Orbital‐ and Millennial‐Scale Changes in Lake‐Levels Facilitate Earthquake‐Triggered Mass Failures in the Dead Sea Basin

The possible link between the occurrence of submarine mass failure and climate‐driven factors is highly disputed. This is due largely to the lack of comprehensive records of mass failures in the geologic record for which ages, triggers, and preconditioning factors can be reliably constrained. Such controls would allow accurate testing of cause‐and‐effect relationships. We present a record that comprises 490 earthquake‐triggered mass failure deposits from the Dead Sea depocenter over the past 220 kyr that permits a robust statistical evaluation and correlation with potential preconditioning factors. Our data set reveals that (a) at the orbital‐ and millennial‐scale, variable sedimentation rates are not a preconditioning factor for these mass failure deposits; (b) at the centennial‐to decadal‐scale, earthquake‐triggered mass failures can occur at any lake‐level state; (c) at the orbital‐ and millennial‐scale, the mass failures are more frequent when lake‐levels were high and punctuated by large‐amplitude fluctuations.

with the same trigger is key to understanding the possible link (Pope et al., 2015;Smith et al., 2013;Urlaub et al., 2013). In addition to the role of water-level, previous studies have suggested that sedimentation-driven overpressure on subaqueous slopes can also decrease slope stability and facilitate mass failures (Leynaud et al., 2007;Sawyer et al., 2017;Ten Brink et al., 2016), but rapid sedimentation alone cannot facilitate mass failures (Cardona et al., 2020;Stigall & Dugan, 2010).
To address the contrasting viewpoints and interpretations noted above regarding the role of climate in driving mass failure frequency, we present a record from the Dead Sea depocenter that spans the last 220 kyr and comprises 490 earthquake-triggered (seismogenic) mass failure events. This case study provides a unique opportunity to shed new light on the debate by distinguishing and separating trigger and preconditioning factors. First, we focus on one specific setting in which the sediments experienced the same climatic and tectonic forcing (Lu, Bookman, et al., 2020;Lu, Waldmann, Ian Alsop, & Marco, 2017;Lu, Waldmann, Nadel, & Marco, 2017;Lu, Wetzler, et al., 2020). Second, it is established that earthquakes act as the trigger for these events (Lu, Waldmann, Ian Alsop, & Marco, 2017;Lu, Wetzler, et al., 2020;Lu et al., 2021). Third, the sedimentary sequence is well-dated (Goldstein et al., 2020) ( Figure S1, Table S1). Fourth, the slope morphology of the basin (Sade et al., 2014) and sedimentary processes at the lake margin (Alsop et al., 2016;Haliva-Cohen et al., 2012) and center (Lu, Waldmann, Ian Alsop, & Marco, 2017;Neugebauer et al., 2014) are well-understood. Therefore, this unique record allows a critical assessment and testing of the links between the occurrence of mass failure, changes in water-level, and sedimentation rate driven by a changing climate.

Mass Failure Deposits in the Dead Sea
The central and deepest part of the Dead Sea marks the depocenter of the Dead Sea Basin, which is the largest pull-apart basin developed along the Dead Sea Fault (Figure 1) Ben-Avraham et al., 2008). The Dead Sea is characterized by lake-level high-stands with large-amplitude fluctuations (>140 m) during glacials, while low-stands with low-amplitude fluctuations (<70 m) occurred during interglacials (Torfstein, Goldstein, Stein, & Knzel 2013;Torfstein, 2019;Waldmann et al., 2010). Aragonite-detritus laminae mainly accumulate during glacials, while halite and homogeneous mud are more widely deposited during interglacials (Kiro et al., 2015;Torfstein et al., 2015).
(a) Type I: Seismogenic sandy turbidites. These deposits comprise sandy turbidites that overlie in situ seismites (Lu et al., 2021) (Figures 2a and 2b), with further sandy turbidites that correlate to historic earthquakes (Lu, Wetzler, et al., 2020) (Figures 2c and 2d). (b) Type II: Laminae fragments-imbedded detritus layers (Figures 2e-2i). These deposits are characterized by sparse fragments of aragonite-detritus laminae within detritus beds and are formed by seismogenic slope failure-induced comminution of aragonite-detritus laminae (Lu et al., 2021). (c) Type III: Slump deposits (Figure 2j). These deposits feature intensely deformed and fragmented aragonite-detritus laminae with distinctive layer folding, which are considered to be sourced from middle and lower lake slopes (Lu, Waldmann, Ian Alsop, & Marco, 2017). (d) Type IV: Chaotic deposits ( Figure 2k). These deposits are characterized by mud containing coarse sands and well-rounded gravel, occasionally with some intraclasts, showing poorly sorted chaotic structures which are interpreted to be sourced from marginal lakeshore areas (Lu, Waldmann, Ian Alsop, & Marco, 2017).
In addition, thousands of intervals of homogeneous mud, turbidite, and debrite (on a centimeter to meter-scale) are also preserved in the Dead Sea depocenter ( Figure S2). However, a seismic or nonseismic origin cannot yet be established for these deposits. Mixed information is expected for a combined data set that compares deposits with different triggers. To test the possible link between changes in water-level and mass failures with the same trigger (Pope et al., 2015;Urlaub et al., 2013), we focus on earthquake-triggered mass failures, whereas mass failures with unclear triggering mechanism(s) are systematically excluded from the analysis.  the lake at a water depth of 2.1 m ( Figure 1). The recovered cores preserve sequences of aragonite-detritus laminae, halite, gypsum, and homogeneous mud ( Figure 3a). We correlate the main stratigraphic units at the orbital-scale between the two cores based on the gamma-ray response from the downhole wireline logs (Coianiz et al., 2019). This allows us to compare sedimentation rates and facies between the two different environments in the lake. LU ET AL.

XRF and Magnetic Susceptibility Scanning
The undisturbed core sections were scanned using an ITRAX core scanner at GFZ (Potsdam) with a Chromium tube at 30 mA current and 30 kV voltage. The scanning resolution is 1 mm and exposure time is 1 s (Neugebauer et al., 2014). We use titanium (Ti) as a proxy for clastic input and calcium (Ca) for carbonate and gypsum to investigate different facies (Lu et al., 2021). To eliminate constant-sum constraints and nonlinear matrix effects, we transform the elemental intensities into log-ratios (ln(Ca/Ti)) (Weltje et al., 2015). The core sections are also scanned at a 1-mm step for magnetic susceptibility by using a Multi-Sensor Core Logger with a Bartington MS2E sensor.

Chronology
Core 5017-1 spans the past 220 kyr (Goldstein et al., 2020;Kitagawa et al., 2017), with most dating uncertainty at <±1 kyr (Table S1; Text S2). We constrain the timing of each mass failure event by the age of the first underlying in situ sediment horizon and by linear interpolation. In some cases, erosion occurs at the base of mass-movements and sediment gravity flows (Alsop et al., 2019;Lu, Waldmann, Ian Alsop, & Marco, 2017), thus the calculated ages constrain a maximum age for the events.

Sedimentation Rate Calculation for the Lake Center and Margin
We calculate sedimentation rate, S, as: where D is the cumulative thickness of sediments deposited over the time interval Δt. Sedimentation rate calculation over the past glacials and interglacials (MIS 7-1) for the lake margin is based on correlation of the main stratigraphic units between Core 5017-1 and Core 5017-3 (Coianiz et al., 2019).
Reliable millennial-scale correlations between the central and marginal cores based on gamma-ray data can only be achieved for the last 20 kyr when evaporite layers formed throughout the lake floor and can be used as stratigraphic markers. Therefore, we obtain millennial-scale variations in sedimentation rate at the lake margin over the past 70 kyr by considering both the marginal Core 5017-3 (20-0 ka) and three Last Glacial (Lisan Formation) outcrops (70-15 ka; Table S2) (Torfstein, Goldstein, Kagan, & Stein, 2013).
We average sedimentation rates derived from the three Last Glacial outcrops at a 5 kyr interval because the sections are not complete at individual sites. We then integrate the averaged outcrop record and the upper slope Core 5017-3 record into one composite record in a 5 kyr bin size (Text S3, Table S2). We use this composite record to represent sedimentation rates on the upper lake slopes for the time 70-0 ka.

Data Binning and Probability Density Calculation for Events Over the Last 70 kyr
The sediments deposited during the Last Glacial in both Core 5017-1 and lakeshore sections are sparsely dated with gaps around 5 kyr. To compare the probability density of seismogenic mass failure deposits (hereafter referred to as events) with lake center and margin sedimentation rates, we group our data set of sedimentation rates and the events in 5 kyr bins.
We calculate probability density of events, P as: where n is the number of events over time interval Δt (=5 kyr) and N is the total number of events over the last 70 kyr. We define the probability density of events as extremely low or the highest when the values are ≤0.04 (standard deviation) and ≥0.08, respectively.
We compare the distribution of probability density of events in different sizes of bins (3.5, 5, 7 kyr). The three sets show similar distributions of probability density of events over the last 70 kyr ( Figure S3), suggesting that the shape of the event distribution is insensitive to changes in bin size. In addition, we also use the Gaussian Kernel and the Sheather-Jones method to determine the optimal bandwidth (3.01) and model the Kernel Density of events. The modeling yields a smooth curve that shows a similar distribution with the probability density of events binned in different sizes ( Figure S3).

Features of Seismogenic Mass Failure Deposits (Events)
Type I deposits show smaller fluctuations in Ca/Ti ratio than the aragonite-detritus laminae intervals (Figures 2a and 2b), suggesting mixing processes. Type II deposits also show smaller fluctuations in the Ca/Ti ratio than the aragonite-detritus laminae intervals underlying the units (Figures 2e-2i), thus also suggesting mixing. Type III and IV deposits both show frequent variations in magnetic susceptibility (Figures 2j  and 2k), indicating large-scale mixing processes and implying that the units have undergone significant transportation. We identify 490 such deposits in Core 5017-1 over the past 220 kyr.

Occurrence of Events Compared to Lake Levels and Sedimentation Rates During 70-0 ka
We identify 30 Type I events, 114 Type II events, 57 Type III events, and 8 Type IV events over the last 70 kyr (Figures 3b and S2; Table S4). The Type II and III events mainly occur during the Last Glacial (Figure S2). The lake-level underwent high-amplitude fluctuations (>140 m; Text S4) during the Last Glacial high-stand, compared to low-amplitude fluctuations (<70 m) during the Holocene low-stand ( Figure 3b) (Torfstein, Goldstein, Kagan, & Stein, 2013). At the centennial-to decadal-scale, the events occurred at all lake-level states of rising, falling, and stable intervening periods (<5 m/kyr; Text S4).
At a millennial-scale, the events have the highest probability density (>0.08) during 35-15 ka when lake-levels show the largest fluctuations but sedimentation rates are low (Figures 3c and 3d; Table S5). In addition, the probability density is extremely low (<0.04) during 15-0 ka when lake-levels show low-amplitude fluctuations but sedimentation rates are highest. The events also occurred frequently (0.04-0.11) during 70-35 ka, when sedimentation rates are low but lake-levels display frequent middle-magnitude fluctuations. The correlation coefficients between the probability density of events and the marginal and central sedimentation rates are both −0.4, indicating a weak negative correlation between them (Figure 3e). Additionally, we find no clear relationship between the thickness of event deposits and lake-level (coefficient: −0.02), and between the thickness of event deposits and marginal sedimentation rates (coefficient: −0.25) (Figures 4a and 4b).
During 42-0 ka, dating uncertainties are <±0.3 kyr (Table S1) and there is high confidence in lake-level history for the majority (70%) of the time interval, with low confidence for the remaining 22%, and only 8% of lake-levels unknown (Figure 3b). About 81% of failure events (N = 115) occurred during time intervals with high confidence in lake-level history. Among them, 87% of events (N = 100) occurred during lake-level rises/falls (spanning 82% of the total time), while 13% of events (N = 15) occurred during quiet intervals (spanning 18% of the total time) (Figure 4c). Our analysis reveals that the distribution of events over the three different lake-level states is not sensitive to the different definitions for "quiet intervals" ( Figure S4; Table S6).
LU ET AL.

Occurrence of Events in the Past Glacial-Interglacial Cycles
The frequency of events in the Penultimate Interglacial (MIS 7), Penultimate Glacial (MIS 6), Last Interglacial (MIS 5), Last Glacial (MIS 4-2), and Holocene (MIS 1) are 1.5, 3.4, 1.1, 2.9, and 1.7 events per kyr, respectively ( Figure 4d). The events occurred more frequently during glacials (with lake-level high-stands and large-amplitude fluctuations) than in interglacials (low-stands and low-amplitude fluctuations) (Text S5; Figure S5). The frequency of events is high during glacials when sedimentation rates are low and it is low during interglacials when sedimentation rates are high (Figures 4d-4f; Table S3). We observe negative correlations (−0.6 and −0.5) between the frequency of events and marginal and central sedimentation rates at an orbital-scale.

Discussion
Types I-IV events are only a partial recorder of the entire paleoearthquake catalog (Lu, Wetzler, et al., 2020), and therefore do not provide any inferences regarding earthquake recurrence. Moreover, the majority of in situ seismites in Core 5017-1 are not overlain by any types of the mass failure-related events, thereby indicating that mass failures do not develop with each significant seismic shaking or that those mass movements were too weak to reach the drill site. Therefore, some additional, nonseismic preconditioning factor(s) must have facilitated the seismically triggered mass failure deposits identified in Core 5017-1.
LU ET AL.

Implications of the Link Between Sedimentation Rates and Occurrence of Events
At the millennial-scale, the correlation coefficients between the probability density of events and lake marginal and central sedimentation rates are both −0.4 (Figure 3e). At the orbital-scale, the coefficient between the probability density of events and lake marginal sedimentation rates is −0.6, and the coefficient between the probability density of events and lake central sedimentation rates is −0.5 (Figure 4f). This new data set differs from observations in other basins where rapid sedimentation results in a reduction of submarine slope stability and thereby facilitates mass failures (Leynaud et al., 2007;Sawyer et al., 2017;Ten Brink et al., 2016). Rapid sediment loading may promote mass failures via increasing pore water pressure, thereby lowering the effective stress in a slope sequence (Leynaud et al., 2007;Sawyer et al., 2017). However, this process also depends on sediment properties (permeability and compressibility) and geometry of the slope sedimentary sequences (Flemings et al., 2002;Urlaub et al., 2015). In the Dead Sea case, comparable sedimentation rates in the basin depocenter and on the slope during MIS2-4 suggests sediment draping and thus no major changes in depositional geometry that could drive lateral fluid flow. Our data set over different time-scales suggests that variation in sedimentation rate is not the main preconditioning factor for mass failures during seismic shaking in the Dead Sea. This suggests that in combination, sedimentation rates were too low and permeability was too high for the generation of excess pore pressure and lowered effective stress in the upper lake slopes due to sedimentation alone.

Effect of Lake-Level Change on the Occurrence of Events
Our data set reveals that at the centennial-to decadal-scale, the events are not statistically correlated with lake-level state. The percentage of events that occurred during quiet intervals (13%) is close to the total time occupied by stable lake-levels (18%). Additionally, the time resolution of lake-level reconstruction and the time lag between the lake-level effects and mass failures are difficult to estimate for individual events at specific times (Text S6). Thus, the statistics at the centennial-to decadal-scale do not allow us to make a definite statement regarding a possible link between lake-level changes and mass failures. However, at the millennial-and orbital-scale, we find a statistical correlation between the lake-level and the frequency of events. Higher frequency of events (during 35-15 ka, MIS 6, and MIS 4-2) are statistically correlated with high lake-levels that were punctuated by large-amplitude falls/rises.
Field investigations and modeling in the Dead Sea Basin have confirmed that incision of stream channels and associated bank failures occurred during rapid falls in lake-level (Closson et al., 2010;Dente et al., 2021;Hassan & Klein, 2002), indicating an overall decrease in slope stability around the lake margins. A similar mechanism of falling water-levels promoting mass failures, has also been reported in Laguna Potrok Aike, southern Patagonia (Anselmetti et al., 2009), and along the continental margin of New Jersey (McHugh et al., 2002) and the South-Chilean active margin (Blumberg et al., 2008).
In the Dead Sea case, relatively low sedimentation rates are statistically correlated with lake-level rises and high-stands at both the millennial-and orbital-scales (Figures 3b-3d and 4e) (Lu, Bookman, et al., 2020;Lu, Waldmann, Nadel, & Marco, 2017), and thus do not support mechanism "(i)." Regarding the enhanced seismicity hypothesis, a complete paleoearthquake (M w ≥ 5) record that covers the Last Glacial is currently lacking and we are therefore unable to evaluate this model.
In the Dead Sea Basin, the very high lake-levels during glacials (>−330 m; Figure 3b) inundated more marginal slopes with steeper gradients (7°-21°) and loaded these slopes with loose sediments (Figures 1c  and S6). These changes can increase the overall subaqueous slope failure potential during seismic shaking and lead to the accumulation of these event deposits in the lake center. Also, the rapid and high-amplitude drops in lake-level during glacials may have enhanced slope erosion and decreased terrestrial slope stability around the lake (Closson et al., 2010;Dente et al., 2021;Hassan & Klein, 2002), creating preconditioned slopes for seismogenic mass failures.

Other Lake-Level Change-Connected Factor(s) and Implications
Another important factor is that sedimentary sequences deposited under different climate conditions and lake-level states on the upper lake slopes (lake margin) comprise different lithologies. Aragonite-detritus laminae are mainly deposited during wet glacials and wet periods of interglacials when the Dead Sea lake-level is relatively high and brine is oversaturated for calcite (Barkan et al., 2001;Stein et al., 1997). The highest probability density of events occurred during 35-15 ka when lake-levels were high (> −330 m; Figure S2) and submerged more marginal slopes with steeper gradients (7°-21°) and charged these slopes with aragonite-detritus laminae.
During 35-15 ka, 66% of the events are Type II deposits which are comparable in composition to aragonite-detritus laminae (Figure 2), with thickness mostly <5 cm (Table S4), and unrelated to other typical preconditioning factor(s) for subaqueous landslides such as enhanced sedimentation rate (Figure 3). Surficial sediment remobilization triggered by earthquakes that is, reported from Chilean lakes and offshore Japan may explain these new observations of very thin layers of mixed sediments from the Dead Sea (McHugh et al., 2016;Moernaut et al., 2017;Molenaar et al., 2021). These factors connected to lake-level change can collectively favor the observed pattern of seismogenic mass failures in the Dead Sea.
Our study implies that to reliably test the cause-and-effect relationships between mass failures and climate-driven factors, for example, changes in water-level and sedimentation rate, at a much higher time-resolution, one should focus on suitable sequences. Such sequences need to comprise well-dated mass failure deposits from one setting with a well-constrained history of water-level changes at the centennial-to decadal-scale. Also needed is a quantification of preconditioning factors through (in-situ) geotechnical data and/or slope stability modeling. Our study highlights the unique potential of deep drilling and lacustrine sequences to better understand and constrain the link between mass failures and potential climate-driven factors, which can serve as an analog for marine investigations.