• from Table S1 of Electronic Supplement to Wechsler et al. (2014)

Table S1. Unit Descriptions for Channels 3 and 4

Unit Number	Description
301	clay with carbonates, brown
305	sandy clay, yellow
308	sand
310	pebbly sand
320	cross bedded sand and sandy gravel
322	sandy gravel
323	sand
324	cross-bedded sandy gravel with manganese staining
324	clayey sand
325	well sorted sand, sometimes with foresets
326	sand, cross bedding
328	pebbles and cobbles
328	basalt gravels, pebbles and cobbles
329	pebbly sand with fossils
329	clayey pebbly sand + shells
330	sandy clay
332	clayey gravely sand
333	sandy clay
334	silty sand
335	clayey sand
337	sand
340	pebbly sandy clay, grey
342	clayey gravel
345	silty clay
349	clayey fine gravel
350	clayey sand
355	clayey gravel
360	clayey coarse gravel and pebbles
365	gravely sand
370	clayey gravel
372	sand
373	clayey gravel
375	clayey sandy gravel
380	clayey sand
382	clayey sandy gravel
384	clayey coarse sand
385	clayey sand
387	sandy gravely clay
388	silty clay
390	sandy gravel
392	sandy clay
394	sandy gravel
395	clayey pebbly gravel
396	pebbly sand
397	clayey gravel
398	clayey sandy gravel
399	clayey pebbly gravel
400	dark brown clay below channel deposits
405	brown clay
415	calciferous clay with shells
420	sandy clay
425	sandy gravelly clay
429	sandy gravelly clay
431	clayey gravelly sand
432	clayey gravelly sand
433	clayey gravelly sand
434	gravelly sand
435	gravelly sand
436	gravelly sand
437	gravelly sand
438	gravelly sand
439	gravelly sand
440	clayey gravel
441	silty clay
442	clayey gravel
443	sandy clay
445	clayey gravel
449	clayey gravel
450	clayey gravel
452	sandy gravel
453	sandy gravel
454	sandy gravel
455	sandy gravel
456	sandy clay
457	sandy gravel
458	sandy gravel
459	silty clay
460	sandy clay
480	clayey gravelly sand
481	clayey gravelly sand
482	clayey gravelly sand
483	clayey gravelly sand
484	clayey gravelly sand
485	clayey gravelly sand
486	clayey gravel
487	clayey gravel
488	clayey gravel
489	clayey gravel
490	gravelly sand
491	silty clay
492	gravelly sand
493	gravelly sand
494	silty clay
495	gravelly clay
496	gravelly clay

• from Wechsler et al. (2014:1333-1337)

Channel 3 is a west-flowing sandy gravelly channel complex that was crossed in several fault parallel trenches (T30, T33, T34, and T38) and in one fault- crossing trench (T45). The channel units were numbered between 300 and 399 (older units higher numbers) for reference. T45, the fault-crossing trench, also represents a longitudinal profile of channel 3 and was used to study earthquake history. The logs of T45 are presented in Figures 3 and Ⓔ S1a–b, and the unit nomenclature is described in Table S1 available in the electronic supplement.

Channel 3 was divided into lower and upper units based on the channel morphology, with the upper units (units 306–329) cutting into and eroding the lower units (units 330–399) east of the fault (T45 log, Figs. 3 and Ⓔ S1a–b available in the electronic supplement). These upper units strata were not observed in fault contact on the west side of the fault. The youngest units (301–305) are capping the faults on the west side and appear unrelated to the upper channel 3 units.

The strata in channel 3 vary in composition from large, rounded pebbles and cobbles at the base of individual subchannels in the lower part of the section, to foreset- bedded gravelly sand and silty clay in the upper part. Furthermore, the upper part (units 310–329) of channel 3 exhibits an anomalous trend on the east side of the fault zone, with indications of an along-fault flow direction (northsouth). The upper part also has many gastropod shells within it, indicating a rise in lake level and probable influence from motion on the eastern fault strand, as discussed later. The foreset bedding (units 320–328) is also consistent with deposition in standing water, which supports the inference of an increase in lake level.

We collected nearly 100 samples of detrital charcoal from the strata of channel 3, of which 36 were dated by 14C mass spectroscopy methods at the Center for Accelerator Mass Spectrometry (CAMS) facility at University of California, Irvine. Twenty-eight of these dates are from trench T45 or from strata exposed in fault- parallel trenches near T45 that could be confidently traced into the strata of T45 (indicated as proxy dates on the trench logs) and are used in this paper. Many of the samples yielded dates that are much older than other samples from the same stratum, including one as old as 12,000 years. Consequently, not all were used in the chronologic model to constrain the ages of faulting events exposed in T45.

Taking the youngest dates from each unit as closest to the actual age of their respective strata, and discarding dates that are out of stratigraphic sequence, we are left with 10 dates that we used in our chronologic model (Fig. 4). From these, the ages of the lower strata from channel 3 range from the mid-fifth century C.E. to the mid-seventh century C.E. whereas a single date from the upper portion of channel 3 yielded a mid-seventh to mid-eighth century date.

As an additional age constraint, we used 5 of the 11 radiocarbon dates from channel 2 that span the entire section of the channel as an upper-bound constraint on the ages of strata in channel 3, based on cross-cutting relations from fault parallel trenches (ⒺFig. S1i available in the electronic supplement). All five dates yielded ages in the late-seventh to mid-eighth century C.E., consistent with the date from the upper part of channel 3 and suggesting there is not much time between the abandonment of channel 3 and formation of channel 2, although the stratigraphic evidence indicates a rise in lake level. A lake level rise may have forced the avulsion from channel 3 to channel 2, an event that may have occurred in only a few decades, or less, as suggested by the 14C dates.

We also used Marco et al. (2005) samples to constrain the youngest event in channel 3. Dates are modeled using OxCal (BronkRamsey, 2009) and IntCal09 calibration-curve (Reimer et al., 2009), as summarized in Figure 5 and Table 2.

Trench T45 exposes evidence for two surface ruptures captured in the stratigraphy of channel 3 (Fig. 3), which we number CH3-E1 and CH3-E2. Event names represent the channel in which they are identified (the first number) and their stratigraphic order from younger (E1) to older (E2). There is also evidence for displacements younger than E1 that involved the overlying strata, and these are most likely the previously documented 1202 and 1759 C.E. earthquakes and any other events that postdate channel 3.

The fault zone itself, as exposed in T45, is divided into two main strands, 1–2 m apart, plus several subsidiary faults. The eastern strand truncates the manganese-stained, cross- bedded upper channel 3 sandy gravel strata (units 320–329). Within and west of the fault zone, units attributed to the lower section of channel 3 are preserved across all fault strands and contain the evidence for the surface ruptures. Hence, the two surface ruptures we identify fall in the time frame of the lower channel 3 section, or between about the mid-fifth to mid-seventh centuries C.E. There are no equivalent upper channel 3 units exposed in T45 west of the eastern fault branch, except the capping units 301–305, which only appear west of the eastern fault branch, and therefore their stratigraphic relation with the younger units west of the fault is unclear. This could be a result of horizontal offset, of local changes in the channel flow near the fault zone, or both.

Event CH3-E1 is expressed on both trench walls as an upward truncation of fault strands that are capped by unit 305. In T45N, within the fault zone along a secondary fault strand, there is a large fissure that contains rotated blocks of coherent stratigraphy floating inside more massive fissure- fill material (between 14 and 15 m). Another fault strand that bounds the fault zone on the west also appears to rupture to the same stratigraphic position (at 17 m), and both faults and the fissure fill are capped by unit 305, whereas all other fault strands rupture to higher levels and presumably moved in later earthquakes. On the south wall of the trench, a single fault strand was found that ruptured up through the channel stratigraphy and was capped by unit 305 (at 16.5 m). Taken together, we consider this strong evidence for a surface-rupturing event that occurred high in the section of the lower channel 3 alluvial fill. We speculate that this event, which produced uplift of the central block within the fault zone, may have caused local damming of channel 3, thereby disrupting the stratigraphy of upper channel 3 deposits.

Event CH3-E2 is also well expressed on both trench walls, with lower channel 3 alluvium of units 380–384 folded or tilted by as much as 30° to the west and in fault contact, followed by truncation of these deformed strata and deposition of undeformed lower channel 3 alluvium. Both the deformed strata and secondary fault strands are capped by unit 375, so we infer this contact to be the event horizon.

The ages of the surface-rupturing events interpreted in the sediments of channel 3 are presented as probability density functions in OxCal (Bronk-Ramsey, 2009) in the chronologic model in Figure 5. Based on this OxCal model, event CH3- E1 falls in the range of 619–684 C.E., with a peak probability at about 653 C.E., whereas CH3-E2 is dated to between 505 and 593 C.E., with a peak probability at about 551 C.E. Because the probability distributions are nearly symmetric, we present them as 653 ± 36 C.E. and 551 ± 42 C.E., with uncertainties reported at 2σ.

• from Wechsler et al. (2014:1337-1342)

The sandy channel complex of channel 4 was exposed in trenches T33, T37, and T39 (Figs. 6–8), with the latter two trenches crossing the fault. Captured within the strata of channel 4, we found evidence for up to six paleoearthquakes, which are numbered CH4-E1 through CH4-E6. The stratigraphic units in channel 4 are numbered 400–499, from youngest to oldest, and represent nearly continuous deposition of sand, gravel, and mud across the fault for several hundred years.

Channel 4 crossed the fault in an area where a long, linear, and narrow pressure ridge is interpreted to have caused a very localized uplift within the fault zone itself. This is obvious in the expression of faulting, as exposed in trenches T37 and T39, where the strata of channel 4 are warped up into the fault. It is partly due to this localized style or expression of the fault that we were able to confidently distinguish individual faulting events.

Trench T33 was excavated west of the fault to identify channel locations, but we use the stratigraphy exposed in the vicinity of trench T37 and T39 to wrap all stratigraphic units in all three exposures, so correlation of strata are certain. From this exercise, two other important elements are apparent. First, the stratigraphy in trench T37 is mostly older than the strata exposed in T39, because T33 shows that strata offlap to the north west of the fault zone. Thus, as seen by the depositional relations between the units in T33 (Figs. 7 and ⒺS1f available in the electronic supplement), strata are progressively older to the south on the west side of the fault, which is consistent with a model in which the faults motion during the period of channel flow created a progression of overlapping subchannels west of the fault, from the oldest in the south to the youngest in the north, consistent with the faults left-lateral motion.

Second, there is a locally significant down-to-the-west component of vertical motion, although a lesser down-to-the-east component is observed on the east side of the pressure ridge. Thus, from the exposures in T33, T37, and T37, the stratigraphy is most consistent with the overall model of left-lateral displacement across a narrow pressure ridge with central uplift and an overall west-side-down long-term component of vertical motion. It is within this context that we discuss the evidence for multiple surface ruptures in the channel 4 complex.

The overall chronology of the channel 4 complex is provided by the radiocarbon dating of 24 samples from these three trenches. As with all detrital charcoal, some samples yielded ages that are too old relative to other sample ages, reflecting either the burning of older wood or their residence within the system for an extended period of time. Nevertheless, 10 of the samples yielded ages in stratigraphic order, assuming the youngest dates from each strata are closest to the actual depositional age of a stratum, and we use these dates to construct a chronologic model to constrain the event ages.

From these dates, channel 4 ranges in age from at least as early as the first century C.E. up through the fourth century C.E., although the lowermost sandy gravel of channel 4 remains undated and may extend back into the latest part of the third millennium, as its age is only constrained by one sample (sample 364, see Table 2) taken from an older channel below channel 4 (Ⓔ Fig. S1c available in the electronic supplement).

Evidence for the oldest of the interpreted events, CH4-E6, is captured in the basal channel 4 deposits exposed in trench T37 (Figs. 6, Ⓔ S1d–e available in the electronic supplement). Ruptures are indicated by both upward termination of individual faults and folding, with angular unconformities identified that resulted from the folding events. Deposition of growth strata and possibly a colluvial wedge, along with fissures capped by undisturbed strata, support the interpretation of an event at this horizon. In the north wall of T37, units 490, 492, and 493 comprise a sandy gravel deposit that is strongly disrupted by many small faults and probable liquefaction. Units 491 and 494 are silty-clay layers and are strongly deformed at the bottom of the trench. This section is folded up onto the pressure ridge and generally maintains a similar thickness, indicating deformation took place after the deposition of unit 490. Units 480 through 489 thin onto the scarp/pressure ridge, indicating deposition after the deformation associated with event CH4-E6, and unit 489 caps many small faults that disrupt strata up through unit 490. Furthermore, units 480–489 thicken across the older folded strata, indicating postevent growth. Similar relationships are observed on the south wall of the T37, with units 480–489 capping many small faults and thinning onto the pressure ridge. These relationships argue that event CH4-E6 produced significant deformation at this site and was likely a relatively large earthquake.

Event CH4-E5 is interpreted from many small faults that break to the top of unit 480 and are capped by units 450–469. On the south wall of T37, stratigraphic growth is seen where units 450–460 thicken in the axis of a small syncline in the area of the greatest change in bedding dip, and these strata thin onto the fold scarp of the pressure ridge, thereby also supporting growth of the primary structure. However, the amount of deformation appears relatively minor when compared with that in event CH4-E6, so we interpret this as a smaller event.

Evidence for event CH4-E4 is observed in both trenches T37 and T33, and there is also evidence at the bottom of the south wall of T39. In T37, several small faults rupture up through units 450–469 and are capped by unbroken strata of units 440–449, and this relationship is consistent on both walls. Included within these many faults is a significant fissure exposed on the south wall that is either capped or filled by unit 449. Further, units 440–449 thin and pinch out onto the fold scarp indicating growth of the fold/pressure ridge after deposition of unit 450. In trench T33, minor faults displace strata up through unit 450 and appear capped by unit 440–449, which appeared as a single stratum in this exposure (Figs. 7, Ⓔ S1f available in the electronic supplement). Altogether, the evidence is strong for the occurrence of an event between deposition of units 449 and 450, although the amount of deformation appears less than that associated with event CH4-E6.

Evidence for the youngest three interpreted events in the channel 4 complex is observed primarily in trench T39 (Figs. 8, Ⓔ S1g–h available in the electronic supplement), although evidence for event CH4-E3 is also seen in T33. In fact, it is in T33 where stratigraphic separation can be demonstrated between events CH4-E4 and CH4-E3, with CH4-E4 breaking up through unit 450 and capped by the sandy-clay strata of units 440–449, and CH4-E3 breaking through 440–449 and into the bottom-most unit (439) of the gravelly sand package of units 430–439. The gravelly sand fills down into the fault, indicating the event occurred during deposition of this channel deposit and was later capped by similar gravelly sand units.

The same deposit (units 440–449) is identified in trench T37, where it is above the event horizon for CH4-E4, and it is exposed in trench T39, where CH4-E3 is interpreted at precisely the same level. On both walls of T39, many fault strands break into the 439–449 units package, with the sand and gravel filling into each fault. Many of the faults re- ruptured in subsequent events, but many did not. Many of the faults are capped by strata of units 430–438, which are similar to unit 439. These units also thicken in a shallow synclinal form that we interpret as produced by event CH4- E3. Finally, strata of units 430–438 thin onto and terminate against the fold scarp of the pressure ridge. One interesting aspect of the T39 exposure is that units 440–449 also appear to thicken into the same synclinal trough as the overlying strata of units 430–439. As these are the basal channel deposits exposed in this trench, this observation suggests that units 440–449 filled a synclinal depression adjacent to the linear pressure ridge, consistent with an event immediately prior to their deposition. This is stratigraphically consistent with event CH4-E4. Evidence for event CH4-E2 is weaker than that of some events, with several small faults terminating at the top of unit 430, and capped by unit 429. On the north wall of T39, units 425–429 thicken in the synclinal trough, arguing for fold growth after unit 430. Also on the north wall, there are several small faults filled with sediments of units 430– 438 (undetermined due to mixing?) that are capped by unbroken strata of unit 429. Collectively, we argue for an event horizon between units 429 and 430 that exhibits relatively minor overall deformation.

Finally, event CH4-E1 breaks up through unit 425 and into unit 420 on several faults on both walls of T39. There is no clear event horizon because the strata are relatively massive, but there is a thickening of unit 420 in the synclinal trough, arguing that the deformation is synchronous with the deposition of unit 420, which we consider as the approximate event horizon. It should be noted that each of the identified events is based on local evidence within the strata of channel 4, whereas there has clearly been substantial additional deformation of the section as a result of more recent ruptures. This is seen by the substantial warping and displacement of units 425 and all overlying strata, where preserved. In the capping lake clay, shears may be evident above recognizable faults that penetrate the base of the clay but are then obscured and completely transparent in the massive clay itself.

The ages of individual rupture events in the channel 4 complex are best constrained by 10 radiocarbon dates from samples taken from within the channel 4 deposits, age constraints from the younger channel 3 complex, and from a single charcoal sample from underneath the channel 4 complex. The lack of good age control on the basal channel 4 deposits in trench T37 limits our ability to precisely date event CH4-E6, although additional historical information allows some additional constraints. Using the OxCal model developed for all of channels 3 and 4 and adding the additional date from underneath channel 4, we calculate the probability distributions for each event from CH4-E6 through CH4-E1 (Fig. 5). Event CH4-E6 is the least well constrained because we have no dates from the basal faulted deposit, and therefore the age is only constrained by a sample that is likely considerably older, as it predates the channel 4 complex altogether. The probability distribution exhibits a boxcar distribution that places the age of the event between 400 B.C.E. and 100 C.E. However, as the basal deposits are clearly associated with channel 4 and are probably not substantially older than other basal deposits of channel 4, we assume the base of channel 4 is on the young side of this distribution, likely making this event fall within the range of the first century B.C.E. to first century C.E.

Events CH4-E2 through CH4-E5 all have direct radiocarbon control, although the strata between events CH4-E3 and CH4-E4 are only dated by a single sample. Nevertheless, the calculated ages for the events are as follows: event CH4-E5 dates between 137 and 206 C.E. with a peak probability at 169 ± 38 C.E.; event CH4-E4 dates between 165 and 236 C.E., with a peak probability at 204 ± 34 C.E.; event CH4-E3 dates between 250 and 310 C.E., with a peak distribution at 277 ± 30 C.E.; and event CH4-E2 dates between 269 and 329 C.E., with a peak probability at 299 ± 30 C.E.

Event CH4-E1 occurred during the deposition of unit 420, and this unit has only a single date that provides some age constraint. The sample was taken from near the upper boundary of this unit, so we assume the sample age postdates the timing of event CH4-E1 because unit 420 appears to thicken in the synclinal axis, which suggests the event occurred early in the deposition of unit 420. With this assumption in place, we calculate the age of event CH4-E1 to be in the range of 294–369 C.E., with a peak probability at 326 ± 36 C.E. If the date actually is in the faulted part of unit 420, then the date of event CH4-E1 may be as much as a century younger. However, it cannot be as young as the basal deposit of channel 3 unless both channels were active at the same time, which we consider unlikely.

• from Wechsler et al. (2014:1342-1343)

We present evidence for eight events that pre-date the 1202 and 1759 C.E. earthquakes identified by Marco et al. (2005). The youngest of them, event CH3-E1, dates to the mid-seventh century C.E. Marco et al. mention 685 C.E. as an upper bound date of the offset units that recorded 2.7 m offset (Marco’s CH2, CH3). However, they do have younger dates from the same unit, so, assuming the older dates are residuals, this upper bound is pushed to the beginning of the eleventh century C.E.

There is no lower age constraint because they did not date samples from below those units. Based on the detrital carbon ages from the channel units discussed above, we consider the period between the first and eighth centuries C.E. to be well covered, with the exception of the first half of the fifth century C.E. (Fig. 9). The nearly continuous sedimentary record for that period recorded eight earthquakes, and it is reasonable to assume this is a complete record of surface-rupturing earthquakes on the JGF for that period. There may be a missing record for the period between the deposition of channel 2 and channel 1.

The historical catalogs mention a destructive earthquake and landslide in Tiberias, on the west coast of the lake, in 850–854 C.E. We do not have channel deposits with ages that cover the period and therefore cannot determine whether this earthquake ruptured at the Beteiha site.

• from Wechsler et al. (2014:1343-1344)

Table 3 and Figure 9 summarize the event ages obtained from the OxCal model and compare them with known his- torical earthquakes from that period. The uncertainty regard- ing the age of the oldest event in channel 4 makes finding the equivalent historical earthquake difficult; however, if indeed the lower part of channel 4 is closer in age to the rest of the channel units, then it is reasonable to assume the event oc- curred around the turn of the millennium and could correlate with the 31 B.C.E. earthquake of Herod’s time. Earlier can- didate events include 92 B.C.E. and an earthquake in mid- second century B.C.E., but the location of those events is not clear (Table 3).

In the historical records there is no mention of any earth- quake anywhere along the DST between 130–303 C.E., and the same is true for the Dead Sea cores (Kagan et al., 2011). Channel 4 units span this period of time, and in them we detected evidence for 3–4 events with model ages that over- lap this time period (events CH4-E5 through E2), and one of those events fall completely within the supposed seismic qui- escence period (CH4-E4). It is well known that the historical earthquake record is incomplete in general, and specifically for that time period (Late Roman) during which the Roman empire was in decline, and the main historical texts describ- ing this period are compilations written by historians who lived during later periods. It is therefore likely that a mod- erate earthquake could occur and not be recorded in either historical sources or by the Dead Sea seismites, yet cause sediment disruption and surface rupture along the JGF. An earthquake of magnitude ∼6–6:5 would rupture the surface at Bet-Zayda Valley and be detectable in trenches (e.g., Liu- Zeng et al., 2007) but not by Dead Sea seismites that are more than 100 km away (figure 9 of Kagan et al., 2011).

In the historical earthquakes catalog for the time period equivalent to the paleoseismic record, there are a few earth- quakes with evidence of damage that are not centered along the JGF or in northern Israel. For example, the 551 C.E. earthquake is thought to have ruptured offshore Lebanon, on the Mount Lebanon Thrust (Elias et al., 2007). Yet we must consider the possibility that the rupture extended to the south, along the Roum fault (Darawcheh et al., 2000) and onto the JGF (Wechsler et al., 2009) or that what is described as one earthquake in the historical record was actually a series of events, one triggering the other, that were amalga- mated in the historical record. Some of the ruptures recorded in the paleoseismic record could be either the rupture tip of an earthquake that originated on another fault segment (north or south of the site) or a triggered event on the JGF that would not merit its own historical mention. An example for an earthquake that originated to the north is the 1202 C.E. event, which was centered on the Yammouneh fault but also ruptured all the way south through the JGF and maybe even farther south.

Other known historical earthquakes that were not cen- tered on the JGF, but which may have either ruptured to the Sea of Galilee or triggered slip or an aftershock include earthquakes in 33, 130, 303, 347, 363, 500, 634, and 660 C.E. (Fig. 9, Table 3). Considering that all detrital charcoal ages are maximum ages for the host deposit because the date reflects the growth and death of the original wood rather than its burning or deposition, some of the events we identify and date in the channel stratigraphy may have a slight bias toward an older age, in which case most may actually be represented in the historical record. In any case, a slight shift in event ages does not affect our overall conclusion that there was a cluster of earthquakes that produced surface rupture along the JGF during the first millennia C.E., followed by a relative dearth of events in the second millennia C.E.

• from Wechsler et al. (2014:1344-1345)

In the past 2000 years, we observe evidence for a total of 10 surface-rupturing earthquakes, of which seven or eight events occurred in the first millennium, compared to just two in the second millennium C.E. This demonstrates that the fault is not behaving in a periodic fashion on a scale of 2000 years and several earthquake cycles. Based on model ages and taking into account the uncertainties in modeled event ages, the overall earthquake occurrence interval for the last 2 millennia is 199 ± 111 years. When computed separately for each millennium, it is 553 ± 32 years for the second millennium and 80 ± 106 for the first millennium (excluding CH4-E6, the oldest event, for which the lower age constraint is poor). These recurrence values do not account for differing earthquake magnitudes and demonstrate how the recurrence interval can be a misleading quantity when trying to estimate the regional earthquake risk.

The past 1000-year period appears deficient in strain re- lease, starting from the Lebanese restraining bend (Däeron et al., 2007), through the Jordan Valley (Ferry et al., 2007) and southward to the Gulf of Aqaba (Klinger et al., 2000). Thus, in terms of moment release, most of the plate boundary has remained locked and has been accumulating elastic strain, as supported by recent GPS data (Sadeh et al., 2012). In contrast, the preceding 1200 years or so experienced a spate of earthquake activity, with large events along the Jordan Valley segment alone in 31 B.C.E., 363, 749, and 1033 C.E. (Guidoboni, 1994; Marco et al., 2003; Guidoboni and Comastri, 2005). Thus, the recurrence interval appears to vary by a factor of two to four during the historical period in the Jordan Valley, as well as at our site.

Studies of the recent seismic activity and frequency– magnitude relations in northern Israel suggest an estimated return period of 340 years for M >6 and 4500 for M >7 earthquakes using the Gutenberg–Richter (GR) relation, with a corresponding slip rate of 1.9 mm/yr (Shapira and Hofstetter, 2002). Hough and Avni (2009) combined available instrumental and historical earthquake data for the region and surmised that the GR distribution is valid for the DST. We compare the earthquake occurrence rate for the Sea of Galilee and Hula area from modern seismic records (Shapira et al., 2007) to the recurrence rate of large earthquakes based on paleoseismic records (Fig. 10), assuming a magnitude equal to or larger than 6.5 ± 0.5 for the surface-rupturing events. The result does not follow the GR distribution; rather it is similar to the characteristic earthquake distribution, with a discrepancy in the rate of small-magnitude earthquakes, as also observed on other faults (Hecker et al., 2013).

This discrepancy can be explained if we consider that the instrumental record is missing both moderate-to-large events and their aftershocks, which can decrease the a- value of the GR distribution by a factor of 2 (Page et al., 2008). A possible explanation for the earthquake behavior observed at the Beteiha site is that it is affected by its southern neighboring segments, as there is the possibility that earthquakes nucleating in the Jordan Valley can rupture through the Galilee stepover to the south of Bet- Zayda (Fig. 2). Ruptures originating from the north are also likely, as demonstrated by the 1202 earthquake (Marco et al., 2005). Another possibility is that large earthquakes on the Jordan Valley, Yammounneh, Roum, or Sergaya segments may trigger smaller aftershock events on the Jordan Gorge segment, in which case the historical record may tend to amalgamate any evidence for multiple, closely timed events into one large event.

The temporal variations in earthquake production may conform to a slip-predictable fault behavior, but more slip per event data are needed to determine the validity of the slip-predictable model for the JGF. A scenario of unzipping of the whole DST system, similar to the behavior of the North Anatolian fault (Stein et al., 1997), can account for periods of lesser activity. The nonperiodic behavior of the DST over the millennial timescale makes it more difficult to meaningfully predict the probability for a large earthquake soon. It may be that the 1995 M 7.3 Aqaba earthquake is the first in a sequence of future earthquakes that will soon be followed by several large earthquakes from south to north. It is also possible that a new cluster of moderate earthquakes was initiated with the October 1759 earthquake on the JGF, in which case we might expect several such events prior to a repeat of a 1202-type very large rupture. Regardless, in terms of moment release, most of the fault has remained locked and is accumulating elastic strain. Therefore, it is imperative to prepare for a large earthquake on the DST, which will occur sooner or later.

• from Wechsler et al. (2018:219-221)

Table 3 and Fig. 8 summarize the event ages and the offsets attributed to each event. The earthquake ages were used for calculating the mean recurrence time between events and the corresponding coefficient of variation on timing (CVt), using a Monte-Carlo approach: the probability density functions (pdfs) of event ages are sampled thousands of times and the mean μ, standard deviation σ and CVt are calculated on the sampled set (Biasi, 2013; Zielke et al., 2015). The resulting mean recurrence time is 190 years and the CVt is 1.05 (Fig. 8b), which implies elevated periods of seismic activity followed by periods of quiescence, or clustered behavior. In the following section we discuss our results in light of the historical record of earthquakes and the known paleoseismic record on neighboring faults.

The 1759 CE and 1202 earthquakes have 0.5 and 2.2 m of horizontal displacement attributed to them, respectively; the 1202 offset is the largest slip-per-event recorded at the site. The pre-1202 offset of Channel 2 of about 1.3 m is interpreted to represent slip in one moderately large earthquake. Channel 3 records one moderately large slip event of about 1.2 m, and possibly one additional small earthquake based on the cross-fault stratigraphy (Wechsler et al., 2014), although the displacement is virtually the same as that of Channel 2 suggesting that events CH2-E1 and CH3-E1 are the same event as indicated by their nearly identical ages.

Deposits in Channel 4 record evidence for six earthquakes in addition to those of Channels 1 through 3 (Wechsler et al., 2014); two of them affected the upper units with resolved additional cumulative slip 2.7+2/−2.5 m. Likely both are moderate in size, each with about 1.3 m of horizontal slip, although it is possible that one of the two events may have significantly larger offset, as in 1202 CE, while the other may be a smaller, 1759-style event. We cannot rule out the possibility that there may be >2 events that contributed to the difference in offset between channels 3 and 4, but we lack records for such events due to a disconformity in the deposition sequence.

From historical records and paleoseismology, the Yammouneh segment ruptured in 1202, between 405 and 945 CE and between 30 BCE and 469 CE (Daëron et al., 2007; Fig. 8). The penultimate Yammouneh rupture could be one of the mid-8th century earthquakes (CH2-E1 or CH3-E1, which are probably the same event). During that period, three distinct earthquakes occurred in the region and were felt from Syria in the north to Petra in the south (Fig. 1a): the first around 746 CE, the second in 749 CE and the third around 757 CE, and while the second of which most likely ruptured the Jordan Valley segment, the first or the third could have been associated with the Yammouneh fault (Ambraseys, 2005, 2009).

The JVF is interpreted to have ruptured in the earthquakes of 1033 CE, 749 and 363 based on both paleoseismic and historical records (Alfonsi et al., 2013, Ambraseys, 2009, Ferry et al., 2011, Fig. 8). Thus, the events with resolved offset at the Beteiha site could have from local earthquakes on the Jordan Gorge segment of the DST. This does not mean that the Jordan Gorge fault ruptured at precisely the same time, as historical records often refer to an earthquake as lasting for weeks or months and a large aftershock on the Jordan Gorge fault could have been amalgamated into the main-shock event. Thus, the observed displacements that we document here (CH2-E1) could have been coseismic with a well-known historical event or they could have been separate events that were induced or triggered by the larger regional events, although in this case the displacement may be smaller than the detection threshold of this study.

The first event documented by Wechsler et al. (2014) in channel 3 sediments (CH3-E1) is considered either quite small (less than 0.5 m offset) or the same as CH2-E1, as discussed above. The lack of datable material from the capping units prevents dating the CH3-E1 event horizon more accurately and distinguishing between it and CH2-E1. To resolve, we use the dates from Channel 2 as the upper bound and date it to the mid-7th to mid-8th century CE (Table 2, Fig. 4). The two candidate historical earthquakes for that period are the 632/634 CE and the 659/660 CE events. The 634 earthquake occurred concurrently with the appearance of a comet, causing great consternation yet little damage in the area (Ambraseys, 2009; Guidoboni, 1994). The 660 earthquake caused extensive damage in the Jordan valley and damage in settlements throughout the region. The earthquake most likely ruptured at least part of the JVF and had documented strong aftershocks (Ambraseys, 2009).

The second event seen in channel 3 sediments (CH3-E2) dates to the 6th century CE. During that period of time several earthquakes are known to have caused damage in the region, including the 502 CE and 551 earthquakes. Both earthquakes caused damage mostly on the Lebanese and Palestinian coast, and according to damage reports, the 551 event was the stronger of the two earthquakes (Ambraseys, 2009). Paleoseismic studies point to the offshore Lebanese thrust system as the most likely source of the 551 earthquake (Elias et al., 2007), and it is possible that such a large event (estimated magnitude 7.5, Ben-Menahem, 1979) may have triggered movement on the JGF via the Roum fault, where paleoseismic evidence places the most recent event sometime after 84–239 CE (Nemer and Meghraoui, 2006). The 4th–8th century period is relatively well documented historically, so a historically “missing” event with >1 m of slip is not likely (Zohar et al., 2016), so we interpret CH3-E2 as either associated with the 502 CE or 551 earthquakes.

The two youngest events in channel 4 sediments, CH4-E1 and E2, are associated with 2.7+2/−2.5 m of offset. This amount may represent two events of comparable magnitude, similar to the channel 3 and channel 2 offsets, or it may represent one larger and one smaller event, similar to the 1202 and 1759 CE offsets. There may even be an additional event between channel 4 and channel 3, which contributes slip to the overall measured offset, although evidence for such an event is lacking due to a gap in the sediment record (Fig. 8). Candidate earthquakes that caused the offset include historical events in the 303, 347 and 363 CE. The 303 earthquake caused damage mostly along the southern Lebanese coast whereas the 347 earthquake caused local damage that is only reported from Beirut, Lebanon. The 363 event was actually 2 earthquakes that caused extensive damage in the region, from Aqaba and Petra, Jordan, south of the Dead-Sea (Klinger et al., 2015) all the way to Paneas, north of the Sea of Galilee, and most likely one of them ruptured the Jordan Valley segment. Archaeological excavations in Sussita, east of the Sea of Galilee, revealed major destruction of an Odeion and Roman basilica, where coins found beneath the collapsed structures are dated to as late as 362 CE (Wechsler and Marco, 2017). It is reasonable to assume that the 363 earthquake rupture reached as far north as the Beteiha valley and that it is part of the measured 2.7+2/−2.5 m offset of the youngest two events in channel 4.

• from Wechsler et al. (2018:221)

In order to quantify the variability of slip per earthquake, we used the measured offset estimates and error margins for the past 6 events (excluding the doubtful CH3-E1) to calculate the coefficient of variation on displacement (CVs), using the same approach we used for the CVt calculations. The slip in events CH4-E1 and CH4-E2 together was assumed to sum up to a total of 2.7 m, but in one sampling run we set one event with 2.2 m of slip and the other with only 0.5 m, while in the second run we set both events to have 1.3 m of slip each. Lastly, we also calculated CVs for the last 4 events only. In all cases, the calculated mean slip-per-event was 1.2–1.3 m, with a CVs of 0.5–0.65, which implies fairly characteristically-sized slip events. Zielke et al. (2015) attribute the low variability of slip to the nature of the dataset they examined (offset geomorphic structures along the fault) and its inability to record offsets that are smaller than some threshold (approx. 1 m).

The higher values of CVs in this study may represent a difference in resolving power, and because we were able to resolve smaller offsets and attribute them to single events our dataset includes more variability. The observation of Zielke et al. (2015) that CVs is smaller than CVt is valid for this study as well. This likely reflects a fundamental characteristic of the JGF, which is seemingly unable to produce large offsets per event such as recorded on neighboring faults such as the Yammouneh fault to the north.

The variability in the amount of slip associated with the different events at the Beteiha site, most of which are about 1.2–1.3 m but which range from 0.5 m to > 2 m, may reflect the different sources of earthquakes on different segments. Based on the historical record combined with the paleoseismic data, the JGF segment appears capable of rupturing in conjunction with its northern and southern neighbors as well as by itself. For example, the 1202 CE earthquake that ruptured primarily the Yammouneh segment in the Bekka Valley (Daëron et al., 2007) also produced slip as far south as the Sea of Galilee, with 2.2 m at the Beteiha site. The 1759 CE earthquake of October (M6.6) ruptured along the Jordan Gorge and the Rachaya faults, and its epicenter was probably located south of the Hula basin (Ambraseys et al., 1994; Gomez et al., 2003; Gomez et al., 2001; Nemer et al., 2008; Sbeinati et al., 2005), and produced 0.5 m of slip at both the Beteiha site (Marco et al., 2005) and the Ateret fortress (Ellenblum et al., 1998).

We interpret the ~1.3 m of offset of Channel 2 and upper Channel 3 to have occurred in the mid-8th century CE. To the south, the 749 CE earthquake is interpreted to have ruptured the Jordan Valley segment based on historical reports of damage and paleoseismology (Ferry et al., 2011). Our mid-8th century event could be synchronous with the large (M7.5) 749 Jordan Valley earthquake, which extends the northern limit for this event, or it could have ruptured as a triggered event and been amalgamated into the 749 CE event, or it may represent rupture in the 746 CE or 757 CE events, both of which are interpreted to have ruptured faults north of the Beteiha Valley (Ambraseys, 2009).

• from Wechsler et al. (2018:221-222)

In order to test the feasibility of multi-segment ruptures on the JGF, a simplified model of the DST from the Jordan Valley segment to the Lebanese restraining bend (not including the Mt. Lebanon thrust) was constructed for Coulomb stress modeling. Fault orientations, dips and locking depths were taken from the literature or assumed based on surface geometry and topography (Table S1). The kinematics of the fault movement were estimated based on existing literature and fault geometry (see Table S1). We did not incorporate a creeping section along the northern JVF, to keep the model simple. In each model run, a rupture was initiated on one fault, based on displacement estimates for the last earthquake on that fault. The resulting Coulomb Stress change on the neighboring faults was calculated using Coulomb 3.3 code (Toda et al., 2011). For comparison, the typical stress drop in a surface-rupturing earthquake can range between 0.5 and 50 MPa (Allmann and Shearer, 2009), and an increase of as little as 0.1 MPa (1 Bar) in the Coulomb stress is considered in the literature as effective for triggering earthquakes (King et al., 1994).

It was found that the stress increases by at least 0.1–0.2 MPa on the JGF in all cases of rupture on nearby faults, and most strongly for large events originating on the Yammouneh or Jordan Valley faults (Fig. 9). Additionally, the model demonstrated that a large earthquake on the Yammouneh fault would un-stress and therefore delay the other Lebanese faults. The coulomb stress transfer model, while being very simple, still shows how the JGF location makes it primed for offset in the event of rupture of a nearby fault, which explains the relatively large number of moderate events observed from the trench data.

Ellenblum et al. (1998, 2015) excavated the archaeological site of Ateret (Vadum Iacub), which was constructed across the JGF about 12 km north of our paleoseismic site. They found an early Hellenistic (333–143 BCE) wall that was offset 6 m, and another late Hellenistic (143–63 BCE) wall that was estimated to be offset only 3.5 m. At our site, the cumulative offset for the last 1700 years is at least 7 m, and based on the overall slip rate it should be about 8–9 m for Hellenistic period features. It would be expected that the two sites would have similar amounts of displacement for similar amounts of time, yet the Ateret site appears to accommodate a lesser amount of cumulative slip, even ignoring the extremely low estimated offset of the late Hellenistic wall. This is also true for the 1202 CE rupture offset, estimated to be 1.5 m at Ateret, versus 2.2 m at the Beteiha site.

This difference of ~30% less slip at the Ateret site may be explained by its location astride a branch of the JGF which is near a trifurcation point and some of the offset is taken by other branches (Fig. 9). A normal fault branches to the north-west immediately south of the archaeological site, causing a left bend in the Jordan River and offsetting Pleistocene strata. Additionally, a fault strand parallel to the Ateret branch and marked as the main continuation of the JGF northward is mapped just north of the junction, east of the site (Fig. 10). Those faults and perhaps other unknown fault(s) that are immediately adjacent to the archaeological site may accommodate some of the offset. This is supported by GPS campaign data across the JGF south of the Ateret site and the branching point, which indicate a locked fault with a slip rate of 4.1 ± 0.8 mm/yr (Hamiel et al., 2016), in agreement with the paleoseismic slip estimates from the Beteiha site (this study).

We resolved slip per event for the past six surface ruptures at the Beteiha paleoseismic site, with slip estimates ranging from 0.5 to 2.2 m, a factor of more than four (Table 3) and a corresponding CVs value of 0.5–0.6. Although at first glance, this range of offsets argues against the characteristic earthquake model (Schwartz and Coppersmith, 1984), when local and regional structure is considered, the model may still apply in some cases. For instance, the largest displacement is associated with the 1202 earthquake, which is known to have ruptured the Bekka Valley with several meters of displacement (Daëron et al., 2007).

The Beteiha site may be near the southern terminus of the 1202 rupture, although it is not known whether the rupture terminated in the Sea of Galilee or propagated as far south as the Jordan Valley. As mentioned earlier, the crusader fortress of Belvoir, located only 13 km south of the Sea of Galilee, was not damaged in the 1202 earthquake indicating that rupture did likely stop at the Galilee releasing step. In contrast, damage from the October 1759 earthquake rupture is centered on the Jordan Gorge fault (Ambraseys and Barazangi, 1989; Sieberg, 1932), and 50 cm of offset at Ateret is attributed to this event, suggesting that this event ruptured only the short, 40 km-long segment of the DST that includes the Jordan Gorge and the east side of the Hula Valley until the NE-bend in the fault, or perhaps even only the 20 km long JGF.

Potentially, this earthquake may have also ruptured the Rachaya fault in Lebanon (Daëron et al., 2005), although conversely, it is possible that the Rachaya ruptured together with the Serghaya fault in the larger November 1759 event (Gomez et al., 2003). Thus, one plausible model is that small, half-meter events may represent rupture of only the 20 km-long JGF, whereas the large 1.2–2.2 offsets represent displacement in larger earthquakes that ruptured either to the north (1202 CE) or south (749, 363 CE).

In transition zones between simple and complex sections of plate boundary faults, repeating earthquakes with similar displacement and rupture extent (i.e., characteristic earthquakes) may be an over-simplified model that should not really be expected to conform to reality. It should be noted that resolving 0.5 m of slip is generally beyond the accuracy limits of this study, so while there may be other 1759-style rupture events, they have not been detected. It is possible that one or more of the channel 4 events are of the smaller 1759-offset type.