Sea surface temperature variability in the central-western Mediterranean Sea during the last 2700 years: A multi-proxy and multi-record approach

. This study presents the reconstructed evolution of sea surface conditions in the central-western Mediterranean Sea during the late Holocene (2700 years) from a set of multi-proxy records as measured on ﬁve short sediment cores from two sites north of Minorca (cores MINMC06


Introduction
The Mediterranean is considered one of the most vulnerable regions with regard to the current global warming (Giorgi, 2006). This high sensitivity to climate variability has been evidenced in several studies on past natural changes (Rohling et al., 1998;Cacho et al., 1999a;Moreno et al., 2002;Martrat et al., 2004;Reguera, 2004;Frigola et al., 2007;Combourieu Nebout et al., 2009). Palaeo-studies focussed mostly on the rapid climate variability in the last glacial period have shown solid evidence of a close connection between changes in North Atlantic oceanography and climate over the western Mediterranean region (Cacho et al., 1999b(Cacho et al., , 2000(Cacho et al., , 2001Moreno et al., 2005;Sierro et al., 2005;Frigola et al., 2008;Fletcher and Sanchez-Goñi, 2008). Nevertheless, climate variability during the Holocene, and particularly during the last millennium, is not so well described in this region, although its understanding is crucial for placing the nature of the 20th century trends in the recent climate history (Huang, 2004).
Some previous studies have already proposed that the Holocene centennial climate variability in the western Mediterranean Sea could be linked to the North Atlantic Oscillation (NAO) variability (Jalut et al., 1997(Jalut et al., , 2000Combourieu Nebout et al., 2002;Goy et al., 2003;Roberts et al., 2012;Fletcher et al., 2012). In particular, nine Holocene episodes of enhanced deep water convection in the Gulf of Lion (GoL) and surface cooling conditions have been described in the region (Frigola et al., 2007). These events have also been correlated to intensified upwelling conditions in the Alboran Sea and tentatively described as two-phase scenarios driven by distinctive NAO states (Ausín et al., 2015).
The existing Mediterranean climatic records for the last 1 or 2 kyr are mostly based on terrestrial archives such as tree rings (Touchan et al., 2005(Touchan et al., , 2007Griggs et al., 2007;Esper et al., 2007;Büntgen et al., 2011;Morellón et al., 2012), speleothem records (Frisia et al., 2003;Mangini et al., 2005;Fleitmann et al., 2009;Martín-Chivelet et al., 2011;Wassenburg et al., 2013), or lake reconstructions (Pla and Catalan, 2005;Martín-Puertas et al., 2008;Corella et al., 2011;Morellón et al., 2012). All of these archives can be good sensors of temperature and humidity changes, but it is often difficult to disentangle the effect of both variables in the proxy records. Recent efforts focussed on integrating these 2 kyr records into regional climatic signals reveal complex regional responses and evidence the scarcity of marine records to form a more complete picture (PAGES, 2009;Lionello, 2012).
Marine records are often limited by the lack of adequate time resolution and/or robust chronologies for detailed comparison with terrestrial records. However, marine records provide a wider range of temperature-sensitive proxies. There are few marine palaeoclimate records available for the last 2 kyr in the Mediterranean Sea (Schilman et al., 2001;Versteegh et al., 2007;Piva et al., 2008;Taricco et al., 2009Taricco et al., , 2015Incarbona et al., 2010;Fanget et al., 2013;Grauel et al., 2013;Lirer et al., 2013Lirer et al., , 2014Di Bella et al., 2014;Goudeau et al., 2015) and they are even more scarce in the western basin. Unfortunately, the existing pool of marine proxy data in the Mediterranean for the last two millennia is too sparse to recognize common patterns of climate variability (Taricco et al., 2009;Nieto-Moreno et al., 2011;Moreno et al., 2012, and references therein).
The aim of the present study is to characterize changes in surface water properties from the Minorca margin in the Catalan-Balearic Sea (central-western Mediterranean) in order to contribute to a better understanding of the climate variations in this region during the last 2.7 kyr. Sea surface temperature (SST) has been reconstructed by means of two independent proxies, Mg / Ca analyses on the planktonic foraminifera Globigerina bulloides and alkenone-derived SST (Villanueva et al., 1997;Lea et al., 1999;Barker et al., 2005;Conte et al., 2006). The application of G. bulloides Mg / Ca as a palaeothermometer in the western Mediterranean Sea is tested through the analysis of a series of core-top samples from different locations of the western Mediterranean Sea and the calibration reviewed consistently. Mg / Ca thermometry is applied in conjunction with δ 18 O in order to evaluate changes in the evaporation-precipitation (E-P) balance of the basin, which are ultimately linked to salinity Pierre, 1999;Barker et al., 2005).
One of the intrinsic limitations of studying the climate evolution of the last 2 kyr is that the magnitude of climatic oscillations is often below the sensitivity of the selected proxies. In order to overcome this limitation we have produced "stack" proxy records from multicores in the same region. The stack record captures the first-order climatic variability from the proxy records and removes the noise, therefore allowing for a more robust identification of regional climatic variability.
The studied time periods have been defined as follows (years expressed as BCE, before common era, and CE, common era): the Talaiotic Period (TP, ending in 123 BCE), Roman Period (RP, from 123 BCE to 470 CE), "Dark Middle Ages" (DMA; from 470 until 900 CE), Medieval Climate Anomaly (MCA, from 900 to 1275 CE), and Little Ice Age (LIA, from 1275 to 1850 CE), with the Industrial Era (IE) as the most recent period. The limits of these periods are not uniform across the Mediterranean (Lionello, 2012), and here the selected ages have been chosen according to historical  events in Minorca and to the classic climatic ones defined in the literature (i.e. Nieto- Moreno et al., , 2013Moreno et al., 2012;Lirer et al., 2013Lirer et al., , 2014.

Climatic and oceanographic settings
The Mediterranean Sea is a semi-enclosed basin located in a transitional zone between different climate regimes, from the temperate zone in the north to the subtropical zone in the south. Consequently, the Mediterranean climate is characterized by mild wet winters and warm to hot, dry summers (Lionello et al., 2006). Interannual climate variability is very much controlled by the dipole-like pressure gradient between the Azores (high) and Iceland (low) system, known as the North Atlantic Oscillation (NAO; Hurrell, 1995;Lionello and Sanna, 2005;Mariotti, 2011;Ausín et al., 2015). However, the northern part of the Mediterranean region is also linked to other mid-latitude teleconnection patterns (Lionello, 2012). The Mediterranean Sea is a concentration basin (Béthoux, 1980;Lacombe et al., 1981) and the excess of evaporation with respect to freshwater input is balanced by water exchange at the Strait of Gibraltar (i.e. Pinardi and Masetti, 2000;Malanotte-Rizzoli et al., 2014). The basin-wide circulation pattern is predominantly cyclonic (Millot, 1999). Three convection cells promote the Mediterranean deep and intermediate circulation: a basin-wide open cell and two separated closed cells, one for the western part of the basin and one for the eastern part. The first one connects the two basins of the Mediterranean Sea though the Strait of Sicily, where water masses interchange occurs at intermediate depths. This cell is associated with the inflow of Atlantic Water (AW) at the Strait of Gibraltar and the outflow of the Levantine Intermediate Water (LIW) that flows below the first (Lionello et al., 2006).
In the north-western Mediterranean Sea, the Northern Current (NC) represents the main feature of the surface circulation transporting waters alongshore from the Ligurian Sea to the Alboran Sea (Fig. 1a). North-east of the Balearic Promontory a surface oceanographic front separates Mediterranean waters transported by the NC from the Atlantic waters that recently entered the Mediterranean (Millot, 1999;Pinot et al., 2002;André et al., 2005).
Deep convection occurs offshore of the GoL due to the action of persistent cold and dry winter winds such as the tramontana and the mistral. These winds cause strong evaporation and cooling of surface water, thus increasing their density, sinking to greater depths and leading to Western Mediterranean Deep Water formation (WMDW; MEDOC, 1970;Lacombe et al., 1985;Millot, 1999). Dense shelf water cascading (DSWC) in the GoL also contributes to the sink of large volumes of water and sediments into the deep basin (Canals et al., 2006).
The north-western Mediterranean primary production is subject to an intense bloom in late winter-spring, when the surface layer stabilizes, and sometimes to a less intense bloom in autumn, when the strong summer thermocline is progressively eroded (Estrada et al., 1985;Bosc et al., 2004;D'Ortenzio and Ribera, 2009;Siokou-Frangou et al., 2010). SST in the region evolves accordingly with the seasonal bloom, with minima SST in February, which subsequently 852 M. Cisneros et al.: Sea surface temperature variability in the Mediterranean Sea increases until maximum SST values during August. Afterwards, a SST drop can be observed in October, although with some interannual variability (Pastor, 2012).

Sediment core description
The studied sediment cores were recovered from a sediment drift built by the action of the southward branch of the WMDW north of Minorca (Fig. 1). Previous studies carried out at this site have already described high sedimentation rates (> 20 cm kyr −1 ; Frigola et al., 2007Frigola et al., , 2008Moreno et al., 2012), suggesting that this location was suitable for a detailed study of the last millennia. The cores were recovered with a multicore system in two different stations located at about 50 km north of Minorca.  (Frigola, 2012;Velasco et al., 1996;Mauffret, 1979). The MIN cores are from sites that are about 300 m deeper than the MR3 ones.
MIN cores were homogeneously sampled at 0.5 cm resolution in the laboratory. In the MR3 cores a different strategy was followed. MR3.1 and MR3.2 were initially subsampled with a PVC tube and split into two halves for X-ray fluorescence (XRF) analyses in the laboratory. Both halves of core MR3.1 (MR3.1A and MR3.1B) were used for the present work as replicates of the same core, and records for each half are shown separately. All MR3 cores were sampled at 0.5 cm resolution in the upper 15 cm and at 1 cm in the deeper sections, with the exception of MR3.1B that was sampled at 0.25 cm resolution. The MR3 cores were composed of brown-orange nannofossil and foraminifera silty clay, which was lightly bioturbated and contained layers enriched in pteropods and fragments of gastropods as well as some dark layers.
Additionally, core-top samples from seven multicores collected at different locations in the western Mediterranean have also been used for the correction of the Mg / Ca-SST calibration from G. bulloides (Table 1; Fig. 1).

Radiocarbon analyses
Twelve 14 C AMS dates were measured in cores MIN1, MIN2, and MR3.3 (Supplement Table S1) using 4-22 mg samples of the planktonic foraminifer Globorotalia inflata handpicked from the > 355 µm fraction. The ages were calibrated with the standard marine correction of 408 years and the regional average marine reservoir correction ( R) for the central-western Mediterranean Sea using Calib 7.0 software (Stuiver and Reimer, 1993) and the MARINE13 calibration curve (Reimer et al., 2013).

Radionuclides 210 Pb and 137 Cs
The concentrations of the naturally occurring radionuclide 210 Pb (Supplement Fig. S1) were determined in cores MIN1, MIN2, MR3.1A, and MR3.2 by alpha spectroscopy (Sanchez-Cabeza et al., 1998). The concentrations of the anthropogenic radionuclide 137 Cs in core MIN1 (Fig. S1) were measured by gamma spectrometry using a high-purity intrinsic germanium detector. The 226 Ra concentrations were determined from the gamma emissions of 214 Pb that were also used to calculate the excess 210 Pb concentrations. The sediment accumulation rates for the last century (Sect. S1.1 in the Supplement) were calculated using the CIC (constant initial concentration) and the CF : CS (constant flux : constant sedimentation) models (Appleby and Oldfield, 1992;Krishnaswami et al., 1971), constrained by the 137 Cs concentration profile in core MIN1 (Masqué et al., 2003).

Bulk geochemical analyses
The elemental composition of cores MR3.1B and MR3.2 was obtained with an Avaatech XRF core-scanner system (CORELAB, University of Barcelona), which is equipped with an optical variable system that allows determining the length (10-0.1 mm) and the extent (15-2 mm) of the bundle of X-rays in an independent way. This method allows obtaining qualitative information of the elementary composition of the core materials. The core surfaces were scraped, cleaned, and covered with a 4 µm thin SPEXCertiPrep Ultralene foil to prevent contamination and minimize desiccation (Richter and van der Gaast, 2006). Sampling was performed every 1 cm and scanning took place at the split core surface directly. Among the several elements measured in this study, the Mn profile was used for the construction of the age models (see Supplement for age model development).

Planktonic foraminiferal analyses
Planktonic foraminifera specimens of Globigerina bulloides were picked together from a size range of 250-355 µm, crushed, and cleaned separately for Mg / Ca and δ 18 O measurements. In core MR3.1B, picking was often performed in the < 355 µm fraction due to the small amount of material (sampling every 0.25 cm). Additionally, quantitative analysis of planktonic foraminiferal assemblages was carried out in core MR3.3 and on the upper part of core MR3.1A by using the fraction size above 125 µm (Fig. S2). The 42 studied samples showed abundant and well-preserved planktonic foraminifera.
The samples for trace elements analyses consisted of ∼ 45 specimens of G. bulloides that were crushed under glass slides to open the chambers. Foraminifera cleaning consisted of clay removal and oxidative and weak acid leaching steps . Samples from core MR3.1A were also cleaned including the "reductive step". Elemental ratios were measured on an inductively coupled plasma mass spectrometer (ICP-MS, Perkin Elmer ELAN 6000) in the Scientific and Technological Centers of the University of Barcelona (CCiT-UB). A standard solution with known elemental ratios was used for sample standard bracketing (SSB) as a correction for instrumental drift. The average reproducibility of Mg / Ca ratios, taking into account the known standard solutions concentrations, was 97 and 89 % for MIN1 and MIN2 cores, and 99 and 97 % for cores MR3.1A, MR3.1B, and MR3.3, respectively.
Procedural blanks were routinely measured to detect any potential contamination problem during cleaning and dissolution. The Mn / Ca and Al / Ca ratios were also always measured to identify potential contaminations due to the presence of manganese oxides and/or aluminosilicates (Barker et al., 2003;Lea et al., 2005;Pena et al., 2005Pena et al., , 2008. To avoid the overestimation of Mg / Ca-SST by diagenetic contamination, Mn / Ca values > 0.5 mmol mol −1 were discarded from core MR3.1B and only those higher than 1 mmol mol −1 were removed from MIN1 and MR3.3. Samples suspected to have detrital contamination with elevated Al / Ca ratios were also removed. No significant correlation exists between Mg / Ca and Mn / Ca or Al / Ca ratios after data filtering (r < 0.29, p value = 0.06).
The Mg / Ca ratios were transferred into SST values using the calibration proposed in the present study (Sect. 5.1). In the case of the MR3.1A record, which was cleaned using the reductive procedure, and as was expected (Barker et al., 2003;Pena et al., 2005;Yu et al., 2007), the Mg / Ca ratios were about 23 % lower than those measured in core MR3.1B without the reductive step. The obtained percentage of Mg / Ca lowering is comparable to or higher than those percentages previously estimated for different planktonic foraminifera, although data from G. bulloides have not been previously reported (Barker et al., 2003). Mg / Ca-SST in core MR3.1A was calculated after the Mg / Ca correction of this 23 % offset by application of the calibration used with the other records.
Stable isotope measurements were performed by means of sonication on 10 specimens of G. bulloides after methanol cleaning to remove fine-grained particles. The analyses were performed in a Finnigan MAT 252 mass spectrometer fitted with a Kiel-IV carbonate microsampler in the CCiT-UB. The analytical precision of laboratory standards for δ 18 O was better than 0.08 ‰. Calibration to Vienna Pee Dee Belemnite (VPDB) was carried out by means of NBS-19 standards (Coplen, 1996).
Seawater δ 18 O (δ 18 O sw ) was obtained after removing the temperature effect on the G. bulloides δ 18 O signal using the Mg / Ca-SST records of the Shackleton palaeotemperature equation (Shackleton, 1974). The results are expressed in the SMOW (Standard Mean Ocean Water) water standard (δ 18 O sw ) after the correction of Craig (1965). The use of specific temperature equations for G. bulloides was also considered (Bemis et al., 1998;Mulitza et al., 2003), but the core-top estimates provided δ 18 O sw values of 2.1-1.5 ‰ SMOW, which were significantly higher than those measured in water samples from the central-western Mediterranean Sea (∼ 1.2 ‰ SMOW) (Pierre, 1999). After application of the empirical Shackleton (1974) palaeotemperature equation, the core-top δ 18 O sw estimates averaged 1.1 ‰ SMOW and were closer to the actual seawater measurements. This, it was decided that this equation provided more realistic oceanographical conditions in this location.

Alkenones
Measurements of the relative proportion of unsaturated C 37 alkenones, namely the U k 37 index, were carried out in order to obtain SST records for the studied cores. Detailed in-formation about the methodology and equipment used can be found in Villanueva et al. (1997). The precision of this palaeothermometry tool has been determined to be about ±0.5 • C (Eglinton et al., 2001). Furthermore, taking into account duplicate alkenone analysis carried out on core MR3.3, the precision achieved results better than ±0.8 • C. The reconstruction of SST records was based on the global calibration of Conte et al. (2006), which considers an estimation standard error of 1.1 • C in surface sediments.

Age model development
Obtaining accurate chronologies for each of the studied sediment cores is particularly critical to allow intercomparison and produce a stack record that represents the regional climatic signal. With this objective, a wide set of parameters have been combined in order to obtain chronological markers in all the studied sedimentary records, including absolute dates and stratigraphical markers based on both geochemical and micro-palaeontological data (Tables S2 and S4). The methodology of age model development is explained in detail in the Supplement.

Mg / Ca-SST calibration
The Mg / Ca ratio measured in G. bulloides is a widely used proxy to reconstruct SST (Barker et al., 2005), although the calibrations available can provide very different results Mashiotta et al., 1999;Elderfield and Ganssen, 2000;Anand et al., 2003;McConnell and Thunell, 2005;Cléroux et al., 2008;Thornalley et al., 2009;Patton et al., 2011). Apparently, the regional Mg / Ca-temperature response varies due to parameters that have not yet been identified (Patton et al., 2011). A further difficulty arises from the questioned Mg / Ca thermal signal in high-salinity regions such as the Mediterranean Sea, where anomalously high Mg / Ca values have been observed (Ferguson et al., 2008). This apparent high salinity sensitivity in foraminiferal Mg / Ca ratios is under discussion and has not been supported by recent culture experiments (Hönisch et al., 2013), which, in addition, could be attributed to diagenetic overprints (Hoogakker et al., 2009;van Raden et al., 2011). In order to test the value of the Mg / Ca ratios in G. bulloides from the western Mediterranean Sea and also review its significance in terms of seasonality and depth habitat, a set of coretop samples from different locations of the western Mediterranean Sea have been analysed. Core-top samples were recovered using a multicorer system, and they can be considered representative of present conditions (Masqué et al., 2003;Cacho et al., 2006). The studied cores are located in the 35-45 • N latitude range (Table 1 and Table 1) is ±0.6 • C. Error of temperature estimates based on our G. bulloides calibration for the western Mediterranean is ±1.4 • C. These uncertainties are higher but still in the range of ±0.6 • C obtained for the Atlantic Ocean in Elderfield and Ganssen (2000) and also 1.1 • C in G. bulloides culture data . sical spring bloom (the most north-western basin) and an intermittent bloom (D'Ortenzio and Ribera, 2009). The resulting Mg / Ca ratios have been compared with the isotopically derived calcification temperatures based on the δ 18 O measurements performed also in G. bulloides from the same samples. This comparison was performed after use of the Shackleton (1974) palaeotemperature equation and the δ 18 O water data published by Pierre (1999), always considering the values of the closer stations and the top 100 m. The resulting Mg / Ca-SST data have been plotted together with those of G. bulloides from North Atlantic core tops previously published by Elderfield and Ganssen (2000). The resulting high correlation (r 2 = 0.92; Fig. 2a) strongly supports that the Mg / Ca ratios of the central-western Mediterranean Sea are dominated by a thermal signal. Thus, the new data set from the Mediterranean core tops improves temperature sensitivity range over the warm end of the calibration. The resulting exponential function indicates ∼ 9.4 % Mg / Ca per • C sensitivity in the Mg uptake with respect to temperature, which is in agreement with the range described in the literature (i.e., Elderfield and Ganssen, 2000;Barker et al., 2005;Patton et al., 2011). The new equation for the Mg / Ca-SST calibration including data from the western Mediterranean Sea and the Atlantic Ocean is as follows: The Mg / Ca-SST signal of G. bulloides has been compared with a compilation of water temperature profiles of the first 200 m measured between years 1945 and 2000 in stations close to the studied core tops (MEDAR GROUP, 2002). Al-though significant regional and interannual variations have been observed, the obtained calcification temperatures of our core-top samples show the best agreement with temperature values of the upper 40 m during the spring months (April-May; Fig. 2b). This water depth is consistent with preferential depth ranges for G. bulloides found by plankton tows in the Mediterranean (Pujol and Vergnaud-Grazzini, 1995) and with results from multiannual sediment trap monitoring in the Alboran Sea and the GoL, where maximum G. bulloides percentages were observed just before the beginning of thermal stratifications (see Bárcena et al., 2004;Bosc et al., 2004;Rigual-Hernández et al., 2012). Although the information available about depth and seasonality distribution of G. bulloides is relatively fragmented, this species is generally found in intermediate or even shallow waters (i.e. Bé and Hutson, 197;Ganssen and Kroon, 2000;Schiebel et al., 2002;Rogerson et al., 2004;Thornalley et al., 2009). However, G. bulloides has also been observed at deeper depths in some western Mediterranean Sea sub-basins (Pujol and Vergnaud-Grazzini, 1995). Extended data with enhanced spatial and seasonal coverage are required in order to better characterize production, seasonality, and geographic and distribution patterns of live foraminifera such as G. bulloides. Nevertheless, the obtained core-top data set offers solid evidence on the seasonal character of the recorded temperature signal in the Mg / Ca ratio.

A regional stack for Mg / Ca-SST records
The Mg / Ca-SST profiles obtained from our sediment records are plotted with the resulting common age model in Fig. 3. The average SST values for the last 2700 years ranged from 16.0 ± 0.9 to 17.8 ± 0.8 • C (uncertainties of average values represent 1σ ; uncertainties of absolute values include analytical precision and reproducibility and also those derived from the Mg / Ca-SST calibration). SST records show the warmest sustained period during the RP, approximately between 170 BCE and 300 CE, except in core MIN2, since this record ends at the RP-DMA transition. In addition, all the records show a generally consistent cooling trend after the RP with several centennial-scale oscillations. The maximum SST value is observed in core MR3.3 (19.6 ± 1.8 • C) during the MCA (Fig. 3c) and the minimum is recorded in core MIN1 (14.4 ± 1.4 • C) during the LIA (Fig. 3e). Centennial-scale variability is predominant throughout the records. Particularly, during MCA some warm episodes reached slightly higher SST than the averaged SST maximum (i.e. 19.6 ± 1.8 • C at ∼ 1021 CE). These events were far shorter in duration compared to the RP (Fig. 3). The highest frequency of intense cold events occurred during the LIA and, in particular, the last millennium recorded the minimum average Mg (g) 20 yr cm −1 stacked temperature anomaly (red plot) with its 2σ uncertainty (grey band). The 80 yr cm −1 (grey plot) and the 100 yr cm −1 (black plot) stacks are also shown. The triangles represent 14 C dates (black) and biostratigraphical dates based on planktonic foraminifera (blue) and are shown below the corresponding core, including their associated 2σ errors.
two subperiods, an early warmer interval (LIAa) and a later colder interval (LIAb) by reference to the 1540 CE boundary. One of the main difficulties with SST reconstructions in the last millennia is the internal noise of the records due to sampling and proxy limitations, which is of the same amplitude as the targeted climatic signal variability. In this sense, we have constructed a Mg / Ca-SST anomaly stack with the aim to detect the most robust climatic structures along the different records and reduce the individual noise. First, each SST record was converted into a SST anomaly record in relation to its average temperature (Fig. 3f). Secondly, in order to obtain a common sampling interval, all   records were interpolated. Interpolation at three different resolutions did not result in significant differences (Fig. 3g). Subsequently, we selected the stack that provided the best resolution offered by our age models (20 yr cm −1 ) since it very well preserves the high-frequency variability of the individual records (Fig. 3g). The obtained SST anomaly stack allows for a better identification of the most significant features at centennial timescales. Abrupt cooling events are mainly recorded during the LIA (−0.5 to −0.7 • C 100 yr −1 ), while abrupt warmings (0.9 to 0.6 • C 100 yr −1 ) are detected during the MCA. Events of similar magnitude have been also documented during the LIA-IE transition. When considering the entire SST anomaly record, a long-term cooling trend of about −1 to −2 • C is observed. However, focussing on the last 1800 years, since the RP maxima, the observed cooling trend was far more intense, at about −3.1 to −3.5 • C (−0.3 to −0.8 • C kyr −1 ). This is consistent within the recent 2 kyr global reconstruction published by McGregor et al. (2015; estimation of the SST cooling trend, using the average anomaly method 1 for the period 1-2000 CE: −0.3 to −0.4 • C kyr −1 ).

Oxygen isotope records
The oxygen isotopes measured on carbonate shells of G. bulloides (δ 18 O c ) and their derived δ 18 O sw after removing the temperature effect with the Mg / Ca-SST signals (see Sect. 3.5) are shown in Fig. 4. δ 18 O c and their derived δ 18 O sw profiles have been respectively stacked following the same procedure for the Mg / Ca-SST stack (Sect. 5.2). In general terms, all the records present a highly stable pattern during the whole period with a weak depleting trend, which is almost undetectable in some cases (i.e. core MIN1).
The average δ 18 O c values range from 1.2 to 1.4 ‰ VPDB and, in general, the MR3 cores show slightly higher values (∼ 1.4 ‰ VPDB) than the MIN cores (∼ 1.2 ‰ VPDB). The lowest δ 18 O c values (1.0 to 1.2 ‰ VPDB) mostly occur during the RP, although some short, low excursions can also be observed during the end of the MCA and/or the LIA. The highest values (1.4 to 1.8 ‰ VPDB) are mainly associated with short events during the LIA, the MCA, and over the TP-RP transition. A significant increase in δ 18 O c values is observed at the LIA-IE transition, although a sudden drop is recorded at the end of the stack record (after 1867 CE), which could result from a differential influence of the records (i.e. MIN1) and/or an extreme artefact (Fig. 4g).
After removing the temperature effect from the δ 18 O c record, the remaining δ 18 O sw record mainly reflects changes in E-P balance, thus resulting in an indirect proxy of sea surface salinity. The average δ 18 O sw values obtained for the period studied ranged from 1.3 to 1.8 ‰ SMOW. The highest δ 18 O sw values (from 2.4 to 1.9 ‰ SMOW) are recorded during the RP, when the longest warm period is also observed, and some values are notable during the MCA too. Enhancements of the E-P balance (δ 18 O sw higher values) coincide with higher SST. The lowest δ 18 O sw values (from 0.8 to 1.5 ‰ SMOW) are recorded particularly during the onset and the end of the LIA and also during the MCA. A drop in the E − P balance has been obtained from approximately the end of LIA to the most recent years. The most significant changes in our δ 18 O sw stack record correspond to increases in the most recent times and around 1200 CE (MCA) and to the decrease observed at the end of the LIA (Fig. 4).

Alkenone-SST records
The two alkenone (U k 37 )-derived SSTs of MIN cores have already been published in Moreno et al. (2012), while the records from MR3 cores are new (Fig. 5). The four alkenone-SST records show a similar general cooling trend during the studied period and they have also been integrated in a SST anomaly stack (Fig. 5e). The general cooling trend involves about −1.4 • C when the entire studied period is considered and about −1.7 • C since the SST maximum recorded during the RP. The mean SST uncertainties in this section have been estimated as ±1.1 • C, taking into account the estimated standard error (see Sect. 3.6). Previous studies have interpreted the alkenone-SST signal in the western Mediterranean Sea as an annual average (Ternois et al., 1996;Cacho et al., 1999a, b;Martrat et al., 2004). The average alkenone-SST values for the studied period (last 2700 years) ranged from 17.0 to 17.4 • C.
The coldest alkenone temperatures (∼ 16.0 • C) have been obtained in core MIN2 during the LIAa and the warmest (∼ 18.4 • C) in core MR3.3 during the MCA. Values near the average of maxima SST (from 17.9 to 18.4 • C) are observed more frequently during the TP, RP, and MCA, while temperatures during the onset of MCA and LIA show many values closer to the average of minima SST (ranging from 16.0 to 16.2 • C). Abrupt coolings are observed during the LIA and some events during MCA (−0.8 • C 100 yr −1 ) and to a lesser extent during the LIA-IE transition (−0.5 • C 100 yr −1 ). The highest warming rates are recorded during the MCA (0.4 • C 100 yr 1 ) and also during the RP.

Mg / Ca-SST vs. alkenone-SST records
In this section, the uncertainties of the alkenone, 1.1 • C, have been calculated from the estimated standard error of the calibration (see Sect. 3.6) and those of Mg / Ca-SST include the analytical precision and reproducibility and the standard error of the calibration. The measured Mg / Ca-SST and alkenone-SST averages are identical within error (16.9 ± 1.4 • C vs. 17.2 ± 1.1 • C), but the temperature range of the Mg / Ca records shows higher amplitude (see Sects. 5.2 and 5.4).
The enhanced Mg / Ca-SST variability is reflected in the short-term oscillations, at centennial timescales, which are larger in the Mg / Ca record with oscillations over 0.5 • C. This larger Mg / Ca-SST variability could be attributed to the highly restricted seasonal character of the signal, which purely reflects SST changes during the spring season. However, the coccolith signal integrates a wider time period from autumn to spring (Rigual-Hernández et al., 2012 and, consequently, changes associated with specific seasons become more diluted in the resulting averaged signal. The annual mean SST corresponding to a Balearic site is 18.7 ± 1.1 • C, according to the integrated values of the upper 50 m (Ternois et al., 1996;Cacho et al., 1999a) of the GCC-IEO database between January 1994 and July 2008. Our core-top records represent the last decades and show SST values closer to the annual mean in the case of alkenone-SST, whereas the Mg / Ca-SST record shows slightly lower values.
The U k 37 -SST records in the western Mediterranean Sea have been interpreted to represent annual mean SST (i.e. Cacho et al., 1999a;Martrat et al., 2004) but seasonal variations in alkenone production could play an important role in the U k 37 -SST values (Rodrigo-Gámiz et al., 2014). Considering that during the summer months the Mediterranean Sea is a very stratified and oligotrophic sea, reduced alkenone production during this season could be expected (Ternois et al., 1996;Sicre et al., 1999;Bárcena et al., 2004;Versteegh et al., 2007;Hernández-Almeida et al., 2011). This observation is supported by results from sediment traps located in the GoL showing very low coccolith fluxes during the summer months (Rigual-Hernández et al., 2013), while they exhibit higher values during autumn, winter, and spring, reaching maximum fluxes at the end of the winter season, during SST minima. In contrast, high fluxes of G. bulloides are almost restricted to the upwelling spring signal, when coccolith fluxes have already started to decrease (Rigual-Hernández et al., 2012. This different growth season can explain the proxy bias in the SST reconstructions, with more smoothed alkenone-SST signals. Both Mg / Ca-SST and U k 37 -SST records show consistent cooling trends of about −0.5 • C kyr −1 during the studied period (2700 years), which is consistent with the recent 2 kyr global reconstruction (McGregor et al., 2015;see Sect. 5.2). The recorded cooling since the RP SST maxima (∼ 200 CE) is more pronounced in the Mg / Ca-SST (−1.7 to −2.0 • C kyr −1 ) than in the alkenone signal (−1.1 • C kyr −1 ). These coolings are larger than those estimated in the global reconstruction (McGregor et al., 2015) for the last 1200 years (average anomaly method 1: −0.4 to −0.5 • C kyr −1 ). It should be noted that the global reconstruction includes alkenone-SST from MIN cores (data published in Moreno et al., 2012).
The detailed comparison of the centennial SST variability recorded by both proxy stacks consistently indicates a puzzling antiphase (Fig. 6b and c). Although the main trends are consistently parallel in both alkenone and Mg / Ca proxies (r = 0.5; p value = 0) as observed in other regions, shortterm variability appears to have an opposite character. Statistical analysis of these differences examined by means of Welch's test indicates that the null hypothesis (means are equal) can be discarded at the 5 % error level: t observed (12.446) > t critical (1.971). This a priori unexpected proxy difference outlines the relevance of the seasonal variability for climate evolution and suggests that extreme winter coolings were followed by more rapid and intense spring warmings. Nevertheless, regarding the low amplitude of several of these oscillations, often close to the proxy error, this observation needs to be supported by further constraints as a solid regional feature.

Climate patterns during the last 2.7 kyr
The SST changes in the Minorca region have implications for the surface air mass temperature and moisture source regions that could influence on air mass trajectories and ultimately precipitation patterns in the western Mediterranean region (Millán et al., 2005;Labuhn et al., 2015). Recent observations have identified SST as a key factor in the development of torrential rain events in the western Mediterranean Basin (Pastor et al., 2001), constituting a potential source of mass instability that transits over these waters (Pastor, 2012). In this context, the combined SST and δ 18 O sw records can provide information on the connection between thermal changes and moisture export from the central-western Mediterranean Sea during the last 2.7 kyr.
The oldest period recorded in our data is the so-called Talaiotic Period (TP), which corresponds to the ages of antiquity such as the period of ancient Greece in other geo-    (Lirer et al., 2014); (e) 30-year averages of the PAGES 2k Network (2013) Europe anomaly temperature reconstruction; (f) Greenland snow surface temperature (Kobashi et al., 2011); and (g) central Europe summer anomaly temperature reconstruction in central Europe (Büntgen et al., 2011). graphic areas. Both Mg / Ca-SST and alkenone-SST records are consistent in showing a general cooling trend from ∼ 500 BCE and reaching minimum values by the end of the period (∼ 120 BCE; Fig. 6a-b). Very few other records are available from this time period, which make comparisons of these trends at regional scale difficult. One of the most prominent features in the two SST reconstructions, particularly in the Mg / Ca-SST stack, is the warm SST that predominated during the second half of the RP (150-400 CE). The onset of the RP was relatively cold and a ∼ 2 • C warming occurred during the first part of this period. This SST evolution from colder to warmer conditions during the RP is consistent with the isotopic record of the Gulf of Taranto (Taricco et al., 2009) and peat recon-structions from north-west Spain (Martínez-Cortizas et al., 1999), and to some extent with SST proxies in the southeastern Tyrrhenian Sea (Lirer et al., 2014). However, none of these records indicates that the RP was the warmest period of the last 2 kyr. Other records from higher latitudes such as Greenland (Dahl-Jensen et al., 1998), and northern Europe (Esper et al., 2014), North Atlantic Ocean (Bond et al., 2001Sicre et al., 2008), as well as speleothem records from northern Iberia (Martín-Chivelet et al., 2011) and even the multiproxy PAGES 2K reconstruction from Europe, suggest a rather warmer early RP than late RP and, again, none of these records highlights the Roman times as the warmest climate period of the last 2 kyr. Consequently, these very warm RP conditions recorded in the Minorca Mg / Ca-SST stack seem to have a regional character and suggest that climate evolution during this period followed a rather heterogeneous thermal response along the European continent and surrounding marine regions.
Moreover, the observed δ 18 O sw stack of the RP suggests an increase in the E-P ratio (Fig. 6a) during this period, which has also been observed in some nearby regions like the Alps (Holzhauser et al., 2005;Joerin et al., 2006). In contrast, a lake record from southern Spain indicates relatively high water levels when the δ 18 O sw stack indicates the maximum in E-P ratio (Martín-Puertas et al., 2008). This information is not necessarily contradictory, since enhanced E-P balance in the Mediterranean could be balanced out by enhanced precipitation in some of the regions, but more detailed geographical information is required to interpret these proxy records from distinct areas.
After the RP, during the whole DMA and until the MCA, the Mg / Ca-SST stack shows a cooling of ∼1 • C (−0.2 • C 100 yr −1 ), which is 0.3 • C in the case of the alkenone-SST stack and the E-P rate decreases. This trend contrasts with the general warming trend interpreted from the speleothem records of northern Iberia (Martín-Chivelet et al., 2011) or the transition towards drier conditions observed in Alboran records (Nieto- . However, SST proxies from the Tyrrhenian Sea show a cooling trend after the second half of the DMA and the Roman IV cold/dry phase (Lirer et al., 2014) that can be tentatively correlated with our SST records (Fig. 6). This cooling phase is also documented in the δ 18 O G. ruber record of the Gulf of Taranto (Grauel et al., 2013). These heterogeneities in the signals from the different proxies and regions illustrate the difficulties in characterizing the climate variability during these short periods and reinforce the need for a better geographical coverage of individual proxies.
The medieval period is usually described as a very warm period in numerous regions in the Northern Hemisphere (Hughes and Diaz, 1994;Mann et al., 2008;Martín-Chivelet et al., 2011), but this interpretation is challenged by an increasing number of studies (i.e. Chen et al., 2013). The Minorca SST stacks also show the occurrence of significant temperature variability that does not reflect a specific warm period within the last 2 kyr (Fig. 6). An important warming event is observed at ∼ 1000 CE, followed by a later cooling with minimum values at about 1200 CE (Fig. 6). Higher temperature variability is found in Greenland records (Kobashi et al., 2011), while an early warm MCA and posterior cooling is also observed in temperature reconstructions from central Europe (Büntgen et al., 2011) and in the European multiproxy 2 kyr stack of the PAGES 2K Consortium (2013). Nevertheless, all these proxies agree in indicating overall warmer temperatures during the MCA than during the LIA. At the MCA-LIA transition, a progressive cooling and a change in oscillation frequency before and after the onset of LIA are recorded. This transition is consistent with the last rapid climate change (RCC) described in Mayewski et al. (2004).
In the context of the Mediterranean Sea, the lake, marine, and speleothem records consistently agree in showing drier conditions during the MCA than during the LIA Chen et al., 2013;Nieto-Moreno et al., 2013;Wassenburg et al., 2013). Examination of the δ 18 sw stack shows several oscillations during the MCA and LIA but no clear differentiation between these periods can be inferred from this proxy, indicating that reduced precipitation also involved reduced evaporation in the basin and that the E-P balance recorded by the δ 18 O sw proxy was not modified. The centennial-scale variability found in both the Mg / Ca-SST and δ 18 O sw stack reveals that higher E-P conditions existed during the warmer intervals ( Fig. 6a and c).
According to the Mg / Ca-SST stack, the LIA stands out as a period of high thermal variability in which two substages can be differentiated, a first involving large SST oscillations and warm average temperatures (LIAa) and a second substage with short oscillations and cold average SST (LIAb). We suggest that the LIAa interval could be linked to the Wolf and Spörer solar minima and that the LIAb corresponds to Maunder and Dalton cold events, in agreement with previous observations (i.e. Vallefuoco et al., 2012).
These two LIA substages are also present in the Greenland record (Kobashi et al., 2011). The intense cooling drop (0.8 • C 100 yr −1 ) at the onset of the LIAb is in agreement with the suggested coolings of 0.5 and 1 • C in the Northern Hemisphere (i.e. Matthews and Briffa, 2005;Mann et al., 2009). These two steps within the LIA are better reflected in the Mg / Ca-SST stack than in the alkenone-SST stack. This is also the case of the alkenone records in the Alboran Sea (Nieto- , which may result from smaller SST variability of the alkenone proxies (see Sect. 5.5).
In terms of humidity, the LIA represents a period of increased runoff in the Alboran record (Nieto- . Available lake level reconstructions from southern Spain also show progressive increases after the MCA, reaching maximum values during the LIAb (Martín-Puertas et al., 2008). Different records of flood events in the Iberia Peninsula also report a significant increase in extreme events during the LIA (Barriendos and Martin-Vide, 1998;Benito et al., 2003;Moreno et al., 2008). These conditions are consis-tent with the described enhanced storm activity over the GoL in this period (Sabatier et al., 2012), explaining the enhanced humidity transport towards the Mediterranean Sea as a consequence of the reduced E-P ratio observed in the δ 18 O sw , particularly during the LIAb (Fig. 6a).
The end of the LIA and onset of the IE is marked with a warming phase of about 1 • C in the Mg / Ca-SST stack and a lower-intensity change in the alkenone-SST stack. This initial warm climatic event is also documented in other Mediterranean regions (Taricco et al., 2009;Marullo et al., 2011;Lirer et al., 2014) and Europe (PAGES 2K Consortium, 2013), which is coincident with a total solar irradiance (TSI) enhancement after Dalton minima. The two Minorca SST stacks show a cooling trend by the end of the record, which does not seem to be consistent with the instrumental atmospheric records. In the western Mediterranean, warming has been registered in two main phases: from the mid-1920s to 1950s and from the mid-1970s onwards (Lionello et al., 2006). The Minorca stacks do not show this warming, but they do not cover the second warming period. Nevertheless, the instrumental data from the beginning of the 20th century in the western Mediterranean do not display any warming trends before the 1980s (Vargas-Yáñez et al., 2010).

Climate forcing mechanisms
The general cooling trend observed in both Mg / Ca-SST and alkenone-SST stacks shows a good correlation with the evolution of summer insolation in the North Hemisphere, which dominates the present annual insolation balance (r = 0.2 and 0.8, p value ≤ 0.007, respectively; Fig. 7). In numerous records from the Northern Hemisphere (i.e. Wright, 1994;Marchal et al., 2002;Kaufman et al., 2009;Moreno et al., 2012), this external forcing has also been proposed to control major SST trends during the Holocene period. In addition, summer insolation seems to have had significant influence in the decreasing trend of the isotope records during the whole period spanned (r = 0.4, p value = 0), as has been suggested in, for example, Ausín et al. (2015). In any case, a different forcing mechanism needs to account for the centennial-scale variability of the records, e.g. increased volcanism in the last millennium (McGregor et al., 2015), although no significant correlations have been observed between our records and volcanic reconstructions (Gao et al., 2008). Solar variability has frequently been proposed to be a primary driver of the Holocene millennial-scale variability (i.e. Bond et al., 2001). Several oscillations observed in the TSI record (Fig. 7a), but the correlations with the Mg / Ca-SST and alkenone-SST stacks are low, since most of the major TSI drops do not correspond to SST cold events. However, some correlation is observed between TSI and alkenone SSTs (r = 0.5, p value = 0). In any case, TSI does not seem to be the main driver of the centennial-scale SST variability in the studied records.   One of the major drivers of the Mediterranean interannual variability in the Mediterranean region is the NAO (Hurrell, 1995;Lionello and Sanna, 2005;Mariotti, 2011). Positive NAO indexes are characterized by high atmospheric pressure over the Mediterranean Sea and increases of the E-P balance (Tsimplis and Josey, 2001). During these positive NAO periods, winds over the Mediterranean tend to be deviated towards the north, overall salinity increases, and formation of dense deep water masses is reinforced as the water exchange through the Corsica Channel is higher and the arrival of northern storm waves decreases (Wallace and Gutzler, 1981;Tsimplis and Baker, 2000;Lionello and Sanna, 2005). The effect of NAO on Mediterranean temperatures is more ambiguous. SST changes during the last decades do not show significant variability with NAO (Luterbacher et al., 2004;Mariotti, 2011), although some studies suggest an opposite response between the two basins, with cooling responses in some eastern basins and warmings in the western basin during positive NAO conditions (Demirov and Pinardi, 2002;Tsimplis and Rixen, 2002). Although still controver-  Figure 8. δ 18 O sw Minorca stack (‰ SMOW) during the last millennium (age is expressed in years CE) plotted with (a) NAO reconstruction (Ortega et al., 2015) and (b) palaeostorm activity in the Gulf of Lion (Sabatier et al., 2012). Notice that the NAO axis is on a descending scale. Grey vertical bars represent negative NAO phases.
sial, some NAO reconstructions on proxy records are starting to become available for the period studied Olsen et al., 2012;Trouet et al., 2012;Ortega et al., 2015). The last millennium is the best-resolved period, and it allows a direct comparison with our data to evaluate the potential link to NAO.
The correlations between our Minorca temperature stacks with NAO reconstructions (Fig. 7) are relatively low in the case of Mg / Ca-SST (r = 0.3, p value ≤ 0.002) and not significant in the alkenone stack, indicating that this forcing is probably not the driver of the main trends in these records, although several uncertainties still exist about the long NAO reconstructions . If detailed analysis is performed focussing on the more intense negative NAO phases (Fig. 7), they mostly appear to correlate with cooling phases in the Mg / Ca stack. The frequency of these negative events is particularly high during the LIA, and mostly during its second phase (LIAb), when the coldest intervals of our SST stacks were observed.
When several different proxy last century records of annual resolution, tested with some model assimilations (Ortega et al., 2015), are compared with the last NAO reconstruction, the observed correlations with δ 18 O sw are not statistically significant. However, the Welch's test results do not allow for the null hypothesis to be discarded. A coherent pattern of NAO variability with our δ 18 O sw reconstruction, with high (low) isotopic values mainly dominating during positive (negative) NAO phases, can be observed in the last centuries (Fig. 8). This pattern is consistent with the described E-P increase during high NAO phases described for the last decades (Tsimplis and Josey, 2001). The SST stacks also suggest some degree of correlation between warm SST and high NAO values (Fig. 7), but a more coherent picture is observed when the SST records are compared to the Atlantic Meridional Oscillation (AMO) reconstruction: warm SST dominated during high AMO values (Fig. 9). This pattern of salinity changes related to NAO and SST to AMO has also been

S S T (ºC )
Mg/C a-S S T U k' 37 -S S T Figure 9. Mg / Ca-SST and alkenone-SST Minorca anomaly stacks during the last centuries plotted with AMO reconstruction (Gray et al., 2004). described in climate studies encompassing the last decades (Mariotti, 2011;Guemas et al., 2014) and confirms the complex but tight response of the Mediterranean to atmospheric and marine changes over the North Atlantic Ocean.
The pattern of high δ 18 O sw at dominant positive NAO corresponds to a reduction in the humidity transport over the Mediterranean region as a consequence of high atmospheric pressure (Tsimplis and Josey, 2001). Accordingly, several periods of increased/decreased storm activity in the GoL ( Fig. 8; Sabatier et al., 2012) correlate with low/high values in the δ 18 O sw , indicating that, during negative NAO conditions, northern European storm waves arrived more frequently in the Mediterranean Sea (Lionello and Sanna, 2005), contributing to the reduction of the E-P balance (Fig. 8). Our data also indicate that, during these enhanced storm periods, cold SST conditions dominated in the region as previously suggested (Sabatier et al., 2012). Nevertheless, not all the NAO oscillations had identical expression in the compared records, which is coherent with recent observations indicating that negative NAO phases may correspond to different atmospheric configuration modes and impact differently over the western Mediterranean Sea (Sáez de Cámara et al., 2015). Regarding the lower part of the record, the maximum SST temperatures and δ 18 O sw recorded during the RP (100-300 CE) may suggest the occurrence of persistent positive NAO conditions, which would also be consistent with a high pressure-driven drop in relative sea level as has been reconstructed in the north-western Mediterranean Sea (southern France, −40 ± 10 cm; Morhange et al., 2013).
It is interesting to note that during the DMA a pronounced and intense cooling event is recorded in the Mg / Ca-SST stack at about 500 CE. Several references document in the scientific literature the occurrence of a dimming of the sun at 536-537 CE (Stothers, 1984). This event, based on ice core records, has been linked to a tropical volcanic eruption (Larsen et al., 2008). Tree-ring data reconstructions from Europe and also historical documents indicate the persistence during several years (536-550 CE) of what is described as the most severe cooling across the Northern Hemisphere during the last two millennia (Larsen et al., 2008). Despite the limitations derived from the resolution of our records, the Mg / Ca-SST stack record may have caught this cooling, which would prove the robustness of our age models (see Supplement for age model development).

Summary and conclusions
The review of new core-top data of G. bulloides Mg / Ca ratios from the central-western Mediterranean Sea together with previous published data support a consistent temperature sensitivity for the Mediterranean samples and allows for the previous calibrations to be refined. The recorded Mg / Ca-SST signal from G. bulloides is interpreted to reflect April-May conditions from the upper 40 m layer. In contrast, the alkenone-SST estimations are interpreted to integrate a more annually averaged signal, although they are biased toward the winter months since primary productivity during the summer months in the Mediterranean Sea is extremely low. The averaged signal of the alkenone-SST records may explain its relatively smoothed oscillations in comparison to the Mg / Ca-SST records.
After careful construction of a common chronology for the studied multicores, based on several chronological tools, the individual proxy records have been grouped in an anomalystacked record to allow a better identification of the main patterns and structures. Both Mg / Ca-SST and alkenone stacks show a consistent cooling trend over the studied period. Since the RP maximum, this cooling has ranged between −1.7 and −2.0 • C kyr −1 in the Mg / Ca record and is less pronounced in the alkenone record (−1.1 • C kyr −1 ). This cooling trend is consistent with the general lowering of summer insolation. The overall cooling is punctuated by several SST oscillations at centennial timescales, which represent maximum SST during most of the RP; a progressive cooling during the DMA; a pronounced variability during the MCA with two intense warming phases reaching warmer SST than during the LIA; and a very unstable and rather cold LIA, with two substages -a first one with larger SST oscillations and warmer average temperatures (LIAa) and a second one with shorter oscillations and colder average SST (LIAb). The described two stages within the LIA are clearer in the Mg / Ca-SST stack than in the alkenone-SST record. Comparison of Mg / Ca-SST and δ 18 O sw stacks indicates that warmer intervals have been accompanied by higher evaporation-precipitation (E-P) conditions. The E-P balance oscillations over each defined climatic period during the last 2.7 kyr suggest variations in the thermal change and moisture export patterns in the central-western Mediterranean.
Comparison of the Minorca SST stacks with other European palaeoclimatic records suggests a rather heterogeneous thermal response along the European continent and surrounding marine regions. Comparison of the new Mediterranean records with the reconstructed variations in TSI does not support a clear connection with this climate forcing. Nevertheless, changes in the NAO and AMO seem to have in-fluenced the regional climate variability. The negative NAO phases correlate mostly with cooling phases of the Mg / Ca stack, although this connection is complex and apparently better defined during the most intense negative phases. Focussing on the last 1 kyr, when NAO reconstructions are better constrained, provides a more consistent pattern, with cold and particularly fresher δ 18 O sw values (reduced E-P balance) during negative NAO phases. Our results are consistent with enhanced southward transport of European storm tracks during this period and previous reconstructions of storm activity in the GoL. Nevertheless, the SST stacks show a more tied relation to AMO during the last four centuries (the available period of AMO reconstructions) in which warm SST dominated during high AMO values. This evidence supports a close connection between Mediterranean and North Atlantic climatology over the last 2 kyr.
The Supplement related to this article is available online at doi:10.5194/cp-12-849-2016-supplement.