The 5017-1 sediment core from the deep Dead Sea basin (31∘30′29′′ N, 35∘28′16′′ E; ca. 300 m water depth in 2010; sediment surface ∼ 725 m b.m.s.l.; Fig. 1) was obtained during the drilling campaign of the ICDP Dead Sea Deep Drilling Project (DSDDP) in winter 2010–2011 (Stein et al., 2011). The record is ∼ 455 m long and comprises two full glacial–interglacial cycles (Neugebauer et al., 2014; Torfstein et al., 2015). Here, we focus on a ∼ 65 m long section from ∼ 180 to ∼ 245 m below lake floor (m b.l.f.). Sediment facies were described with an accuracy of 1 cm based on line-scanning images of the split sediment cores. Magnetic susceptibility data in 1 mm resolution were routinely obtained for the entire 5017-1 record (see Neugebauer et al., 2014, for details).

For micro-facies analyses we applied a combination of petrographic thin-section microscopy and high-resolution µXRF element scanning. A total of 26 large-scale thin-section samples (10 × 2 cm) were prepared representing changes in facies types along the section. Preparation largely followed the standard procedure for soft sediments (e.g. Brauer et al., 1999) but were performed under dry conditions to avoid salt crystallisation during the preparation process. Thin sections were analysed with a petrographic microscope (Leica DMLP) and images were taken with a digital camera (Olympus DP72). Fluorescence was analysed using a Nikon AZ100M microscope, operated with violet and polarised light conditions, and Nikon photo software (NIS Elements AR 4.3).

The µXRF measurements were acquired every 200 µm for 10 s using the ITRAX µXRF core scanner at GFZ, Germany. The core scanner is equipped with a Cr tube operated at 30 kV and 30 mA to irradiate the split-core sediment surface. This non-destructive method acquires element intensities of Si, S, Cl, K, Ca, Ti, Fe, Br and Sr (Neugebauer et al., 2014), which are presented as count rates (counts per second – cps). The element intensity records reflect relative changes in the composition of the Dead Sea sediments, but they are also influenced by physical sediment properties (e.g. density, water content, grain size) and the sample geometry. The easiest and most convenient way to minimise the physical and geometrical sample effects is by the transformation of element intensities into ratios or log ratios (Weltje and Tjallingii, 2008).

Laminated sediments were sampled for grain size distributions with 1 cm3 sample volume at 1–3 cm vertical resolution and a total of 363 samples. Sample preparation included decomposing organic matter using 30 mL H2O2 (30 %) and distilled water (1:1 concentration) and breaking aggregates with Calgon detergent ((NaPO3)6, 1 %) and ultrasonic bath. The particle size distribution was measured using an LS 13 320 laser diffraction particle size analyser for (1) the total sample and (2) the carbonate-free sample after dissolution through HCl (32 %, dilution of 1:9 with distilled water). Less than 1 g of sediment was required for measurement.

In total, 22 gravel layers detected in core 5017-1 were sampled for petrographic analyses. The samples were wet-sieved for five grain size fractions (> 4, 2–4, 1–2, 0.5–1 and < 0.5 mm), for which strewn slides were prepared for microscopic inspections. Here, we focus on two gravel units occurring within the studied core section (180–245 m b.l.f.).

For X-ray powder diffraction (XRD) measurements 25 samples were collected from about the same depths as thin sections to complement microscopic inspections. Powder X-ray patterns were collected using a PANalytical Empyrean powder diffractometer with Cu Kα radiation, automatic divergent and antiscatter slits and a PIXcel3D detector. The diffraction data were recorded from 5∘ to 85∘ 2Θ via a continuous scan with a step size of 0.013 and a scan time of 60 s per step. The generator settings were 40 kV and 40 mA.

Total organic carbon (TOC) and calcium carbonate (CaCO3) contents have been determined from 19 of these samples using an elemental analyser (EA3000-CHNS Eurovector). First, 5–10 mg dried and homogenised sample material was weighed in Sn capsules for total carbon (TC) determination. Subsequently, second sample aliquots of 3–4 mg of the samples were decalcified in Ag capsules in three steps through treatment with (1) 3 % HCl, (2) 20 % HCl and (3) drying at 75∘ C for TOC determination. Data were calibrated with standards (BBOT, sulfanilamide, and for TOC additionally Boden3) and empty Sn and Ag capsules. The relative standard deviation is < 1 %. CaCO3 contents were calculated from the difference of TC–TOC.

The sediments of the analysed ∼ 65 m long section of core 5017-1 mainly consist of laminated marl of the aad facies (alternating aragonite and detritus; e.g. Machlus et al., 2000), gypsum and halite deposits (Neugebauer et al., 2014). Commonly, detrital material is composed of clay to silt-sized calcite, quartz, dolomite and minor feldspar and clay minerals. The thickness of detrital layers ranges from < 1 mm to several centimetres and their colour is greyish to black, if iron sulfides are present (pyrite or greigite), or brownish and greenish, if terrestrial organic or algal remains are dominant. Aragonite formed as 5–15 µm small stellate aggregates of orthorhombic crystals building ∼ 0.1–4 mm thick white laminae. Monoclinic, euhedral ∼ 10–60 µm gypsum crystals build ∼ 0.2–3 cm thick beige layers. Larger, up to 1 mm gypsum crystals appear scattered within detrital layers. Cubic halite crystals ranging in length from ∼ 1 mm to several centimetres are either embedded in predominantly detrital marl or build thick deposits. These thick halite deposits contain only minor detrital material and are often layered.

Six micro-facies types were identified (Fig. 2):

green aad: alternating white aragonite and greenish detrital marl laminae (∼ 1 mm thick couplets; Fig. 2a), the greenish laminae exhibit some diatoms and very strong fluorescence pointing to a significant amount of chlorophyll preserved in the sediment (Fig. 2f);

In one exceptional case at ∼ 239 m composite depth, a ∼ 35 cm thick layer of well-sorted gravels with < 2 % clay to sand matrix appears (Figs. 3 and 4). The petrographic composition of this gravel deposit is identical to that of the other mud-supported gravels (Fig. 3). This gravel layer is from a core section that suffered a major loss of core material during the drilling process (core 5017-1-A-92-1). From the 130 cm long core drive only a cumulative thickness of 35 cm gravels and almost no fine material were recovered in the liner. Therefore, the sedimentological contacts to over- and underlying sediments are not preserved and are unknown (Fig. 3). Unfortunately, this prevents investigation of sediment structures in the context of the complete depositional environment. Interestingly, in the 10 cm wide core catcher of this core drive, matrix-supported gravel has been caught. This core catcher sample largely resembles the basal layers of the above-mentioned thick turbidites and slumps.

Median grain size values of the laminated sediments, excluding the halite-facies types hd and lh/hh, vary between ∼ 7 and ∼ 10 µm for samples with and without CaCO3 (i.e. after dissolution of CaCO3; see the Supplement), respectively. These grain size distributions indicate mainly clay (∼ 54 and ∼ 43 % with and without CaCO3, respectively), very fine silt (∼ 45 and ∼ 55 % with and without CaCO3, respectively) and very little sand (∼ 0.1 and ∼ 0.9 % with and without CaCO3, respectively). Gypsum-detritus samples (gd facies) revealed the coarsest mean grain size of ∼ 8 µm (∼ 12.5 µm without CaCO3) and the highest sand fraction (∼ 0.5 and ∼ 2.2 %, with and without CaCO3, respectively) due to gypsum which was not removed during sample treatment. The aad-n and aad-II micro-facies show similar and low mean grain sizes of ∼ 6 µm (∼ 9 µm without CaCO3), while the green aad type exhibits a slightly higher mean value of ∼ 7 µm (∼ 11 µm without CaCO3). Also, the silt and sand fractions of the green aad type are enhanced in comparison to the other two aad types (see the Supplement).

The differentiation of the laminated micro-facies types gd, aad-II, aad-n and green aad is supported by total organic carbon and calcium carbonate contents (Fig. 2g). The gd facies is characterised by lowest TOC values of 0.25–0.5 and ∼ 18 % CaCO3, whereas the aad-II facies (0.35–0.57 % TOC, 30–47 % CaCO3) and the aad-n facies (0.6–0.7 % TOC, ∼ 47 % CaCO3) exhibit higher values. The green aad facies is characterised by highest TOC (0.65 and ∼ 0.9 %) and CaCO3 contents (∼ 40–50 and 62 %).

The elements Si, S, Cl, K, Ca, Ti, Fe, Br and Sr were obtained by µXRF scanning and used to characterise the Dead Sea sediments (Fig. 2; see also Neugebauer et al., 2014, 2015). Aragonite laminae are revealed by high Sr / Ca values, gypsum is represented by high S / Ca values and halite is best characterised by the Cl / Br ratio (Fig. 2). The elements Si, K, Ti and Fe are constrained to siliciclastics in the detrital sediment fraction. More ambiguous is the interpretation of Ca that occurs in aragonite, gypsum and detrital calcite. The Ca / (Sr+S) ratio indicates the detrital carbonate fraction because the authigenic Ca sources, i.e. aragonite and gypsum, are removed. The Ti / Ca ratio represents the relative siliciclastic fraction (Fig. 2). The sum of Ca / (Sr+S) and Ti / Ca ratios best represents the total amount of carbonate and siliciclastic detritus (Fig. 4). A correlation plot of these two element ratios shows low, but significant, correlation (R2 = 0.4) for the carbonate and siliciclastic detrital fractions (Fig. 2h); the plot also indicates an additional Ca-bearing detrital fraction.

The analysed section of core 5017-1 is subdivided into four main lithostratigraphic units (Fig. 4). Units I and III are predominantly composed of halite (controlled by hd and lh/hh facies), some gypsum and detrital marl. Units II and IV present primarily aad (green aad, aad-n and aad-II) and gd facies. These lithostratigraphic units are tied to the stratigraphic framework (Neugebauer et al., 2014) and the U–Th chronology (Torfstein et al., 2015) of the 5017-1 core.

The lowermost unit I (245–237.5 m composite depth) is the upper part of a ca. 40 m thick halite sequence, the thickest halite deposit in the entire core, and is part of the last interglacial Samra Formation (Neugebauer et al., 2014). This unit has very low magnetic susceptibility values and Sr / Ca ratios with high Cl / Br ratios (Fig. 4). The above-mentioned well-sorted gravel deposit, mainly composed of limestone and dolomite clasts and halite, was identified ∼ 2 m below the top of this halite unit (Figs. 3 and 4). U–Th ages proposed a sedimentary hiatus between ca. 116 and 110 ± 3 ka marked by this gravel layer (Torfstein et al., 2015).

The ∼ 25 m thick unit II (237.5–212.5 m) presents mainly aad and gd facies. It is divided into two subunits: (1) subunit II-a (237.5–228 m) comprises aad-II, aad-n and gd facies and is characterised by low Sr / Ca and Cl / Br ratios and distinct peaks in magnetic susceptibility. (2) Subunit II-b (228–212.5 m) differs from subunit II-a as, in addition to the above facies, it presents three thick sequences characterised by green aad facies and partly high Sr / Ca ratios. Unit II is characterised by frequent, up to several metres thick, graded detrital layers and slump deposits (Fig. 4). U–Th ages place unit II between ca. 108 ± 5 and 93 ± 7 ka (Torfstein et al., 2015), i.e. an interval of ∼ 3–27 000 years. Preliminary varve counting on the core photographs of this unit reveals a minimum of 4050 ± 250 varves, which is at the lower end of the range and uncertainty of the U–Th ages. It is likely that much of the sediment was eroded through the frequent mass-waste events. Unit II builds the upper part of the Samra Formation of core 5017-1 as defined by Neugebauer et al. (2014).

Unit III (212.5–201.5 m) is dominated by halite deposits of the hd and lh/hh facies, which is well reflected in high Cl / Br ratios. The lower ca. 4 m of this unit could not be recovered due to the hardness of the salt. Some centimetre- to decimetre-thick occurrences of aad-n, aad-II and gd facies are intercalated in the halite deposits, as reflected by higher magnetic susceptibility, Ca / (Sr+S) + Ti / Ca and Sr / Ca ratios. Unit III was deposited between ca. 93 and 87 ± 7 ka (Torfstein et al., 2015) and probably marks the transition between the Samra and Lisan formations in the deep-basin core 5017-1 (Neugebauer et al., 2014). Compared to the chronology of the outcrops at the margin where the Samra–Lisan transition has been traditionally considered at 75–70 ka (e.g. Waldmann et al., 2009), probably because of transgressive truncation, the deep core may indicate that the transition occurred ca. 15 000 years earlier (Torfstein et al., 2015).

The uppermost unit IV (201.5–180 m) compares to unit II and is characterised by the three aad facies, as indicated by higher Ca / (Sr+S) + Ti / Ca and Sr / Ca ratios and the absence of halite (Fig. 4). In contrast to unit II, where magnetic susceptibility values strongly fluctuate, constantly low magnetic susceptibility characterises unit IV (Fig. 4). This unit can be divided into three subunits: (1) subunit IV-a is composed of aad-n, aad-II and gd facies; (2) subunit IV-b is a green aad section; and (3) subunit IV-c is composed of aad-n and aad-II. Several centimetre- to metre-thick slumped deposits and graded detrital layers occur in unit IV. At a composite core depth of ∼ 195 m the sediment is ca. 85.5 ± 8 ka, and 6 m above unit IV (i.e. at 174.5 m depth) an age of 70.5 ± 5 ka has been reported (Torfstein et al., 2015). The interpolated age of the upper boundary of unit IV at 180 m depth is ca. 75 ± 6 ka. Unit IV builds the lowermost part of the Lisan Formation of core 5017-1 (Neugebauer et al., 2014).

Lake levels of the water bodies occupying the Dead Sea basin are sensitive responders to changing hydro-climatic conditions in the lake's catchment (Enzel et al., 2003, 2008; Bookman et al., 2006). Lake level reconstructions based on onshore sequences indicate a total amplitude of lake level fluctuation of at least ∼ 270 m, with lowest levels of ∼ 430 m b.m.s.l. occurring during parts of the last interglacial, the last termination, potentially the Holocene and – anthropogenically induced – in modern times (e.g. Bookman (Ken-Tor) et al., 2004; Bartov et al., 2007; Waldmann et al., 2009; Stein et al., 2010). The highest lake level of Lake Lisan of ∼ 160 m b.m.s.l. was reached during the last glacial maximum (Bartov et al., 2003). These exposed sediments at the Dead Sea margins also showed that in general different lake levels resulted in different sedimentary facies (e.g. Machlus et al., 2000; Migowski et al., 2006). Hence, facies types can be considered relative lake level indicators, but without assigning an absolute level change (Figs. 4 and 5). Unlike the near-shore environment, where lateral changes can alter the sedimentary facies, which may lead to erroneous relative lake level interpretations, in the deep basin such lateral changes are uncommon and, therefore, facies changes are better related to changes in relative lake levels. These relative lake levels are crucial to infer regional, basin-scale hydroclimatic changes that control the direction of lake level trends (i.e. rising or falling), which are the net product of the respective positive or negative lake budget over decades to millennia. To avoid complexities in inferring minor relative lake level changes and to remain reasonable within the resolution of the U–Th chronology, we concentrated only on reconstructing the millennial-scale facies alterations and interpret them in terms of relative lake level variations.

The typical sediment facies during rising levels and the resulting episodic high stands of both the deep last glacial Lake Lisan and the much shallower Holocene Dead Sea is the aad facies composed of alternating aragonite and detritus (e.g. Machlus et al., 2000; Bookman (Ken-Tor) et al., 2004). As the lake is devoid of bicarbonate, deposition of aad requires large amounts of bicarbonate supply by freshwater reaching the lake through runoff during the winter rainy season to trigger precipitation of primary aragonite (Stein et al., 1997; Barkan et al., 2001). Three different subtypes of aad were distinguished in the investigated sediment section through micro-facies analyses. (i) Green aad (Fig. 2) comprises greenish detrital laminae containing green algae remains and represents highest lake levels and less salty limnological conditions. This facies depicts the sediments deposited in core 5017-1 during the Last Glacial Maximum high stands (Neugebauer et al., 2014), when Lake Lisan reached its maximum extent (e.g. Begin et al., 1974; Bartov et al., 2002). (ii) The aad-n and (iii) the aad-II facies are similar, except that aad-II is characterised by commonly thicker but irregularly spaced aragonite laminae (Fig. 2). This may indicate insufficient supply of bicarbonate to support regular annual aragonite formation. Therefore, the aad-II facies was likely deposited during episodes of somewhat lower lake levels compared to the aad-n facies. The aad-II facies also differs from the ld facies type (laminated detritus), which exhibits coarser detritus (50–60 µm) than aad (8–10 µm; Haliva-Cohen et al., 2012) and which is a characteristic facies for intermediate lake levels of the interglacial Samra and Ze'elim formations (e.g. Migowski et al., 2006; Waldmann et al., 2009; Neugebauer et al., 2014). The ld facies type was, however, not detected in the studied core section, which is supported by the constantly very fine grain sizes of the sediments (Supplement).

The deposition of well-laminated or massive gypsum (gd facies, Fig. 2) is associated with mixing of the water body due to lake level fall and a thinning of the upper freshwater layer (Torfstein et al., 2008). Halite deposition is related to a negative water balance during times of decreased lake levels (e.g. Lazar et al., 2014). Here, we distinguish between a mixed halite-detritus facies (hd) and layered or homogeneous, consolidated halite (lh/hh facies, Fig. 2). Whereas the presence of detritus suggests freshwater influx during extreme runoff events, deposition of thick halite indicates episodes of lowest lake levels.

Lake level trends inferred from micro-facies analysis are supported by µXRF element scanning data (Figs. 4 and 5). Halite sequences associated with a negative water balance are well expressed in increased Cl / Br ratios. The detrital input depends on the erosion in the catchment, aeolian deposition over the lake and the catchment, and freshwater supply to the lake. The relative detrital input can be estimated using the Ca / (Sr+S) ratio (for carbonate fraction) and the Ti / Ca ratio (for siliciclastic fraction). The Sr / Ca ratio resembles the aragonite amount that increases with enhanced supply of freshwater. The combination of these ratios by summing up both detrital fractions and aragonite and subtracting halite results in a curve that can be interpreted as a proxy for water balance (Fig. 5), with negative values for halite and gypsum deposits and positive values for detritus and aragonite.

Gravel deposits are rather common in the deep basin and have been identified as matrix-supported material mainly in basal layers of turbidites and slumps that are up to several metres thick and reflect mass-waste deposits, which can be triggered by either extreme runoff or seismic events and slope instabilities (Kagan and Marco, 2013; Neugebauer et al., 2014; Waldmann et al., 2014). Only in one case at ∼ 239 m composite sediment depth has a 35 cm thick well-sorted gravel deposit lacking fine-grained components been documented (Figs. 3 and 4). This gravel has been interpreted as beach deposit and in turn used to argue for a major drawdown or even almost desiccation of the lake at the end of the last interglacial (Stein et al., 2011; Torfstein et al., 2015). Combined U–Th ages and oxygen isotope stratigraphy suggest a ∼ 116 to 110 ka hiatus at around the position of the gravel deposit, which is assumed to support the drawdown hypothesis (Torfstein et al., 2015). However, both petrographic composition and grain characteristics of the well-sorted gravel are identical to gravel in basal layers of thick slumps and turbidites as the one deposited only 6 m above (at ∼ 233 m composite sediment depth; Fig. 3). This suggests the possibility of a similar source and even the same transport mechanism. Due to the massive core loss of 65 % in the core section where the well-sorted gravel was found, no direct information about the in situ contacts of this gravel to over- and underlying sediment units is available (Fig. 3a) and its primary sedimentological context remains unknown. However, the core catcher material supports the interpretation of even the well-sorted gravel as the vestige of a major mass-waste deposit, since it consists of matrix-supported gravel exactly resembling basal layers from at least 22 turbidites and slumps occurring in the entire record. It is likely that the fine-grained sediment components of the turbidite were washed out during the drilling process. Low core recoveries and loss of material often occurred in sediment sections with alternating hard halite and soft mud (Neugebauer et al., 2014) as in this section of the record above the major halite deposit. Therefore, we are convinced that the well-sorted gravels are the result of a drilling artefact and should not be interpreted as an in situ beach layer but rather as the washed-out relict of the basal sediments of a major mass-waste deposit. The deposition of a thick turbidite could have also caused the supposed hiatus, although its length should be critically tested.

Accepting our reinterpretation of the well-sorted gravel as primarily a drilling relict of a turbidite instead of an in situ beach deposit implies that the Dead Sea was not necessarily desiccated at the end of the last interglacial, although it might have been at a low stage. This consideration accords better with thermodynamic calculations and water balance simulations concluding that the chemistry of the brine and the geometry of the basin should prevent the lake from drying up (Yechieli et al., 1998; Krumgalz et al., 2000). First, the specific chemical composition of the Dead Sea brine (mainly Mg, Na, Ca and Cl) allows for a very high salinity with a low water activity and vapour pressure to be reached. Therefore, the rate of evaporation decreases with increasing salinity. Second, the low surface-area-to-volume ratio of the lake basin limits the amount of evaporated water. In addition, the relative humidity of the air above the brine has to be close to zero in order to further evaporate a highly concentrated brine, which, however, was never observed (Katz and Starinsky, 2015) and is considered unlikely, especially at very low lake levels, due to the wind-protected topography of the deep Dead Sea basin.

Relative lake levels have been reconstructed for the ∼ 117–75 ka interval based on the six micro-facies types introduced above (Figs. 4 and 5). Relatively lower lake levels are reflected by the halite-dominated units I (∼ 117–108 ka) and III (∼ 93–87 ka). Intermediate to relatively higher lake levels are inferred for the aad facies-dominated units II (∼ 108–93 ka) and IV (∼ 87–75 ka) because this facies indicates increased fresh water inflow.

The age estimate of unit I indicates that the low stand of Lake Samra commenced during the later part of the last interglacial and may have continued until ∼ 108 ± 5 ka (Fig. 5). However, there is no information for the time interval from ∼ 116 to 110 ± 4 ka, due to the erosional unconformity revealed from the chronological data (Torfstein et al., 2015). The deposition of ca. 2 m of halite above the hiatus indicates that low levels of the lake continued until ∼ 108 ± 5 ka because, for times of halite deposition in the Dead Sea basin, a lake level below 400 m b.m.s.l. can be assumed (Neev and Emery, 1967; Bookman (Ken-Tor) et al., 2004; Waldmann et al., 2009; Stein et al., 2010).

During ∼ 108–93 ± 6 ka (unit II) a trend of general increase in lake level is indicated by the succession from aad-II to aad-n and finally to the green aad facies (Fig. 4). Intercalated gypsum deposits from ∼ 108 to 100 ka indicate frequent short-term drops in lake levels. In the last glacial Lisan Formation such gypsum deposits were associated with reduced precipitation, intensified winds and probably increased evaporation during Heinrich events (Bartov et al., 2003; Torfstein et al., 2008, 2013; Rohling, 2013). A Lake Samra high-stand between ∼ 100 and 93 ka is in agreement with a level from exposures at the lake's margins, where a relatively high level of ∼ 320 m b.m.s.l. was proposed (Fig. 5; Waldmann et al., 2009, 2010).

An abrupt lake level decline and a subsequent millennial-scale low stand, probably below 400 m b.m.s.l., is inferred from the ∼ 7 m thick halite deposit during the ∼ 93–87 ± 7 ka interval (unit III, Figs. 4 and 5). Within this unit some aad-n, aad-II and gd facies alternate with the thick and mainly layered halite deposits indicating superimposed, probably centennial-scale, lake level fluctuations. This halite sequence represents the final stage of the Samra Formation and marks the last appearance of halite for the next ∼ 70 000 years (Neugebauer et al., 2014) until the early Holocene salt formation (e.g. Yechieli et al., 1993; Stein et al., 2010). The late Lake Samra halite indicates a more pronounced lake level drop than the limited lake level decline inferred from coarser clastic deposits in the exposed lake margin sediments (Fig. 5; Waldmann et al., 2009).

During ∼ 87–79 ± 7 ka (units IV-a and IV-b) lake level increased again as evidenced from the succession from aad-II and aad-n facies, intercalated by some gd facies, to green aad facies (Fig. 4). At ∼ 79 ka (unit IV-c) the lake probably shortly declined again as indicated by aad-II facies, before continuing to rise again at ∼ 77 ± 6 ka (Fig. 5). Earlier studies of the exposed sediments of the Samra and Lisan formations suggested that a depositional unconformity between ∼ 75 and 70 ka separated these two formations at the lake's margins (Bartov et al., 2003; Waldmann et al., 2009; Torfstein et al., 2013). Above this assumed unconformity, aad facies characterise the lower and upper members of the Lisan Formation (e.g. Bartov et al., 2002). Below the unconformity, the onshore Samra Formation is composed of reddish ld facies, sands and gravels (Waldmann et al., 2009). In the deep core, however, there is no obvious sedimentological indication for a low stand of the lake at around this time, but aad facies apparently continuously deposited since ∼ 87 ka. This suggests that Lake Lisan commenced ca. 10–15 kyr earlier than was assumed from the exposures where its deposits were in part truncated. This difference between shallow- and deep-basin sediments might be explained by (1) the abundant occurrence of slumping deposits and graded layers in the deep core (Fig. 4) or (2) by a lake level rise from a lower level to the level of the observed unconformity at the margins during that time. Combining these two possibilities suggests that these slumping deposits might indicate transgressive erosion at the outcrop locations during times of lake level rise of the early Lake Lisan (Bartov et al., 2007). This is likely causing unconformities in the near-shore marginal areas of the basin. The large number of slump deposits within this sediment section might point to several short-term level oscillations during this generally rising level trend, but there is no further evidence for this proposition.

The Dead Sea is situated at a key transitional zone between predominantly Atlantic and tropical-influenced climates. The zone of interaction between both climate regimes is expected to have changed during major climatic transitions like from glacial to interglacial modes and vice versa. In contrast to the scarce and sometimes contradicting information from the Levant, several records from the entire Mediterranean realm provide evidence for teleconnections of large-scale climate variations with the North Atlantic climate regime during the early last glacial. The alternation of cold stadial intervals from ∼ 111 to 108 ka (GS 25) and ∼ 90 to 85 ka (GS 23+22; Rasmussen et al., 2014), as reflected in Greenland ice cores (Fig. 5; Wolff et al., 2010) and North Atlantic ice-rafting events C24 and C22+C21 (Chapman and Shackleton, 1999), and warmer interstadials (GI 24+23 and GI 21; Rasmussen et al., 2014) are well expressed in the western (marine core MD952042 off the Iberian margin; Sánchez Goñi et al., 1999), central (Lago Grande di Monticchio; Fig. 5; Brauer et al., 2007; Martin-Puertas et al., 2014) and eastern Mediterranean (Tenaghi Philippon: Tzedakis, 2005; Lake Van: Litt et al., 2014; and marine sediments: Cheddadi and Rossignol-Strick, 1995). The same pattern of large-scale fluctuations is proposed from Lebanon (Lake Yammoûneh; Develle et al., 2011; Gasse et al., 2015) and the Soreq and Peqin speleothem records in Israel (Fig. 5; Bar-Matthews et al., 1999, 2000, 2003). Finally, the Dead Sea record also reveals low water levels at ∼ 110–108 ± 5 and ∼ 93–87 ± 7 ka, reflecting dry periods corresponding to Northern Hemisphere stadials, and higher lake levels at ∼  108–93 ± 6 and ∼ 87–75 ± 6 ka coinciding with Greenland interstadials (Fig. 5).

The long interstadials interrupted by short stadials in the North Atlantic realm during the build-up phase of the large continental ice sheets (e.g. Mangerud et al., 1996, 1998; Clark et al., 1999; Svendsen et al., 2004) are related to changes in Northern Hemisphere orbital insolation (Fig. 5; Laskar et al., 2004). At the same time, orbital insolation-driven changes of the Intertropical Convergence Zone controlled the monsoon system and led to a strengthening of the African summer monsoon and widespread vegetation cover in the Sahel and in the southern Sahara regions (Fig. 5; deMenocal et al., 2000; Tjallingii et al., 2008; Herold and Lohmann, 2009). Enhanced precipitation in eastern Africa induced the formation of organic-rich sapropel layers S4 and S3 (Fig. 5) in the eastern Mediterranean Basin due to enhanced Nile River runoff (e.g. Rossignol-Strick, 1985; Rohling et al., 2015). These changes in freshwater flow further influenced the isotopic composition of the eastern Mediterranean Sea water, i.e. the source region for precipitation in the region of the Soreq and Peqin speleothems (Fig. 5; Bar-Matthews et al., 2000, 2003). This is an indirect mechanism explaining a monsoonal influence in the speleothem records of the Levant.

Two issues must be discussed when precipitation increase and decrease are considered for explaining rising and falling trends in lake levels and their maxima: (a) winter vs. summer precipitation, and (b) tropical vs. Atlantic–Mediterranean sources. Different scenarios have been proposed but the source of moisture for the precipitation leading to higher Dead Sea lake levels during the early glacial and related atmospheric teleconnections is still debated. Northward shifts of the tropical rain belt as far north as the Levant have been excluded (Tzedakis, 2007, and references therein), and Enzel et al. (2015) also argued that summer rains associated with either the African or Indian monsoons are unlikely even in the southernmost point of the Dead Sea watershed. Based on hyperarid soils, Amit et al. (2006, 2011) demonstrated that the southern Negev, including the southern watershed of the Dead Sea, has been hyperarid since the early Pleistocene. This is further supported by a diminishing speleothem growth southward and a proposal for a predominating Atlantic–Mediterranean source of winter precipitation in the Negev during relatively short episodes of the last interglacial (Vaks et al., 2010). This apparently contradicts the hypothesis that a northward shift of summer rains from monsoonal sources to the southern Levant contributed to the slightly increased Dead Sea lake level during the early last glacial (Torfstein et al., 2015). Less attention has been paid to seasonal shifts in precipitation as a factor for lake level fluctuations. One reason is that most model studies focus on the summer season (e.g. Liu et al., 2004; Herold and Lohmann, 2009), while information about the winter season atmospheric circulation during intervals of maximum insolation is still scarce. One exception is the study by Kutzbach et al. (2014), which suggests that an increase in winter storm tracks could have caused the wetter intervals in the Levant during maximum Northern Hemisphere seasonality.

Present-day observations identified a third possible mechanism of moisture supply to the southern Levant. Winter to spring tropical plumes originating from the tropical eastern Atlantic and western Africa usually transport moisture across the Sahara into the southern Levant deserts when the subtropical jet is at a southern latitudinal position (e.g. Kahana et al., 2002; Rubin et al., 2007; Tubi and Dayan, 2014). Increasing the frequencies of such atmospheric circulation patterns that cause widespread ample rainstorms is probably the only way to increase the runoff yield in southern Negev drainage basins (e.g. Enzel et al., 2012) to a volume that will be noticed as a level change, although minor, in the Dead Sea. Low-latitude tropical plumes have been also proposed as a moisture source in the past when Northern Hemisphere insolation reached maxima during times of the last interglacial Lake Samra (Waldmann et al., 2009, 2010).

Disentangling the interactions of low-latitude/tropical and mid-latitude (Atlantic and Mediterranean) moisture sources and related mechanisms that triggered the reconstructed long-term and large-scale lake level fluctuations of the Dead Sea during the first 40 millennia of the last glacial is challenging and remains partly speculative. One reason for this difficulty might be that orbital-driven changes in insolation and seasonality are the common external trigger for both high- and low-latitude climatic fluctuations during that time. Nevertheless, the striking coincidence with palaeoclimatic records across the Mediterranean suggests a strong role of the Atlantic–Mediterranean atmospheric circulation for the moisture supply to the Levant during the last glacial inception.

The observed coincidence of a cold North Atlantic and dry southern Levant during the last glacial inception is apparently in contradiction with the long-term observations of glacial high stands and interglacial low stands at the Dead Sea. This apparent difference in the Dead Sea lake level response to North Atlantic climate changes at different timescales might be explained by threshold effects in the growth of the Fennoscandian ice sheet. Once the ice shield reached a certain height, it became a morphological barrier causing a major system shift in the atmospheric circulation pattern (Webb III et al., 1993) which, in turn, forced the Mediterranean storm tracks to shift southward and be funnelled and intensified towards the central Levant. This would explain the proposed doubling of annual rainfall in this region (Enzel et al., 2003, 2008) and the high Lake Lisan levels during the last glacial (Rohling, 2013). During the glacial inception, different atmospheric boundary conditions prevailed, likely because the ice shield elevation was still below the threshold, which forces the large-scale circulation pattern to change. To test this hypothesis, more high-resolution proxy records from the southern Levant and advanced modelling studies are needed.