–Tunisian Strait (Essallami et al., 2007).

To understand better the origin of these salinity changes, Melki et al. (2009) compared them with those obtained from the core SU81-18 recovered in the North Atlantic Ocean off Portugal (Duplessy et al., 1992, 1993). They found that the first three events were associated with a similar salinity decrease of the incoming Atlantic water entering the Mediterranean Sea in the upper part of the Gibraltar Strait, but the western Mediterranean freshwater balance was not affected at these epochs. The basin acted as today as a concentration basin (P + R < E), whereas the Holocene event, absent in the North Atlantic Ocean, was found to be of local origin. It indicates an increase of precipitation over the whole Mediterranean region.

Consequently, the effect of the deglaciation melt water in the Gulf of Lion was mainly transferred from the North Atlantic via the Gibraltar Strait, whereas the Alpine glaciers contribution in the last deglaciation could have been potentially significant only for the northern Mediterranean sites (Stocchi et al., 2005). It was restricted to the Adriatic Sea and the Gulf of Genoa (Lambeck et al., 2004).

Values of δ¹⁸O_foraminifera measured on G. bulloides from the Mediterranean core tops (Fig. 1) and their corresponding calcification temperatures (Tiso) are summarized in Table 1. The isotopic temperature (Tiso) was calculated using the paleotemperature equation of Shackleton (1974) by combining these δ¹⁸O_f measurements with the oxygen isotopic ratio of seawater (δ¹⁸O_sw) extracted from a gridded atlas (LeGrande and Schmidt, 2006) at the appropriate location of each core. Results for Mg/Ca measured on G. bulloides are also shown in Table 1. The Mg/Ca of G. bulloides were converted to temperatures (SST_Mg/Ca) based on Elderfield and Ganssen (2000) (Mg/Ca = 0.81 e^0.081T) empirical equation for G. bulloides. Using this Mg/Ca-temperature empirical calibration results in surface temperatures (SST_Mg/Ca) that are systematically higher than the isotopic temperatures (Tiso).

Under modern conditions, G. bulloides develops preferentially during the spring bloom in the Mediterranean Sea (Pujol and Vergnaud-Grazzini, 1995). This is demonstrated, for instance, by the strong correlation over all the Mediterranean Sea, between modern April-May SSTs (Levitus, 1982) and isotopic temperatures estimated using δ¹⁸O values of G. bulloides (Kallel et al., 1997a).

However, if we focus only on data from the western Mediterranean Sea, particularly in the Gulf of Lion (Table 1), we note that isotopic temperatures estimated using δ¹⁸O values of G. bulloides from core tops average ∼ 16.5 °C, and are, therefore, ∼ 2 °C warmer than April-May SST (Atlas WOA5). In this area, the planktonic spring bloom is mainly linked to the start of the vertical stratification of the water column associated with the water warming after winter overturning. As the vertical winter overturning is particularly intense in the Gulf of Lion, it is likely that the stratification and development of G. bulloides spread over a longer period and/or start later than in the other areas of the Mediterranean Sea. Thus, in the Gulf of Lion, the best correlation with the isotopic temperatures estimated using δ¹⁸O values of G. bulloides is obtained with the mean spring (April, May, June) SST, and not with April-May SST.

Past April-May-June SST (SST_MAT) were estimated along the core MD99-2346 from the counts of planktonic foraminifera published by Melki et al. (2009), using the modern analogue technique. These temperatures estimates are shown in Fig. 2A. The dissimilarity coefficients associated with this SST reconstruction are generally lower than 0.25 except for the Last Glacial Maximum (Fig. 2B; between 21.09 and 18.07 cal kyr BP.) when they can reach 0.3 (Fig. 2C). All estimates were obtained with low standard deviations (not exceeding 1.25 °C).

The high-resolution SST_Mg/Ca record for the last 28 kyr, can be compared to that obtained using modern analogue technique applied to fossil foraminiferal assemblages (SST_MAT) (Fig. 2A). Records display similar patterns. In both cases, the last glaciation is colder than the Upper Holocene by about 7 °C and the post-glacial warming of the Bollïng/Alleröd (B/A), which began at about 14.7 cal kyr BP., was interrupted by a climatic deterioration and an abrupt cooling at the same epoch as the European and North Atlantic Younger Dryas (YD) between about 12.8 and 11.5 kyr BP. (Bard et al., 1987; Duplessy et al., 1991; Rasmussen et al., 2006). This pattern of temperature change is similar to that observed in other Mediterranean areas such as the Alboran Sea, the Tyrrhenian Sea, the Sicilian-Tunisian Strait and the eastern Mediterranean basin (Cacho et al., 1999; Emeis et al., 2000, 2003; Essallami et al., 2007; Kallel et al., 1997b; Paterne et al., 1999).

Significant differences between the two records of the Gulf of Lion are observed, however, during the Younger Dryas and the Lower Holocene. The YD cooling phase is shorter in the Mg/Ca record (of 12.93 at 11.49 kyr BP.) than in the modern analogue technique record (of 13.45 to 11.49  kyr BP.). The timing of the SST_Mg/Ca cooling is more consistent with that defined in the North Atlantic for the Younger Dryas (12.8 to 11.7 kyr BP.; Bard et al., 2000; Heinrich, 1988; Rasmussen et al., 2006). Moreover, the foraminiferal record indicated that SSTs were higher at the beginning of the Holocene than the Upper Holocene SSTs by about 2 to 3 °C. Such a structure is not as marked on the Mg/Ca record.

More importantly, Fig. 2A clearly indicates that while the structures of two temperature records are broadly similar; the SST_Mg/Ca estimates are, however, several degrees warmer than SST_MAT. The shift is constant during the last glaciation and Holocene but changes significantly through time during the last deglaciation. This is probably due to the offset of the biological responses with respect to hydrological changes in temperature. Thus, in order to determine the average offset between the two temperature records, we suggest to retrieve the deglaciation interval, and we calculated the mean differences for the interval 0–8 kyr BP. (Holocene) and 15–25 kyr BP. (glacial period). Over the Holocene, the temperature offset is about 3.8 °C, and is +4 °C for the glacial period.

A recent work pointed out that anomalously high Mg/Ca ratios measured by conventional ICP-AES analysis of planktonic foraminifers from the Mediterranean Sea (and in particular in the eastern basin) are mainly related to the presence of early diagenesis, high Mg-calcite overgrowths (10–12%) (Boussetta et al., 2011; Hoogakker et al., 2009; van Raden et al., 2011). However, along the core MD99-2346, an interpretation of the XRD profiles, deconvoluted using the IGOR^® software, did not reveal any measurable amount of Mg-rich calcite on G. bulloides shells. If it exists, in trace amounts, it is below the XRD detection-limit (Sabbatini et al., 2011). Nevertheless, the deposition of secondary inorganic calcite containing 10–12% of Mg, even in small quantities, has the potential to change noticeably Mg/Ca values obtained on whole planktonic shells (which contain only a few mmol mol-1 Mg/Ca; Sexton et al. (2006)).

In the eastern Mediterranean basin, where the diagenetic coating appears to be generally more important, Boussetta (2011) showed that there is considerable local variability in its amount even at the sample scale. Unpublished data, dealing with the evolution of the diagenetic coating along a core recovered in the Levantine basin, show that the amount of diagenetic calcite also varies over time (Boussetta, 2011, PhD thesis). Such variability in the Mg-rich calcite encrustation, both spatial and temporal, is not compatible with the relative constancy, within the rather stable time period of the Holocene and the LGM, of the temperature shift between SST _Mg/Ca and SST _MAT that we note in the core MD99-2346. We can therefore reasonably conclude that the presence of a diagenetic coating cannot account for this systematic Mg enrichment observed in core MD99-2346.

The Mediterranean Sea is a semi-enclosed basin showing a strong salinity gradient from west (∼36 psu near the Strait of Gibraltar) to east (∼40 psu) (Boyer et al., 2002; Lascaratos et al., 1999). These salinities are higher than the average for the World Ocean. Recently, several studies have emphasized the potential role of high salinity as a bias on Mg/Ca thermometry, with sensitivities ranging from 4 ± 3% Mg/Ca per psu unit observed in culture studies (Kisakürek et al., 2008; Lea et al., 1999; Nürnberg et al., 1996) up to 15–59% per psu suggested by field studies on Mediterranean core top samples (Ferguson et al., 2008).

In order to evaluate the possibility that Mg/Ca temperature anomalies from the western Mediterranean Sea are affected by surface salinity, we reconstructed the Gulf of Lion surface salinity record, by combining SST estimates derived from foraminifera using modern analogue technique and the foraminiferal δ¹⁸O values, following the method introduced by Duplessy et al. (1991) and Kallel et al. (1997a). Then, these salinities are compared with the potential shift (ΔT) between Mg/Ca temperatures (SST_Mg/Ca) and inferred temperatures using modern analogue technique (SST_MAT).

As can be seen (Fig. 3), the surface salinity record obtained from core MD99-2346 reveals large-scale fluctuations that are neither correlated with Mg/Ca temperatures nor with the shift (ΔT) between SST_Mg/Ca and SST_MAT. This is particularly visible during the cold periods synchronous to the Heinrich events 2 and 1, during the Younger Dryas and during the Lower to the Middle Holocene. Abrupt and strong salinity decrease events of 2 to 3‰ are observed during these periods (Melki et al., 2009) and they are not reflected by corresponding changes in Mg/Ca temperatures or in the shift relative to SST_MAT.

To confirm the hypothesis that salinity does not play any dominant role in the Mg uptake into Mediterranean foraminiferal tests, we looked if a salinity effect can be detected from Mg/Ca measured in G. bulloides core tops samples (Boussetta et al., 2011) and from plankton tows (van Raden et al., 2011) recovered from the western Mediterranean Sea.

In the western Mediterranean basin, SEM observations and X-Ray Diffractometry Analyses indicate that the post-mortem calcite overgrowth is minor (Boussetta et al., 2011) and that the most of this calcite being likely removed during the chemical cleaning for Mg/Ca analyses (Sabbatini et al., 2011).

For these samples we estimated the difference (Δ Mg/Ca) between the measured Mg/Ca ratio (Mg/Ca_measured) and that expected (Mg/Ca_T) from the calibration of Elderfield and Ganssen (2000), by the use of estimated temperatures from oxygen isotope composition of G. bulloides. The residuals (Δ Mg/Ca) were then compared with salinity changes (Fig. 4). This figure clearly shows that there is no significant correlation between salinity and plankton tow or core top sample Δ Mg/Ca supporting our conclusion that the Mg/Ca enrichment of G. bulloides along core MD99-2346 cannot be explained by salinity.

The Mg/Ca-temperature calibration for G. bulloides that we used in this study has been established by Elderfield and Ganssen (2000) using surface sediment material from open ocean, particularly the North Atlantic Ocean on a latitudinal transect from, 30° to 60°N at about 25°W. It should be noted that Elderfield and Ganssen (2000) have removed from their database several G. bulloides Mg/Ca values, which appeared “unusually high” compared to values obtained in the same surface temperature conditions on other planktonic foraminiferal species. These authors observed that Mg-enriched samples of G. bulloides are mainly found on shallower core tops, but they do not provide a clear explanation about why only the G. bulloides species appears to show such anomalous Mg/Ca values. We believe that the existence of anomalously high G. bulloides Mg/Ca ratios clearly suggest the problem of developing a Mg/Ca thermometry on this species, although — at this stage — no clear explanation exists about its cause.

This Mg enrichment seems, therefore, to be a signal acquired during the development of G. bulloides and not at the sediment surface. This interpretation is supported by the data obtained by van Raden et al. (2011) showing that the high Mg/Ca values also characterize G. bulloides shells retrieved from plankton tows in the western Mediterranean. The high calcite saturation state of Mediterranean surface waters cannot explain this Mg enrichment. Culture experiments on planktonic foraminifera, including G. bulloides, have shown that less Mg rather than more Mg is incorporated into the tests during calcification at higher saturation states (Kisakürek et al., 2008; Lea et al., 1999; Russell et al., 2004).

Two hypotheses will have to be tested in the future, but unfortunately it cannot be done with the currently available data set:

• 1. is the empirical equation obtained by Elderfield and Ganssen (2000) incorrect because it should have taken into account the G. bulloides Mg/Ca values considered by these authors as “abnormally” high?;
• 2. is the Mediterranean G. bulloides associated to a particular genotype that has developed a specific process of biomineralization characterized by high Mg uptake (vital effect)?

Darling et al. (2003) demonstrate, in the Santa Barbara Channel (Pacific Ocean), that the same genotype of G. bulloides (type IId) occurs in all the changing hydrographic regimes associated with this region throughout the annual cycle with the exception of January, when they recorded the additional presence of the high-latitude G. bulloides type IIa. On the other hand, it has been shown that different morphotypes of G. ruber, which are most likely representative of different genotypes have distinct Mg/Ca (Steinke et al., 2005). Consequently, genotype difference, if it exists, between G. bulloides individuals used in the Atlantic calibration and those analysed in the Mediterranean Sea might be responsible for the observed anomalously high Mg/Ca in the Mediterranean Sea.

Deciphering the origin of this consistent Mg enrichment of G. bulloides shells in the Mediterranean Sea requires further investigations, based on modern material.