We use an earthquake catalogue from 1 January 1983 until 31 August 2017 within 28–34 ∘N and 33–37∘ E, recorded by ∼140 stations whose distribution has changed in time and space. Most of the data are from the Israel Seismic Network (ISN), the Comprehensive Nuclear-Test-Ban Treaty (CTBT) and the Cooperating National Facility (CNF). Some additional data were incorporated from other regional networks: GE; the GEOFON global network of Deutsches GeoForschungsZentrum, Potsdam (GFZ); the Jordanian Seismic Observatory (JSO); and the seismic network of Cyprus (CQ). These earthquakes, which have been monitored by the Seismological Division of the Geophysical Institute of Israel, comprise a catalogue of ∼17600 earthquakes. They were relocated (Fig. 2) to generate a new catalogue with more precise locations of hypocentres (Wetzler and Kurzon 2016). As part of the relocation process, ∼900 earthquakes were excluded for various reasons, e.g. events that were recorded by less than four stations and large location errors (including the Md=5.8 1993 event in the Gulf of Elat, which does not affect our marking of faults since it was nucleated outside our high-resolution geological data). Before 1983 the locations are less reliable. Hence, the relocated catalogue consists of ∼16700 events of 0.1≤M≤7.2 (Fig. 2). Earthquakes with unknown magnitudes received a default value of M=0.1. The magnitude and the location of the Mw=7.2 1995 Nuweiba earthquake were fixed according to Hofstetter et al. (2003).

In order to assess the applicability of the following seismic processing and analysis, we define the network coverage area as the zone in which the hypocentres are relatively well constrained. This is examined and determined here as the polygon that covers all seismic stations that recorded at least 350 arrivals and consists of the smallest number of polygon sides that link between the stations (Fig. A1 in Appendix A).

In order to quantitatively characterise the regional seismicity and associate the earthquakes with mapped faults, we examine two parameters: (a) earthquake kernel density and (b) seismic moment (M0) kernel density. Both parameters are obtained through the following spatial data processing. A regional scan is carried out in a 0.5 km interval 2-D grid, in the horizontal coordinates. For each grid point, both parameters are calculated utilising all recorded events within a 6 km radius. The parameters are calculated based on the kernel density estimation as an approach to obtain the spatial distribution through a probability density function, using the distance to weight each event from a reference point (each grid point, the common centre of its adjacent events). This circular-shape-based approach prevents any directional bias.

The 6 km limitation, the Gaussian function and its standard deviation of 2 (for the kernel estimation) were tuned and chosen to (a) capture different seismic patches along active faults, (b) be significantly larger than the location horizontal median error (∼1.2 km; Wetzler and Kurzon, 2016), (c) assign a higher weight to events closer to the evaluated grid point and (d) include as many events as possible for achieving statistical significance at each of the grid points.

The earthquake kernel density parameter, ρNk, is calculated by counting all the weighted events within a 6 km radius from each grid point, dividing their sum by the sampler area (πr2) and normalising by the duration of the earthquake catalogue: 1ρNk=∑n=1Ne-d(n)22σ2Tπr2, where N is the total number of events within the radius r, d(n) is the distance between an event n and the circle centre, σ is the standard deviation of the Gaussian function and T is the duration of the earthquake catalogue. Units are events per squared kilometre per year (events km-2 yr-1).

The M0 kernel density parameter, ρM0k, is obtained by first calculating the seismic moment released by each event separately, using the empirical relation between M0 and ML, as obtained by Shapira and Hofstetter (1993) after converting units from dyne × cm to N×m: 2log⁡M0=10+1.3ML. Secondly, each amount of energy is weighted according to the distance of the corresponding event from the circle centre (like the calculation of the earthquake kernel density). Then, we sum the weighted M0 released from all the events within a 6 km radius, divide the sum by the circle area (πr2) and normalise by the duration of the catalogue: 3ρM0k=∑n=1NM0(n)e-d(n)22σ2Tπr2, where N is the total number of events within the radius r, M0(n) is the seismic moment released from an event n according to Eq. (2), d(n) is the distance between an event n and the circle centre, σ is the standard deviation of the Gaussian function and T is the duration of the earthquake catalogue; units are joules per squared kilometre per year (J km-2 yr-1).

This category (solid black in the map) includes potential sources for large to major earthquakes in the region. According to paleo-seismic and/or geodetic investigations (Table 1; Fig. A2), these faults are associated with Holocene average sinistral slip rates of 1–5 mm yr-1. Equally important, all the faults in this category are relatively long with a preferable slip orientation according to the present stress field (Jaeger et al., 2007; Eyal and Reches, 1983). Our database (Fig. 5) includes fault segments from this subgroup that are located up to 150 km away from Israel. As noted in Sect. 4, the only recorded large earthquake, the 7.2 Mw Nuweiba event, occurred on the Aragonese fault and was associated with mean slip of 1.4–3 m (Baer et al., 1999).

South of Lebanon, geodetic measurements show ∼4–5 mm yr-1 sinistral slip rate (Masson et al., 2015; Hamiel et al., 2016, 2018a, b). Faulting in Lebanon is partitioned to a few branches (Fig. 5) and the specific rates are less constrained. While the Yammuneh and the Serghaya faults can undoubtedly be considered as independent sources for significant earthquakes, the status of the shorter Rachaiya and Roum fault branches is less clear. Nevertheless, according to the present state of information (for example, Nemer and Meghraoui, 2006), we cannot rule them out and they remain part of this group.

Previous analyses of maximum earthquake magnitude based on historical earthquakes or on background seismicity predicted magnitudes of ≤7.8Mw for the largest segments (e.g. Stevens and Avouac, 2017; Klinger et al., 2015; Hamiel et al., 2018a).

This subgroup (pale blue lines in the map) consists of fault zones with lengths of several to dozens of kilometres that are associated with the DST. Based on several previous works (Table 2), we estimated the slip rates along these fault zones as 0.5–1 mm yr-1. All the fault segments are located inside (or partly inside) the overlap zone defined by the two seismological analyses (Fig. 6).

The subgroup includes the Hasbaya fault in Lebanon, the fault zone in the western and eastern margins of the Dead Sea, the marginal faults of the Hula basin, the Carmel-Tirza fault zone (CTF) and the Elat fault (Fig. 5). The partitioning of the slip rate across parallel segments in any given fault zone is usually below the geodetic measurement (or the information) resolution. Therefore, the segments presented in Fig. 5 are representative but not necessarily the most active within a given system.

Due to the lack of reliable historical and paleo-seismological evidence, the evaluation of maximum possible magnitude on these faults is less certain and requires several assumptions. First, we consider here a local rupture on segments from a given system and disregard a rupture of the entire system as part of an extremely large earthquake on the main strike-slip faults (as evaluated separately in Sect. 5.1). In addition, we assume that the longest possible subsurface rupture length is similar to the length of the segment's surface trace. For example, the Carmel fault, the northern fault in the CTF, is up to 40 km in length (on- and offshore). According to some published scaling relationships, rupturing along its entire length can be associated with up to ∼7Mw earthquakes (Wells and Coppersmith, 1994; Stirling et al., 2013). However, here we assume again that such magnitudes must be interconnected with an earthquake along a much larger DST segment (Agnon, 2014) and not confined to a local segment. We therefore assume a maximum rupture length of ∼10–20 km along faults from this subgroup and correspondingly to maximum magnitudes of Mw<6.5 (Wells and Coppersmith, 1994). We note that the data on the Elat fault is based only on evidence from its northern edge (e.g. a catastrophic event at 2.3 ka inferred by Shaked et al., 2004), while the rates at its offshore parts are less constrained. Further work on its subsurface section and the connection to the main sinistral displacement is required for better evaluation of its seismic potential.

We additionally note that large earthquakes along the Cyprian Arc (Fig. 1) can also generate tsunamis that might affect the coastline of Israel (Salamon et al., 2007). This source is not analysed and mapped here but should be taken into account in regional seismo-tectonic models.

The hazard of surface rupture is defined as the likelihood of an earthquake that will rupture the surface within a certain time window. This likelihood is based on knowledge about the past and present fault kinematics and dynamics. The determination of the relevant time reference for young faulting is usually dictated by different constrains and applications. In the United States, faults are commonly considered to be active for planning constructions if they have ruptured the surface at least once in the past 10 kyr. However, regional conditions, such as sedimentary cover or available age dating of pertinent geological units, can affect this determination. For example, faults that are defined as “Active” in the “Design Provisions for Earthquake Resistance of Structures” in Israel are those that ruptured the surface in the past 13 kyr (Heimann, 2002). This is the age of the top of the lake formation that covers significant parts of the Dead Sea valleys.

The time reference for special constructions such as dams and nuclear power plants is usually much longer, because the possible damage to the construction has severe regional implications. According to the International Atomic Energy Agency (IAEA) Safety Fundamentals (IAEA, 2010), capable faults are those with evidence of displacement over thousands or millions of years, depending on how tectonically active is the area. Here, the Quaternary period is selected as the time reference for sensitive facilities due to two main reasons: (a) we assume that faults that were active during the present regional stress regime (Zoback, 1992) are more likely to activate in the near future. The regional stress state within the Quaternary period represents the current stress field well (Eyal and Reches, 1983; Hofstetter et al., 2007; Garfunkel, 2011; Palano et al., 2013). We note that “regional stress field” (Zoback, 1992) as a criterion for active faulting is closely related to the “tectonic regime” suggested by Galadini et al. (2012). (b) Quaternary geological units are mostly well defined in the region.

The primary and secondary criteria for sorting the faults are listed in a descending order of categorisation, meaning that faults are initially examined according to the first criterion, and, only if they do not match it, they are examined according to the second criterion, and so on. Where geological evidence is absent, we utilise a seismological criterion (Fig. 6), under the assumption that faults associated with seismically active subzones are more likely to have ruptured the surface in the Quaternary compared to others.

Finally, because of the limitation of our database, mapped capable faults (Fig. 7) are limited to Israel and its vicinity.

Main strike-slip faults of the DST are identified here as the main sources for large regional earthquakes (Fig. 7).

Faults that have no field relationship with Quaternary formations consequently show no direct evidence for Quaternary faulting. We therefore designed the next criteria under the rationale that they expand the database with faults that reasonably have been active since the Quaternary, based on the following criteria.

First-order branches and the marginal faults of the DST. a.

First-order branches of faults that are mapped following the primary criteria. A fault branch is defined here as splitting at an acute angle from another fault. The throw direction of the fault and its branches are also taken into account.

Regions with intermediate seismicity rates present a challenge for hazard evaluation; whilst the hazard might be perceptible, the seismic data and the geological evidence for recent surface rupture are sparse compared to very active zones. Considering that the earthquake phenomenon is a stochastic process and its predictability is limited, we develop a methodology for mapping and characterising hazardous faults by taking advantage of incorporating interdisciplinary information with statistical seismological analyses.

Two regional fault maps are presented: one is relevant for regional ground shaking models (Fig. 5) and the other for surface rupture near facilities that are particularly vulnerable to this hazard (Fig. 7). In addition to the approach of classifying faults by the recency of faulting or by their recurrence intervals (Machette, 2000, and references therein), we utilise other criteria such as seismological patterns (Sect. 4) and tectonic configuration (Sect. 6.3). In particular, we use the distribution of the earthquake kernel density and the seismic moment kernel density to test the relevancy of faults for different hazards. Figure A3 reveals that most of the capable faults, which are mapped based on the geological criteria, could have entered the map also by the seismological criterion (ignoring its 6 km fault length limitation). The match between the geologically categorised faults and the area defined by the seismological analysis reinforces the methodological concept of utilising the two seismological distributions that are, to a large extent, independent of one another. Moreover, faults that are defined here as “main seismic sources” according to specific tectonic conditions (i.e. slip rate, geometry, structure) are well correlated with the zone defined by our seismological analysis (Fig. 6). This emphasises the significance of this analysis, especially when slip rates are slow or under debate (as in Sect. 5.2).

The internal hierarchic categorisation of faults in both maps (Figs. 5 and 7) enables separating different fault groups and can later be implemented if a specific hazard is considered or if risk evaluation is applied. However, we note that although faults are marked by hierarchical criteria, the different categories in many cases complement each other rather than show the hierarchy of the activity level. The grid-based distributions of the obtained seismicity parameters are utilised here together with fault geometry parameters (length and orientation) for defining capable faults. The advantage of this integration is expressed where the seismological criterion (Sect. 6. 3) defines capable faults in zones where young formations are scarce (Fig. 7). Just as important, our database of gridded seismicity, with possible adjustments, can be implemented as an independent source for hazard evaluations, and as a complement to the regional databases of mapped faults in zones of subsurface faults.

Although our methodology is demonstrated for Israel and its vicinity, the approach is universal and is particularly useful in domains of intermediate seismicity rates or limited field evidence. The criteria, when implemented in other regions, should be adjusted according to the regional and local seismo-tectonic settings. For example, our seismicity-based analysis is not considering the orientation and the inclination of the fault surface when epicentre locations and fault traces are correlated together, because most of the faults in the study area are characterised by steep dips. This cannot be neglected in low-angle fault zones or the convergence regime. Finally, our approach of hierarchic tectonic criteria for categorising faults can be applied in principle also when local siting of an infrastructure is considered. However, faults with extremely long recurrence intervals, located along zones that are not covered by young formations, might be difficult to detect, even when seismo-tectonic criteria are considered. Moreover, faults that constitute a mechanical potential for slip, such as conjugate fault sets (Eyal and Reches, 1983) or old faults that can be reactivated by stress triggering (Stein et al., 1997), are not defined as capable in our regional analysis, unless further geological or seismological evidence for Quaternary activity exists. Therefore, local siting, in particular of sensitive infrastructure, might require stricter criteria both for surface rupture and ground shaking, depending on the specific requirements.

The DST accommodates most of the seismic activity but also contains zones of very sparse seismicity (Fig. 6). The seismicity distribution maps (Figs. 3 and 4) exhibit enhanced seismicity in the pull-apart basins and reduced activity in the long straight segments. The heterogeneous distribution can be explained by the tendency of stress amplification and failure to concentrate locally within zones of geometric irregularity, such as releasing bends (e.g. Segall and Pollard, 1980; Reches, 1987), whereas the long segments can accommodate higher stresses that are released in single earthquakes of more seismic moment release (Sagy and Lyakhovsky, 2019). At the northern section of the Jordan Valley long segment, section I is the least active part of the DST during the last ∼35 years. Shallow crust creep along the northern part of this segment at a rate of approximately half the total plate motion (Hamiel et al., 2016), and potential partitioning of the DST activity to the CTF (Sadeh et al., 2012; Hamiel et al., 2018b) might reduce the seismicity rate in section I (Fig. 6). Sections II and III, at the middle and the northern sections of the Arava segment, are also associated with sparse seismicity but to a lesser extent. With no indication for creep, the reduction of seismicity might be attributed to local locking of the main fault. Structural and lithological contrasts in fault junctions (e.g. the SNB and ∼ NNE striking faults) might also affect the increasing or decreasing of local seismicity along the segments.

Figures 3 and 4 indicate a ∼ SE–NW-trending seismological lineament with intensified seismicity in its southeast (IV in Fig. 6, referred to here as the east Sinai zone). This lineament seems to branch off the DST in a zone of a structural boundary, between the deep tectonic basins of the Gulf of Elat (Ben-Avraham, 1985) and the Arava valley, a “structural and topographic saddle with hardly any `rift-valley' in its centre” (Garfunkel, 1981). Since the seismic activity implies that it may run further northwest, we refer to it as the Elat–Bardawil lineament (EBL). Its orientation, sub-parallel to the CTF, the Suez rift and the Red Sea spreading centre, might indicate a similar extensional feature (see Fig. A4). This possibility is supported by geodetic analysis (Palano et al., 2013), a focal mechanism solution within this zone (Abdelazim et al., 2016), and by the orientation of nearby Quaternary faults (Fig. 6) and other fault traces in Sinai, outside our high-resolution data (e.g. Eyal et al., 1980). However, currently there are no available high-resolution maps to confirm the existence of faults associated with the seismicity in the east Sinai zone. We interpret the seismicity within the EBL as related to the reactivation of subsurface faults that were either formed during the post-Eocene Red Sea rifting or even older faults. Further research is required for better characterisation of this activity and its relationship to the regional tectonics.

Finally, relatively long E–W-trending faults (SNB) cross the south of Israel and Sinai, and some of them are marked as Quaternary faults (Figs. 7 and A3). However, there are no geologic or geodetic indications for any activity along them since the early Pleistocene, and the associated seismic activity mostly concentrates in their junctions with the DST. We therefore assume that these dextral oblique-slip faults are inactive in the present regional stress field, and their reactivation may generally decrease with increasing distance from the DST.