Ground Motions in the Zevulun Valley(Haifa Bay) Deep Structure

Ground motions in the Zevulun Valley (Haifa Bay) Deep Structure Effects

Thesis submitted in partial fulfillment of the requirements for the Master of Sciences degree

By Alina Goldberg 317518090

Under the supervision of:

Prof. Michael Tsesarsky, Ben Gurion University

Prof. Zohar Gvirtzman, Geological Survey of Israel

January 2020

Ground motions in the Zevulun Valley (Haifa Bay) Deep Structure Effects

Thesis submitted in partial fulfillment of the requirements for the Master of Sciences degree

By Alina Goldberg 317518090

Alina Goldberg ______Date: 14.01.2020

Under the supervision of:

Prof. Michael Tsesarsky ______Date: 14.01.2020

Prof. Zohar Gvirtzman ______Date: 14.01.2020

Signature of chairperson of the committee for graduate studies ______Date: ______

January 2020

Table of contents

1 Introduction ...... 1

2 Scientific Background ...... 4 2.1 Ground motion amplification ...... 4 2.2 1D, 2D and 3D methods ...... 6 2.3 Geological setting and seismicity of the studied region ...... 8 2.3.1 Seismicity ...... 8 2.3.2 Zevulun Valley geological background ...... 9 2.3.3 Potential seismic reflectors ...... 12

3 Methods ...... 12 3.1 SW4 Wave propagation software ...... 13 3.1.1 Governing equations and time functions ...... 13 3.1.2 Seismic sources ...... 15 3.1.3 Simulation setup ...... 15 3.2 Output options ...... 18

4 Model Setup ...... 19 4.1 Geological data ...... 19 4.2 Raster file construction ...... 20 4.2.1 Zevulun Valley model structure ...... 21 4.2.2 Geological model inaccuracies ...... 22 4.3 Simulation setup ...... 23 4.3.1 Reference model ...... 23 4.3.2 Simulation parameters ...... 25 4.3.3 Summary of Zevulun Valley simulations ...... 28 4.4 Output plan ...... 28

5 Results ...... 30 5.1 PGV results ...... 30 5.2 Seismograms, arrival times and spectral analysis ...... 34 5.2.1 Northern Source (JGF) – observations ...... 36

5.2.2 Southern Source (KNR) – observations ...... 38 5.3 Result interpretation ...... 48 5.3.1 PGV analysis ...... 48 5.3.2 Energy and duration of motion ...... 49 5.3.3 Spectral analysis ...... 53

6 Discussion ...... 56 6.1 Amplification maps ...... 56 6.2 Amplification in the Qishon Graben ...... 64 6.3 Simulated frequency content ...... 66 6.4 Source magnitude ...... 67 6.5 Model Limitations ...... 69

7 Conclusions ...... 70

8 References ...... 72

List of figures

Figure 1- A map of areas suspected of ground motion amplification during an earthquake in the northern region of Israel. 1 Figure 2 - Research zone map of the main tectonic boundaries and ZV structures. 3 Figure 3 - The Mw = 4.3 Landers 1992 aftershock recordings (after Boore 2004). 6 Figure 4 - Top Judea elevation map and ZV geological sections. 11 Figure 5 - Visualization of the Tiff formatted curvilinear surfaces comprising the 3D model. 21 Figure 6 - Visualization of regional GITT05 model. 24 Figure 7 - The distributed slip model used to simulate a strike-slip fault in the ZV model. 26 Figure 8 - Research area, source and synthetic stations locations. 30 Figure 9 - PGV and AR results for the JGF simulations. 32 Figure 10 - PGV and AR results for the KNR simulations. 33 Figure 11 - Reflectors at depth for the six representative stations of the ZV. 35 Figure 12 - Results obtained from the PBZN station, DSM-JGF model. 43 Figure 13 - Results obtained from the NVHB station, DSM-JGF model. 43 Figure 14 - Results obtained from the KMKB station, DSM-JGF model. 44 Figure 15 - Results obtained from the AFK station, DSM-JGF model. 44 Figure 16 - Results obtained from the EHM station, DSM-JGF model. 45 Figure 17 - Results obtained from the PBZN station, DSM-KNR model. 45 Figure 18 - Results obtained from the NVHB station, DSM-KNR model. 46 Figure 19 - Results obtained from the KMKB station, DSM-KNR model. 46 Figure 20 - Results obtained from the AFK station, DSM-KNR model. 47 Figure 21 - Results obtained from the EHM station, DSM-KNR model. 47 Figure 22 - Full PGV maps for (a) DSM-JGF and (b) DSM-KNR. 49 Figure 23 - KNR-DSM Arias Intensity results. 51 Figure 24 - Velocity magnitude propagation with time in the KNR-DSM model. 52 Figure 25 - JGF-DSM Arias Intensity results. 53 Figure 26 – Amplified frequencies correlating to a reflector at the ZV stations. 56 Figure 27 - First resonance frequency peak range and amplificatin ratios from Zaslavsky et al. (2008) compared to the present work’s DSM-KNR model results. 58 Figure 28 - Amplification ratios map combined with polygons of the Krayot Cities. 63 Figure 29 - Component ratio analysis for the PBZN station for KNR-DSM model 65 Figure 30 - Spectral amplification at the PBZN, NVHB and KMKB stations, after Shani-Kadmiel et al. (2018). 67

List of Tables

Table 1. Material properties of the simulation based on GITT05 general model. 23 Table 2. Material properties of the main units in the ZV. 26 Table 3. Grid size structure detailed by mesh refinement levels. 27 Table 4. PGV values and amplification values for each of the four simulations. 31 Table 5. PGV values derived from synthetic seismograms in m/sec. 34 Table 6. Comparison of spectral analysis of Zevulun Valley conducted by 4 different studies. 61

Abstract

Sedimentary basins are a common subsurface geological structure in the northern region of Israel. Characteristically, sedimentary structures are filled with relatively young sediments with low seismic velocities, which are known to amplify seismic waves during an earthquake.

The coastal plain of Haifa Bay (Zevulun Valley) is a densely populated urban zone, with nearly 1 million residents and well-developed industrial facilities such as a marine port, oil refineries, and other chemical plants containing hazardous materials. The subsurface of the Valley consists of a complex geological sedimentary structure of two grabens and a horst. Thus, the area is widely researched in the goals of determining the seismic hazard it presents, and particularly the ground motion amplification that can be expected atop it’s complex sedimentary structure in the case of a strong motion earthquake.

In Israel, due to limited coverage of the seismic network and low background seismicity, the empirical research is not sufficient since it requires a large amount of data from previously recorded events. Hence, numerical modeling is essential to study the critical areas that face increased seismic hazard in the future.

In this work, 3D finite difference numerical modeling was performed using the SW4 software to simulate the seismic waves propagation in the complex subsurface structure of the Zevulun Valley during a strong motion earthquake (Mw = 6) generated from two source locations along the Dead Sea Transform. The basin to no-basin amplification maps show maximum amplification ratios of 12.7 for the point source simulations, 11 for the southern source and 10 for the northern source DSM simulations. This leads to the understanding that amplification ratios are similar for different source locations and varied ground velocities. The highest amplification ratios were detected above the

deepest parts of the Graben. Hilazon Graben displayed amplification ratios of up to 4 and no notable amplification was detected atop the central horst.

The spectral analysis, limited to f  0.75 Hz, performed using the H/H and H/V methods, confirmed two suspected geological boundaries as strong seismic reflectors; At both grabens, low frequencies (f < 0.3 Hz) are amplified due to the presence of the top Judea reflector, with spectral amplification ratios of 9 at the station adjacent to the deepest parts of the Qishon Graben, 1.5 at the station located nearby the margins of the graben and ratios of 3.5 at the Hilazon Graben. At the Qishon Graben an additional frequency peak of f = 0.45 Hz is associated with the top Mavki’im/Patish Formations with spectral amplification ratios of around 3. At the central horst, no spectral amplification was detected due to the shallow depth of the top Judea reflector. H/V analysis amplification results didn’t fully coincide with the PGV map ratios and the results of the H/H method, suggesting in most cases much higher and unsupported amplification values. An overall amplification ratio map of the entire ZV area is presented in this work, challenging the lower amplification values that were introduced in a previous work based on ambient noise measurements constraining 1D subsurface models.

1 Introduction

Sedimentary basins are a common subsurface geological structure in the northern region of Israel, including a line of rhomb-shaped-grabens along the Dead Sea Transform (DST): The Dead sea, Kinneret and the Hula Basins. The Harod Valley, the Jezreel Valley and the Zevulun Valley (ZV), lie along the Carmel Fault Zone (CFZ). Sedimentary basins are commonly filled with relatively young and poorly consolidated sediments with low seismic velocities. These sediments are known to amplify incoming seismic waves during an earthquake (Boore 2004). Sedimentary basins suspected of ground motion amplification during a massive earthquake in northern Israel are colored in pink in Figure 1 (Gvirtzman & Zaslavsky 2009).

Figure 1- A map of areas suspected of ground motion amplification during an earthquake in the northern region of Israel. Green color represents a rock site, white color represents an ordinary soil site, black color represents soil sites suspected to be amplified due to a hard bedrock beneath, pink is for sedimentary basins with assumed extreme ground motion amplification (Gvirtzman & Zaslavsky 2009).

The coastal plain of Haifa Bay (ZV) is a densely populated urban zone, the largest in northern Israel, with critical civil infrastructures and industrial facilities such as a marine port, oil refineries, and other chemical plants containing hazardous materials. The Bay area includes the city of Haifa in the south, the city of Akko in the north and several large towns (Krayot) in between. The subsurface of the ZV is a complex, faulted sedimentary basin, comprised of two grabens separated by a horst (Figure 2). At the surface, the ZV is covered with unconsolidated Quaternary sediments (Kurkar Formation). The combination of soft sedimentary layers and deep subsurface structure increase the seismic hazard, while the varied land uses contribute to the seismic risk of the ZV.

In the past two decades, the Israeli Geophysical Institute (GII) has led an elaborate field campaign (Zaslavsky 2006 and Zaslavsky 2007), using ambient noise measurements to determine amplification values in the ZV. This research was based on Nakamura’s 1989 hypothesis that site response could be estimated from the spectral ratio of the horizontal versus vertical components (HVSR) of microtremors observed at the same site. Zaslavsky et al. (2008) presented distribution maps of the first and second resonance frequencies and their associated amplitudes for the Qishon Graben (QG, Figure 2), summarizing the work they have done for 10 years. An independent computational study was conducted by Gvirtzman & Louie (2010), who performed 2D numerical analysis of wave propagation for several detailed geological sections of the ZV. By the means of the basin to no-basin ratio method they showed that the 2D analysis yields higher amplification values than the instrumental 1D analysis, especially for the deep basin structure of the QG. They also provided new information on the dominant frequencies and their related amplification ratios for the Afek Horst (AH) and the Hilazon Graben (HG, Figure 2). In a recent study, Shani-Kadmiel et al. (2018) used a portable network of 6 seismic stations to record ground motions in the ZV for 16 months during the years 2014-2015, collecting and analyzing data resulting from 15 M > 3.5 earthquakes. By comparing their results to the previous works mentioned above, they

3 showed that amplification values in the deeper parts of the QG are even higher than those obtained by the 2D study.

The primary goal of this research is to estimate the amplification of ground motions atop the ZV during a strong motion earthquake using high resolution 3D numerical model of the ZV. This study is expected to complement the existing instrumental and 2D numerical modeling by addressing 3D effects and covering areas not modeled or instrumented in previous research. Secondary goals include the study of the expected peak ground velocity (PGV) values and the duration of motion during a large seismic event and evaluation of the potential seismic hazard the ZV presents in future earthquakes.

Figure 2 - Research zone map. (a) A map of the main tectonic boundaries and faults affecting the ZV – Dead Sea Transform (DST), Carmel Fault Zone (CFZ) and the Cyprus Arc. (b) ZV structures – Qishon and Hilazon Grabens, Afek Horst, the Carmel and Ahihud Faults.

2 Scientific Background 2.1 Ground motion amplification

There are two main physical reasons to ground motion amplification: The main cause of amplification is the decrease in seismic wave velocity near the surface when a wave propagates from the surrounding rock to an upper softer layer of unconsolidated sediments, soils or weathered rock. The elastic energy of a shear wave is given by 휌Vsu̇ 2, where 휌 is the density, Vs is the shear wave velocity and u̇ is the particle velocity. The conservation of elastic energy requires constant energy flux along the wave’s path (assuming geometrical spreading and intrinsic attenuation are negligible across the interface). For a wave traveling from a hard rock with properties ρ1, vs1 into a soft rock with properties ρ2, vs2:

2 2 (1) ρ1vs1u̇ 1 = ρ2vs2u̇ 2

The amplitude increase-factor A, equal to the impedance ratio, is given by:

2 u̇ 2 ρ1vs1 (2) A = 2 = u̇ 1 ρ2vs2

Since generally the density ratio is much smaller than 2, the amplification ratio can be related to velocity ratio as:

v (3) A ~ s1 vs2

The second reason for ground motion amplification is the interference of waves reflected from the free surface with the waves reflected from the underlying bedrock, creating a standing wave effect amplifying and prolonging the motion. For a simplified case of a vertical wave traveling through a layer with thickness H and homogenic material properties (ρ, Vs), the resonance frequency (or “fundamental frequency”) is given by the 1D equation:

v (4) f = s 4H

The equation outlines the fact that fundamental frequency is determined by the soft layer and not of the hard bedrock beneath.

However, the mentioned reasons only consider the vertical 1D effect, which depends on the layer thickness and material properties. Over the past tens of years, sedimentary basins exhibit 2D and 3D complexities affecting the amplification of motion during strong motion earthquakes; Mexico City, Los Angeles and Kobe are only few examples of mega- cities built atop large sedimentary basins, that suffered structural damage caused by strong ground motions and prolonged duration of motion during past high magnitude earthquakes in their areas (Singh et al. 1988, Holmes & Somers 1996, Aguirre & Irikura 1997). In these events, structures that were built on hard rock suffered relatively less damage than those built on top of soft sediment. An example from the Mw = 4.3 Landers 1992 aftershock earthquake is presented in Figure 3. The Figure shows recordings from two different stations located at the same epicentral distance from the source, one is situated on top of a soil site (soft sediments) and the other on top of a rock site. It is evident that the soil site amplitudes are significantly higher than those of the rock site, mainly in the horizontal components (Boore 2004).

The possible multi-dimensional effects are:

1. Horizontally propagating surface waves can resonate across the basin’s width:

v (5) f = s 2L where L is the basin’s width and Vs is shear wave velocity in the soft layer.

2. Focusing and scattering of waves induced by the geometry of the basin’s floor acting as a lens. A curved basin floor can focus and amplify ground motions by a factor of 3 or more (Shani-Kadmiel et al. 2014).

3. Constructive interference of body waves and surface waves near the basin’s edge, known as ‘basin edge effect’ can be disastrous in seismic events. In the 1995 Kobe, Japan earthquake it was the main cause of damage to many structures due to their location along the basin’s edge (Kawase 1996).

Figure 3 - The Mw = 4.3 Landers 1992 aftershock recordings, showing amplification of motion atop a soil site compared to a rock site (after Boore 2004).

2.2 1D, 2D and 3D methods

One dimensional ground response is a common practice in seismic hazard analysis. This method assumes the response of the sediments is mainly caused by SH waves propagating vertically from the underlying rock upwards, in a laterally homogenous medium (Kramer 1996). The 1D method is simple enough and often able to capture

7 essential aspects of the response, yet it neglects to capture more complex effects caused by multiple refraction and reflection in the basins. Using small earthquake recordings, Tucker et al. (1984) suggested that 1D analysis could predict the average response of sediments near the center of a valley but not near the edges. Silva (1988) showed that 1D analysis in basins may underestimate amplification by a factor of 2 to 4.

2D modeling for site effects research is also commonly used, for example Olsen et al. (1996) used 2D simulation to model large amplifications and long durations recorded in the Salt Lake Basin in Utah, from mine blasts. Their results suggested that deep-basin resonance, reverberations in the near-surface low-velocity layer, attenuation and topographic scattering influence site amplification in the researched basin. Graves et al. (1998) used 2D modeling to study unexpected damage and amplification patterns at basin edges. Their simulations indicated that the shallow basin-edge structure (1 km deep) formed by the Santa Monice fault, creates a large amplification in motions immediately south of the fault scarp, in very good agreement with the mainshock damage patterns, recorded ground motions and locations of elevated site response. Bonilla et al. (2006) used computational modeling to simulate wave propagations in the Grenoble basin. Their numerical study has shown higher amplifications in the case of 2D simulations over 1D simulations, due to a strong impedance ratio between the basin sediments and the surrounding bedrock.

The 2D seismic modeling approach can have an advantage over the 3D modeling; Accurate modeling of soft sediments in many cases requires low velocities and a high frequency limit, increasing the 3D computational costs. However, 3D modeling is a key to the understanding of the complex wave behavior imposed by the source and the 3D underground structure along the propagation path, allowing examining the subsurface geology in all 3 directions, and capturing the spatial variation of the layers, which cannot always be done using 2D modeling.

3D modeling of basin response is a relatively new modeling approach, gaining popularity in the research field in recent years. For example, Wei et al. (2017) studied the 2015

Gorkha (Nepal) earthquake sequence, using 3D basin modeling as a complimentary method to the analytical research, to better understand the role of the basin structure in trapping the waves that led to extensive ground vibration. Their results show that 3D basin model can match observed waveforms reasonably well at -0.3 Hz and lower frequency and that the 3D simulations indicate that the basin structure trapped the wavefield and produced an extensive ground vibration. Moczo et al. (2018) performed 3D as well as 2D and 1D simulations for various typical sedimentary valleys to investigate sensitivity of earthquake ground motion (EGM) to impedance contrast, attenuation, velocity gradient and geometry, comparing the amplification factors of the three types of simulations. Their conclusion was that 1D estimates, as well as 2D estimates at several sites are insufficient due to the key structural parameters that cause the amplification and due to the fact 3D models generate more efficient trapping of surface waves.

2.3 Geological setting and seismicity of the studied region 2.3.1 Seismicity

The seismicity in northern Israel is dominated by two main tectonic boundaries, the Dead Sea Transform (DST) and the Cyprus Arc (CA), (Figure 2). The DST is an active transform boundary between the African and the Arabian plates. The DST has a proven geological, historical and instrumental record of producing significant earthquakes such as the 1927 Mw 6.3 Jericho earthquake and the 1995 Mw 7.2 Nueba earthquake (Hamiel et al. 2009, Marco et al. 1996, Agnon 2014). Analysis of GPS data across the northern section of the Jordan Valley Fault (JVF) and Jordan Gorge Segment (JGS) revealed a slip rate of ~4.1 mm/yr for these sections (Hamiel et al. 2016). The CA is a subduction zone, about 200 km northwest of Israel, that has produced significant earthquakes in the past, such as the 1996 Mw 6.8 event. Another, local source is the Carmel fault, part of the Carmel Fault Zone (CFZ), which bounds the research area from the south. The 1984 Mw

5.3 earthquake which occurred in the area between the Carmel fault and the Jezreel valley has been associated with the Carmel fault (Hofsttetter et al. 1996). Moreover, GPS measurements clearly indicate a left-lateral slip accompanied by fault normal extension along the Carmel Fault with 0.7 mm/yr left-lateral and 0.6 mm/yr extension rates, which means elastic strain is currently being accumulated and might suggest a possible moderate-to-large earthquake along the Carmel fault in the future (Sadeh et al. 2011).

Instrumentally recorded seismic data in Israel is limited, due to the region’s low seismicity and sparse coverage of the seismic network. However strong earthquakes have occurred in the past and are inevitable in the future. Hence, ground motion analysis and the study of basin effect is very important and relevant. This research will estimate, using 3D numerical modeling, the ground motions in the ZV in Haifa bay during a strong (Mw 6) earthquake.

2.3.2 Zevulun Valley geological background

The ZV is situated along the northern shore of Israel, in the Haifa Bay area. The geological structure of the study area is rather complex; In the Valley, a set of normal faults striking E-W forms a structure of two Grabens - Qishon and Hilazon, separated by the AH (Figure 2). The Valley is confined by two main faults; The NW-SE trending Carmel Fault (sometimes named “Yagur Fault”) from south and the Ahihud Fault from north (Figure 2). The Carmel Fault is a part of the Carmel-Gilboa-Faria Fault System. The fault separates the Carmel Mountain from the Zevulun and Jezreel Valleys, forming a steep, 300 to 500 m high slope, with a general vertical throw of over 1000 m (Salamon et al. 2013). The Ahihud fault is part of the Ahihud fault system (AFS), extending from the Bet-Kerem Tectonic Valley in the east to the Mediterranean shelf in the west (Zilberman et al. 2011). A characterization of specific faults inside the Valley is still not available and therefore will not be taken into consideration in this model.

The QG lies in the southern part of the ZV. The width of the graben is 4-6 km. The graben started developing during the Eocene, governed by tectonic activity and sea level fluctuations. During the Miocene, the Graben was filled with calcareous sediments,

10 conglomerates and some evaporites. During the Pliocene transgression, a thick complex of clays and marls was deposited. Late Pliocene – early Pleistocene brought a phase of uplift and faulting, deepening the QG (Kafri & Ecker 1964, Mero 1983).

The main lithostratigraphic units of the sedimentary column in the study area are the Judea, Mt. Scopus, Avedat, Saqiye and Kurkar Groups (Mero 1983, Kafri & Ecker 1964). The W-E and N-S geological sections of the valley together with their shear velocities, based on the work of Gvirtzman et al. (2011) can be seen in Figure 4. The Judea Group of Late Cretaceous (Cenomanian-Turonian) age consists mainly of dolomites and limestones. Its thickness in the research area varies between several meters near the Carmel fault to hundreds of meters at western central part of the graben (Fleischer & Gafsou 2003). The Mt. Scopus Group is composed of Senonian – Paleocene Formations. The formations comprising this Group (Menuha, Ghareb, Mishash and Taqiye) are represented by pelagic chalk and marl with some chert, phosphorite, and oil shale. The Eocenian Avedat Group is mainly represented by alternations of chalk and chalky limestone, with chert nodules. The Saqiye Group lies uncomformably atop the Groups mentioned above. In our study area, The Saqiye Formations can be found only in the QG and are absent from the rest of the valley. The Bet-Guvrin Formation (late Eocene – Early Miocene) mostly consists of chalky marl. Mavki’im Formation is composed of anhydrite formed during the Messinian salinity crisis. The Mavki’im Formation changes laterally with the Patish Formation, reef-limestone largely composed of corals. The Yafo Formation is the youngest of the Group, dated to Pliocene age, represented by a massive, relatively homogeneous layer of clays. The Kurkar Group uncomformably overlays all the mentioned Groups and covers the ZV leaving no surface expression of the grabens. The Kurdane Formation (upper Pliocene – Lower Pleistocene) consists of sandy to calcareous limestone, rich in fossils. Above the Kurdane Formation there is a sequence of interbedded strata of chalky sandstones, silty layers and unconsolidated sands (Zaslavsky 2008, Gvirtzman & Don 2005, Gvirtzman et al. 2011).

Figure 4 - (a) Top Judea elevation map. Red line represents the Carmel fault. Black lines represent the internal faults of the valley. (b) A-A' geological section of the valley. (c) B-B' geological section of the Valley. (b), (c) are based on the work of Gvirtzman et al. (2011).

2.3.3 Seismic reflectors

Gvirtzman et al. (2011) established several lithostratigraphic surfaces in the ZV structure that may act as seismic reflectors owing to their geotechnical characteristics.

1) The top of the Judea Group is the most prominent reflector in the coastal plain; It can be found at varying depths throughout the entire Valley, representing the highest impedance ratio (as known from ambient noise measurements and drilling logs) against the overlying formations. At the QG the Judea Group can be found at depths of more than 1000 m. At the central horst, it is identified as major reflector against the overlying chalks of the Mt. Scopus Gr. and the Kurkar sands. In the HG the Avedat chalks are alternately situated above the Judea units, but represent a relatively low impedance ratio. 2) Top Avedat, although consisting mostly of chalks and marls, has limestone horizons which can cause amplification due to the softer sediments of the Bet-Guvrin Fm. Above at the QG. In this work, the material properties of each formation were chosen homogeneously, therefore the limestone horizons in the Avedat Gr. will not show. 3) Top Mavki’im (alternately, Patish) Gr. is located below the soft clays of the Yafo Fm. and at some places the Kurkar sands. It represents a strong reflector at the QG only, since there is no record of the Saqiye Group outside the QG inside the Valley. 4) The Kurdane Fm. is a local reflector within the Kurkar Gr, detected in ambient noise research. It is complicated to map the Kurdane layers due to their random appearance, therefore this reflector was not included in the 3D model. (Gvirtzman & Don 2005, Zaslavsky 2006, 2007, Gvirtzman & Louie 2010, Gvirtzman et al. 2011).

3 Methods

The study of site effects during strong earthquakes (M > 6) can be conducted using two main approaches: 1. The instrumental approach involves an elaborate documentation of previous events. This requires a history of strong magnitude earthquakes in the research area and a dense seismic network deployed across the research domain.

2. The computational approach requires advanced computational capabilities, which require high-quality data of the source, path and site parameters. Obtaining the necessary information often involves guesswork and assumptions, thus raising the epistemic uncertainty of the results.

The best approach will be the combination of the two: verification of the computational model based on past events and then extrapolation of the results for an event that hasn’t occurred yet, with higher certainty. However, the rare occurrence of high magnitude events in the region and the limited coverage of the Israeli seismic network in this area leads to substantial challenges in using the instrumental method, hence making the numerical modeling an important complementing tool for seismic hazard analysis in the ZV.

3.1 SW4 Wave propagation software

SW4 (Seismic Waves, 4th order) is a computer code for simulating seismic wave propagation using a node-based finite difference approach. The numerical method satisfies the principle of summation by parts, which guaranties energy stability of numerical solution (Petersson & Sjogreen 2017). The model was specifically designed to allow massive parallelization.

SW4 is used by specifying a geological material model, a seismic source and the duration of the simulation. The software then solves the seismic wave equations and outputs the resulting motions as specified by the user. SW4 solves the seismic wave equations in Cartesian coordinates. It is therefore appropriate for local and regional simulations, where the curvature of the earth can be neglected.

3.1.1 Governing equations and time functions

SW4 simulates the motion due to a seismic event by solving the elastic or visco-elastic wave equations in displacement formulation, in the three-dimensional spatial domain. The linear hyperbolic partial differential equations contain second derivatives with

14 respect to space and time. The general elastic wave equation in second order formulation is

(6) ρutt = ∇ ∙ 풯 + F(x, t), x ϵ Ω, 0 < t < T

(7) u(x, 0) = 0, ut(x, 0) = 0, x ϵ Ω

Here, ρ is the density, u(x, t) is the displacement vector, and 풯 = 풯(u) is the stress tensor. The computational domain Ω is a box shaped region where one side (optionally) follows the shape of the topography.

The forcing term 퐹 in the governing equation consists of a sum of point forces and point moment tensor source terms. For a point forcing we have:

퐹푥 (8) 퐹(푥, 푡) = 푔(0, 푡, 푡0)퐹0 (퐹푦) 훿(푥 − 푥0) 퐹푧

Where 푥0 = (푥0, 푦0, 푧0) is the location of the point force in space, 푔(0, 푡, 푡0) is the time function with offset 푡0 and frequency parameter 휔 and 훿(푥 − 푥0) is the Dirac delta. The source time function can be selected from a set of predefined functions, or by spline interpolation of a user defined discrete time-series. For example, the Gaussian time- function:

휔 2 2 (9) 푔(푡, 푡 , 휔) = 푒−휔 (푡−푡0) /2 0 √2휋

The key parameters used by the function are the angular frequency 휔 and time offset

푡0.

The angular frequency is set to match the previously chosen 푓푚푎푥 given by the equation

2휋푓 (10) 휔 = 푚푎푥 2.5

Also matching the fundamental frequency 푓0 by the following equations:

휔 (11) 푓 = , 푓 = 2.5푓 0 2휋 푚푎푥 0

To eliminate numerical artifacts rising from the initial conditions t0 is taken such that:

6 (12) t ≥ 0 ω

3.1.2 Seismic sources

Earthquake source dynamics are a key to understand the physics of earthquake initiation and propagation (Madariaga 2007). Seismic source for modeling has evolved from point source single couples (Nakano 1923) to more complex sources, i.e. moment-tensor sources and various kinematic sources (Maruyama 1963). The point source is the most basic approach to seismic energy propagation through the medium. It is enough for modeling small sources situated sufficiently far from the studied area. Details of the rupture process for the point source are expressed in the moment-tensor source time function M0 (Madariaga 2007). For large and proximal sources, the point source becomes less relevant and the source geometry and propagation of the rupture must be considered. Shani-Kadmiel et al. (2016) has developed a generic, kinematic finite-fault source model that accounts for fault geometry, propagation of the rupture and slip distribution, the Distributed Slip Model (DSM), which was used in this work.

3.1.3 Simulation setup

The basic step in numerical simulation setup is to develop a conceptual model that serves the research goal best. For seismic wave propagation modeling this includes necessary simplifications of the geological structure, seismic source model and location, among other essential inputs. The next step is assigning material properties to the conceptual model. Each point in the model receives values of density (ρ), pressure-wave velocity (Vp), shear-wave velocity (Vs), pressure-wave attenuation factor (Qp) and shear- wave attenuation factor (Qs). The final step is turning the velocity model into an SW4 model. This step involves choosing a series of parameters that are essential to meet the numerical and mathematical conditions dictated by the SW4 software. The three main parameters that should be determined are the grid spacing for the finite difference grid

(Δh), the maximum frequency solved in the simulation (fmax) and the minimum number of grid points per wavelength (P).

Choosing the right parameter values starts with understanding the limits of the computation abilities. The main factors limiting this work are the actual running time TA and the available Random-Access Memory (RAM). The running time is controlled by the number of available processors, can be determined by the modeler and must agree with the schedule set for the research. The available RAM affects the number of grid points (NPTS) and consequently the attainable size of the simulation. To determine the maximum size of the simulation we consider also the approximated memory per grid point (MPGP). For visco-elastic Cartesian grid configuration 475 bytes/GP are required. Given 360 GB of available RAM on a 68 processors cluster:

푡표푡푎푙 푅퐴푀 360∗109 (13) max 푁푃푇푆 = = ≈ 760 ∗ 106 푀푃퐺푃 475

This grid size represents the maximum number of grid points that would generate a sufficiently accurate solution for the simulation. It is possible to perform a slightly larger simulation with the cost of time and precision. The running time is affected directly by the number of grid points. The accuracy of the solution increases with the number of grid points, at the expense of longer running time.

Once the number of grid points for the simulation is determined, it is used to calculate the grid spacing ∆h and the maximum frequency fmax as follows:

푥∙푦∙푧 (14) ∆ℎ3 = 푁푃푇푆

Here x,y,z are the model dimensions.

min 푉 (15) 푓 = 푠 푚푎푥 푃∆ℎ

P, the number of grid points per shortest wavelength, is a normalized measure of how well a solution is resolved on the computational grid. According to the SW4 manual (Petersson & Sjogreen 2017) the accuracy is acceptable with 10 ≥ P ≥ 6.

The challenge in choosing the correct parameter values lies within the relationship between the parameters:

3 3 푥∗푦∗푧∗ 푓푚푎푥∗푃 (16) 푇퐴 ∝ 푁푃푇푆 = 3 min 푉푠

The model dimensions, maximum frequency and number of grid points per shortest wavelength increase the size of the simulation and therefore the running time. The grid spacing, and minimum shear velocity decrease the size of the simulation and the running time.

Simulating site effects due to an earthquake at a specific site involves creating a much bigger simulated area than just the research domain; The simulated area must contain the site, the path to the source, the source and an extra space for absorbing boundary reflection, thus resulting in a simulation consisting of high number of grid points and long running time. The SW4 program allows Cartesian local mesh refinement to be used to make computational mesh finer near the free surface. The mesh refinement is performed in the vertical direction and each Cartesian grid is constructed from user specified refinement levels. In this approach, the grid size in all three spatial directions is doubled across each mesh refinement interface. Since sedimentary basins are near- surface, low velocity structures the mesh refinement can provide a high-resolution solution for the basin itself, increasing the grid spacing with depth bellow the structure, where that kind of accuracy is no longer significant, thus leading to a substantial saving in memory and time.

However, mesh refinement leads to increase in supergrid size; The supergrid is an extra, surrounding layer added around the domain of the simulation to reduce artificial reflections from the far-field boundaries. The thickness of the layer is determined with multiplying the highest grid spacing used in the simulation by 30. The supergrid must be considered and included in the specified model dimensions. This means that too many refinement levels, although translating to smaller NPTS, may lead to a large supergrid size, exceeding computational abilities.

3.2 Output options

The outputs chosen to be presented in this work include PGV maps, amplification basin to no basin maps, synthetic velocity seismograms, spectral velocities, H/H and H/V amplification ratios. The data was obtained using SW4 available main output options:

1. Time history – a 3-component solution can be attained for any X, Y, Z location within the computational domain by a user specified command. The data then is presented in the form of velocity seismograms (in this work), since velocity is more sensitive to lower frequencies of the ground motion than acceleration (Kramer, 1996). Spectral analysis was performed using the Fast Fourier Transform (FFT) to study the frequency content and spectral ratio values, enabling to differentiate the material related amplification from structural related amplification.

2. 2D cross-sectional data – two-dimensional (horizontal or vertical) cross-sectional data arrays can be produced for a specified time step. They can be used for visualizing the solution, making the images for a movie or checking material properties. For the purposes of this study the 2D matrices were used to produce Peak Ground Velocity (PGV) maps of the research area. The magnitude of the ground velocity is calculated using the length of the vector of the joined velocities in all directions:

2 2 2 (17) 푣 = √푣푧 + 푣푦 + 푣푥

The PGV is the absolute maximum value of ground velocity reached within the simulation, using only horizontal PGV, neglecting the z direction. In this work, basin to no basin PGV value ratios were used to calculate the amplification value of the ground surface in the ZV.

4 Model Setup

The aim of this research is to study the effects of complex subsurface geological structure on the ground motion amplification atop the ZV. Correlation between the amplification values and the subsurface reflectors requires the construction of an accurate 3D subsurface model and the preparation of the simulation framework. These are achieved in four main steps:

1. Collecting data on the subsurface geology of the valley from available sources. 2. Organizing the information in a format readable by the SW4 software. 3. Designing the simulations and choosing the parameters discussed in section 3 (‘Methods’). 4. Planning the output options.

4.1 Geological data

Over the past decades many local studies were conducted in the ZV for earthquake hazard assessment; One of the earliest studies was led by Kafri & Ecker (1964), on the general subsurface conditions, local site stratigraphy and composition of the soil in the Valley, based on borehole information and geological cross sections across the study area. Mero 1983 added to the previous research using new collected seismic data and borehole data from the 1960’s and 1970’s. The depth of the Top Judea Group, which is the common and main reflector throughout the valley, was compiled by Zaslavsky 2007 from the structural maps of Mero (1983), the studies of Fleisher & Gafsou (2003) and Bar Yossef et al. (2003). The geology of the shelf and continental margin offshore Haifa is described widely by Almagor & Hall (1980), Eytam & Ben-Avraham (1992), and Almagor (1993).

Gvirtzman et al. (2011) presented six structural maps of the main stratigraphic boundaries in the ZV: Base Kurkar, Top Mavki’im/Patish, Top Bet-Guvrin, Base Saqiye, Top Mount Scopus and Top Judea. The structural maps were based on data collected from 200 water boreholes (by Gvirtzman & Peleg 2006) and additional available seismic

20 information. The six structural maps were stored in “Geotiff” format file, a curvilinear surface containing spatial elevation data of the units. These Geotiff formatted surfaces served as the geological basis for the simulations presented in this work (Figure 5).

4.2 Raster file construction

SW4 supports a fully 3D heterogeneous material model that can be specified in several formats. The format chosen to fit the research aim best is the raster file (rfile). The rfile uses a binary, block-structured, data format that allows material properties to be represented with finer spatial resolution near the free surface of the earth. The first block describes the topography/bathymetry information (Petersson & Sjogreen 2017). The following blocks contain the subsurface data. Each block is assigned the properties (X,Y,Z, Δh, material properties) of the layer(s) within it. Each layer may be assigned different depth and material properties within the same block. The SW4 software reads and processes the information stored in each layer, creating a geometrical model. The final number of blocks depends on the extent of the research area and the required resolution. For example, the ZV model was constructed using two blocks; The first block is the topography block. The second block contains the properties of six layers, each with different varying depth and material properties. Each of the six layers within the second block is visualized in Figure 5.

Figure 5 - Visualization of the Tiff formatted curvilinear surfaces representing each of the main six geological subsurface reflectors in the ZV.

4.2.1 Zevulun Valley model structure

The ZV rfile was constructed of 2 blocks. Each block is a rectangular shaped cube, with the dimensions 34 x 22 x 6 km. This extent was decided upon based on the data stored in the Geotiff files together with the geographic boundaries of the valley. The grid spacing chosen for the rfile is 100 m both in the vertical and the horizontal direction.

Since the grid spacing of the simulation is limited by various factors (described in section 3, ‘Methods’), and the SW4 manual states that the grid spacing of the rfile should be larger than the general grid spacing of the simulation to achieve effective smoothing, 100 m grid spacing was selected for the rfile. The projection used is Transverse Mercator.

The first block contains the topography information above sea level. For the purposes of this work which focuses on the subsurface geology, the topography of the geological cross section eventually was flattened to a level of 0 meters above sea level. The actual elevation values in the ZV are up to 30 meters above sea level (at Eastern parts), which makes the flattened topography a reasonable approximation.

The second block contains the subsurface information for the valley; The material block is constructed by vertically stacking the geotiff curvilinear surfaces on top of each other, thus creating a geological frame to the valley. Preliminary work on the surfaces included reprojection, resizing, resampling and most importantly checking for stratigraphic inaccuracies, e.g. it was verified that the Top Judea reflector is always located bellow all other units. The seven main units in the valley: Kurkar Group, Yafo Formation, Mavki’im/Patish Formation, Bet-Guvrin Formation, Avedat Group, Mt. Scopus Group and the Judea Group. The material properties of each unit were inserted into the outline to complete the material model. Each layer in the model represents a spatially homogeneous geological unit. The exception is the Judea Group; In this model, the Top Judea reflector represents the top of the bedrock, which becomes denser with depth. Consequently, a velocity gradient was assigned from the Top Judea boundary downwards, starting with the shear wave velocity of 2000 m/sec (the shear wave velocity of the Judea Group), until blending with its surrounding general model at 2 km depth with the shear velocity of ~ 3500 m/sec.

4.2.2 Geological model inaccuracies

The process of constructing the ZV geological model involved some simplifications: - The 3D model is an idealized basin structure that represents the ZV but does not account for the extensive faulting within the valley.

The topography was flattened. The southern border of the ZV is Mount Carmel, and topography has proven to amplify ground motion as well. -The Kurdane reflector, a hard formation within the Kurkar Group was not modeled at this work. First, the formation has a very random appearance, which is very hard to model. Secondly, this is a very shallow and thin formation, that would be impossible to model within the narrow frequency range of this work.

The top Judea surface in the rfile has lower seismic velocity than the GITT05 model at the surface, creating a sharp difference in shear wave velocities (1.5 ratio), leading to a certain amplification in the rfile domain boundaries, affecting also the YGR reference station which is supposedly unamplified.

4.3 Simulation setup 4.3.1 Reference model

Calculating the amplification factor of ground motions requires a reference. There are two types of references which will be used throughout this work:

1. Reference model - a regional, laterally homogenous layered model (no-basin). The no- basin model simulates the propagation of the seismic energy without effects related to impedance ratio due to the presence of soft sediments, basin effects, edge effects, etc. The ratio between the basin model and the no-basin model yields an amplification factor based on the highest velocity values recorded during every simulation. The reference “no basin” model in this work is a layered model of Israel’s subsurface velocities, derived from the unified velocity model described by Gitterman et al. (2002), known as the GITT05 model. The material properties assigned to each layer are summarized in Table 1.

Table 1. Material properties of the simulation based on GITT05 general model.

Depth [km] Vp [km/s] Vs [km/s] 휌 [gr/cm3] Qp Qs

0 4.75 2.85 2.49 532 266

2 5.72 3.41 2.66 744 372

9.76 6.27 3.66 2.77 860 430

16 6.53 3.79 2.84 928 464

28 6.97 4.02 2.96 1058 529

Figure 6 - Visualization of the regional GITT05 model, and the approximate location of the ZV.

2. Reference site - as was discussed in the introduction section, ground motion amplification and consequently, more physical damage have been observed atop soft sediment sites but not atop hard rock sites, located at equal epicentral distances from the earthquake source. Thus, to estimate the amplification as a function of frequency content, a rock site (top Judea Group) in Kibbutz Yagur, adjacent to the research area, served as a reference station for this work, similarly to the instrumental deployment of Shani-Kadmiel et al. (2018).

4.3.2 Simulation parameters

Model size - All four simulations have the same size and parameters, apart from the source type and location and the general geographic coordinates of the model. The simulations have a length of 85 km, width of 104 km and depth of 19 km. The horizontal measurements were selected to accommodate the physical dimensions of the ZV, the path to the source, the source epicentral location and the supergrid layers. Moreover, the model extent must be sufficiently large to accommodate almost 15 km of faulting northwards for the DSM simulations. The vertical dimensions were chosen to include the locking depth of the source (9 km) and the supergrid layer size.

Zevulun Valley material properties – The shear wave velocities and densities of the geological units in this work were based on the paper of Gvirtzman and Louie (2010). Pressure-wave velocities were calculated by Vp = √3•Vs. The quality factors were determined by Qs = 20Vs (in km/s) for Vs<1.5 km/s, Qs = 100Vs (in km/s) for Vs>1.5 km/s, and Qp = 1.5Qs (Gvirtzman & Louie 2010). The properties for all valley units are summarized in Table 2. Material properties of the main units in the ZV.

The seismic source - Two types of seismic sources were used to initiate seismic-wave propagation in the ZV: a single point source (PS) and a finite fault with Distributed Slip (DSM). The DSM is comprised of a sequence of 15655 point-sources distributed over a rupture area of 50 km2, simulating a strike slip fault with energy distribution of 80%

18 northward directivity (see Figure 7), resulting with a total moment release of 10 (M0), the equivalent of Mw 6. The Mw 6 was chosen due to computational limitations; Since

26 magnitude order is a function of rupture area, higher magnitude would increase the size of the model beyond computational ability. The focal mechanism of the source is a 90° dipping strike-slip fault with a strike of 5 degrees and a rake of 0 degrees.

Table 2. Material properties of the main units in the ZV.

Vp [m/sec] Vs [m/sec] 휌 [gr/m3] Qp Qs

Kurkar Gr. 520 300 2000 6 9

Yafo Fm. 1040 600 2200 12 18

Mavki'im Fm. 2600 1500 2500 150 225

Bet-Guvrin Fm. 1380 800 2500 16 24

Avedat Gr. 1910 1100 2500 22 33

Mt. Scopus Gr. 1560 900 2500 18 27

Judea Gr. 3460 2000 2700 200 300

Figure 7 - The distributed slip model used to simulate a strike-slip fault in the ZV model.

Source depth was set to 9 km in accordance with the locking depth suggested by Hamiel et al. (2016), for both the JGF and the northern segment of the JVF.

Each point source is assigned a Gaussian velocity-time function given by:

휔 2 2 (18) 푔(푡, 푡 , 휔) = 푒−휔 (푡−푡0) /2 0 √2휋

With a frequency parameter ω = 1.88 radians/sec, giving a fundamental frequency of 푓0 = 0.3 Hz. The simulations were performed using the deterministic low-frequency approach (fmax<1 Hz), with a maximal frequency of 0.75 Hz, based on the lowest shear- wave velocity in the model Vsmin =300 m/sec, the minimum grid points per wavelength P=8 and the grid spacing h = 50 m.

The Gaussian velocity-time function is typically used in earthquake simulations since over large time periods the displacement tends towards a constant value and the velocity and acceleration tends towards zero.

Grid size – The grid for the simulation was constructed using mesh refinement; The 6000 m (from top downwards) were modeled with a spacing of 50 m, obeying the SW4 manual rule of hmodel < hrfile for smoothing purposes. First refinement doubles the grid spacing (h=100) for the next 1200 m. Second and final refinement at 7200 m depth, doubles the grid spacing to h=200 m until the bottom of the simulated block, at 19000 m. The supergrid, given h=200 m, forms a 6000 m layer. The resulting number of grid points is ~ 453x106, determined based on the mesh refinement levels, detailed in Table 3.

Table 3. Grid size structure detailed by mesh refinement levels. h is the grid spacing, Nx, Ny, Nz are the number of grid points in each direction resulting from the grid spacing.

Grid Ztop h Nx Ny Nz Points

0 0 50 2081 1701 121 428.3.106

1 6000 100 1041 851 13 115.2.106

2 7200 200 521 426 60 133.2.106

Simulation time - Simulation time was set to 80 seconds to allow the slowest waves to propagate across the entire simulated domain. Each simulation was executed on NES using 68 processors and took approximately 145 hours.

4.3.3 Summary of Zevulun Valley simulations

To understand the effect of source location on the ground velocities and the amplification values atop the ZV structures, two main sources were chosen along the DST; The Jordan Gorge fault (JGF) and the Kinnarot source (KNR), northern and southern sources respectively (Figure 8). The JGF represents an active segment of the Jordan fault according to the ‘map of Active and potentially active faults for Israeli Standard 413’ (Sagy et al. 2017). The Kinnarot source represents the most northern part of the Jordan Valley Fault (JVF). Hamiel et al. (2016), used detailed GPS measurements across the northern section of the JVF and the JGF. They found a slip rate of 4.1 ± 0.6 mm/year at both faults, the JGF fully locked above the locking depth (10 km) and the northern source of the JVF creeping from a depth of 1.5 ± 1.0 km to the surface with a creep rate of 2.5 ± 0.8 mm/year. In this work, the sources were modeled both as a point source and as a DSM source, resulting in four main simulations. The source types, epicentral locations and the shortcut names that will represent the simulations in this paper are:

• Point Source, Jordan Gorge fault (PS-JGF) • Point Source, Kinnarot valley (PS-KNR) • DSM source, Jordan Gorge Fault (DSM-JGF) • DSM source, Kinnarot valley (DSM-KNR)

Both sources – JGF and KNR, are located about 50 km from the ZV PBZN station.

4.4 Output plan

The output options that are of interest to this research are 2D cross-sections and time- velocity arrays achieved by placing synthetic stations across the research domain. Obtaining 2D data requires specifying the point of time or time intervals at which to save

29 the data array. Twelve synthetic stations were set at key locations inside the ZV, to get optimal coverage of the main subsurface structures comprising the valley. Synthetic stations were also placed at the main populated and industrial areas in the research domain. Six stations were assigned to known coordinates based on a portable network deployed by Shani-Kadmiel et al. (2018) – the ZV net. For 16 months during 2014-2015, seismic ground motions at the ZV were recorded by the network. The stations are: Yagur (YGR), Bazan (PBZN), Neve Avraham (NVHB), Kfar Hasidim (KHSD), Kfar Maccabi (KMKB) and Afek (AFK). Three more stations were placed in the major ‘Krayot’ cities, to measure ground motion in: Kiryat Yam (KY), Kiryat Motzkin (KM) and Kiryat Bialik (KB). Two stations were placed inside the HG: Ein Ha Mifratz (EHM) and Hilazon (HLZ) and finally, Haifa airport (AP) and Refael facility (Figure 8).

For the following analysis, five main station and one additional reference station are suggested to best represent the valley: PBZN, NVHB, KMKB, AFK, EHM and YGR reference station. Since real data recordings already exist for these six stations (Shani- Kadmiel et al. 2018), it is reasonable to compare the existing empirical data to the results obtained with the 3D numerical model for a more extensive analysis.

Figure 8 –(a) Research area and the source locations. (b) ZV synthetic stations location. Red triangles represent the ZV net, that was used to analyze the results in this study. White triangles represent all other stations. 5 Results 5.1 PGV results

Ground motions atop the ZV for each of the modeled source scenarios are shown in Figure 9 and Figure 10. The results are presented as a series of PGV maps, received from an 80 second simulated ground motion. The left panel are the results of the point source (PS) simulation while the right panel are the results of the DSM simulations. For every PGV map the amplification ratio (AR) map is attained by dividing the basin model PGV values by these of the reference model. Table 4 contains the summary of the PGV and AR values for each simulation.

Table 4. PGV values and amplification values for each of the four simulations. Each value is representative of a PGV map and a basin to no basin amplification map.

PS-JF PS-KNR DSM-JF DSM-KNR

PGV [m/sec] 0.267 0.302 0.098 0.16

Peak Amplification 13 13 10 11 Value

The general key observations are: 1. Point source simulations generate higher PGV values than DSM sources, with up to 0.3 m/sec for point source, and less than 0.16 m/sec for DSM source. 2. The highest PGV values are observed atop the QG. 3. The location of the source in relation to the basin controls the amount of energy that enters the basin, and therefore the PGV values. E.g. the Jordan fault source causes smaller ground velocities in the QG than the Kinnarot source because of its northern location. 4. The ground motion amplification ratio inside the QG varies from 6 to 13 for the point sources, 10 for DSM-JF and 11 for DSM-KNR model. The maximal amplification ratio values are observed atop the deepest section of the graben. 5. In the HG we observe a maximum amplification ratio of 4. 6. Although the PGV values for the DSM models are lower than for the PS, the amplification ratios obtained for DSM simulations are close to PS amplification values.

A Point Source model simulates the release of seismic energy all at once, from a single point, thus producing higher amplitudes and PGV values than the Distributed Slip Model. The PS results are reported in this work as part of the modeling process. No further results will be shown or discussed, since the PS model does not physically describe the source mechanism of the DST.

JGF PGV results

Velocity [m/s] Velocity

Figure 9 - (a), (b) PGV results for the JGF simulations. (c), (d) Amplification ratios for each simulation, respectively.

KNR PGV results

Figure 10 - (a), (b) PGV results for the KNR simulations. (c), (d) Amplification ratios for each simulation, respectively.

5.2 Seismograms, arrival times and spectral analysis

At this part, the results will be presented for the northern source and the southern source separately demonstrating how each source location affects the ground motion of the three major parts of the ZV structure: the QG, AH and the HG. The peak values recorded in the seismograms are summarized in Table 5. The depth to main reflectors for the six stations is presented in Figure 11. The following analysis is based on the SW4 coordinate system where X axis represents NS and Y axis represents EW.

Table 5. PGV values derived from synthetic seismograms in m/sec.

JGF KNR

PBZN 0.0253 0.0933

NVHB 0.0226 0.0601 KMKB 0.0160 0.0467

AFK 0.0151 0.0314

EHM 0.0409 0.0434

YGR 0.0095 0.0313

Figure 11 - Reflectors at depth for the six representative stations of the ZV, derived from the 3D subsurface model built using the tiff curvilinear surfaces (Figure 5).

Figure 12- Figure 21 contain the results of our analysis for the two DSM models, for each of the five Valley stations. Figure 12- Figure 16 are for JGF. Figure 17 - Figure 21 are for KNR. YGR is the reference station and is in all the Figure 12 - Figure 21. Each Figure 12 - Figure 21 consists of six images as follows: (a) Synthetic velocity seismogram of the basin station, representing 80 seconds duration of simulation. (b) Synthetic velocity seismogram of the reference station, representing 80 seconds duration of simulation. (c) Fourier spectra derived from each stations time-history, for the three components X,Y,Z. Whereas typically the geometric mean of the horizontal components is presented in spectral analysis, we decided to present both X and Y components in order to detect and differentiate the seismic phases entering the basin. (d) H/H analysis – basin horizontal spectral velocities to reference station horizontal spectral velocities. (e) H/V analysis - basin horizontal spectral velocities to basin vertical spectral velocities. (f) Velocity-depth profile, with notations of the major seismic reflectors of the station.

5.2.1 Northern Source (JGF) – observations

Qishon Graben

The following three stations are located atop the QG – PBZN, NVHB and KMKB. The PBZN and the NVHB stations are situated above a complete geological section of the QG, including the Saqiye Group. In general, the amplification is mostly dominated by the X direction. The spectral ratios show higher amplitudes for the H/V analysis (both horizontal components considered), while the H/H analysis generates more moderate amplitudes and will be described in this next part.

The PBZN station is located on top of the deepest sedimentary column among the stations, approximately 1400 m deep. The ground motion duration recorded at this site (Figure 12a) is considerably longer than at the other stations inside the graben or the YGR reference station. The station recorded 15 seconds of amplified waves at both

37 horizontal directions with ground velocities up to 0.025 m/sec. At the reference station YGR, very small PGV values are observed, less than 0.01 m/sec. Spectral analysis shows 2 main Y direction dominated peaks at ~ 0.2 and ~ 0.35 Hz with ratios of 9 and 6 respectively (Figure 12d).

The NVHB station represents a 1170 m deep geological section, demonstrating ground motions of up to 0.022 m/sec (Figure 13a). While the X component behavior is similar to the PBZN station, there are no high resonance-dominated amplitudes detected in the Y direction for this station. Yet, the amplitudes received take a longer time to diminish. The dominating peak is at 0.2 Hz (in the Y direction) with an amplification value of 3 (Figure 13d).

The KMKB station is situated in the eastern, shallower parts of the graben with a 400 m deep sedimentary column. This area of the graben lacks the Saqiye Group from its geological record. Ground motion velocities of up to 0.016 m/sec were recorded at the station (Figure 14a). The spectral amplification values at this station are no higher than 2 in both horizontal directions, at frequencies of ~ 0.15 and ~ 0.3 Hz and ~0.5 Hz (Figure 14d).

The KMKB station’s depth, geological column and results of analysis are very similar to the KHSD station, therefore only the KMKB stations results are presented here.

Afek Horst and Hilazon Graben

The AFK station is representative of the AH and the EHM station of the HG. One more synthetic station was positioned in a different location inside the HG, but it showed similar results to the EHM station and therefore is not presented here. The results for AFK and EHM are presented in Figures Figure 15 - Figure 16.

The station located on top of the horst shows small velocity amplitudes, with maximum value of 0.015 m/sec (Figure 15a. The spectral amplification ratios received for this station are very low, up to 2 times amplification at ~ 0.35 Hz in the H/H analysis (Figure

15d). A significantly higher value of 6.5 ratio has been received in the H/V analysis (Figure 15e).

The HG station produces the highest ground velocities in the entire valley for the JGF source scenario, with a maximum amplitude of 0.04 m/sec (Figure 16a). Spectral amplification is around 6 at the frequency of 0.35 Hz (Figure 16d).

5.2.2 Southern Source (KNR) – observations

Qishon Graben

The Kinnarot source generates considerably higher velocity amplitudes than the Jordan Gourge Fault source; PBZN station displays the highest ground motion values in the valley, with the peak value of 0.093 m/sec in the X direction and a rather high 0.05 m/sec amplitude in the Y direction (Figure 17a). The seismogram recorded considerably high amplitudes in both horizontal direction for 15 consecutive seconds. High amplification of the horizontal components can imply energy trapping caused by SH and Love waves reflected from the vertical walls of the basin, e.g. edge effect. The H/H spectral analysis shows 2 main peaks, at ~ 0.2 Hz (Y direction) and 0.5 Hz (X direction) with the amplification ratios of 9 and 6.5 respectively (Figure 17d).

At the NVHB station, maximum amplitude of 0.06 m/sec (Figure 18a) was observed. The high amplitudes in this station recordings are dominated distinctly by the X component for a period of 10 seconds, then all phases decrease almost to zero, then another resonance (of a lesser amplitude) commences for 12 seconds. Spectral amplification ratios are up to 4 at 0.17 Hz and 3 at 0.35 Hz (Figure 18d).

The remaining graben station KMKB shows 7 seconds of high amplitude of up to 0.046 m/sec (Figure 19a). The spectral ratio observed with the H/H method is 4 at 0.12 Hz (Figure 19d).

Afek Horst and Hilazon Graben

The AFK station displays higher velocity amplitudes for the southern source (up to 0.03 m/sec, as seen in Figure 20a), but generally shows the same characteristics as in the northern source scenario. The spectral ratio is 3 for ~ 0.12 Hz (Figure 20d). The highest velocity value received at the YGR reference station is also 0.03 m/sec, with similar wave characteristics as of the recorded AFK station (Figure 20b).

The Hilazon station exhibits slightly different behavior at this setting (Figure 21a); relatively high amplitudes can be observed in the X direction (with maximum of 0.043 m/sec), and smaller amplitudes at the Y direction for a 13 seconds period. At that point, a 10 second relaxation is observed in all directions, followed by a less powerful, yet noticeable, reverberation for another 10 seconds in both horizontal components. The spectral ratios received for 0.3, 0.7 Hz are ~ 2 and ~4, respectively (Figure 21d).

JGF Results

Figure 12 - Results obtained from the PBZN station; a) synthetic seismogram for PBZN. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 13 - Results obtained from the NVHB station; a) synthetic seismogram for NVHB. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 14 - Results obtained from the KMKB station; a) synthetic seismogram for KMKB. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 15 - Results obtained from the AFK station; a) synthetic seismogram for AFK. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 16 - Results obtained from the EHM station; a) synthetic seismogram for EHM. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

KNR Results

Figure 17 - Results obtained from the PBZN station; a) synthetic seismogram for PBZN. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 18 - Results obtained from the NVHB station; a) synthetic seismogram for NVHB. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 19 - Results obtained from the KMKB station; a) synthetic seismogram for KMKB. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 20 - Results obtained from the AFK station; a) synthetic seismogram for AFK. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

Figure 21 - Results obtained from the EHM station; a) synthetic seismogram for EHM. b) synthetic seismogram for YGR reference station. c) Fourier spectra of the basin station. d) Amplification spectral ratios obtained using the H/H method. e) Amplification spectral ratios obtained using the H/V method. f) Velocity-depth profile based on Figure 11 and Gvirtzman & Louie (2010) shear velocities for each layer in the ZV model.

5.3 Result interpretation 5.3.1 PGV analysis

The PGV maps and the synthetic seismic recordings show higher ground velocities for the southern source than for the northern source - this is most evident in the QG and the AH, located at the southern and central parts of the valley. Atop of the deepest sections of the QG (PBZN and NVHB stations) the measured PGV values are 3.5 times higher for the southern source and at the margins of the graben (KMKB station) the recordings show nearly 3 times higher ground velocities. At the AH, represented by the AFK station the northern Kinnarot source causes twice as high PGV values than the southern Jordan Gorge Fault source. The differences cannot be explained by distance from the source to the site, as it is similar for both sources. For example, the distance from the JGF source to the PBZN station is 55.5 km, while the distance from the KNR source to the same station is 55.2 km, only 300 m difference. These notable differences in amplitudes can most probably be explained by the amount of seismic energy reaching the valley as a function of the source location and rupture directivity. As demonstrated in Figure 22, the energy spreads through the medium in the form of four velocity lobes, according to the shear wave double couple model radiation pattern for vertical strike- slip faults. It is clearly shown in the Figure that in the southern source simulation (KNR) the energy from the westward lobe propagates directly towards the valley. However, in the case of the northern source simulation (JGF) most of the westward lobe energy is radiated north of QG mostly into the HG, where highest values of ground velocity for the JGF simulation are recorded. Another important observation in Figure 22 are the PGV values at the source in comparison to the PGV values in the valley. Both simulations show that the ZV exhibits similar to source region PGVs, with KNR even more than so in the deepest parts of the QG.

DSM-JGF DSM-KNR

Figure 22 - Full PGV maps for (a) DSM-JGF and (b) DSM-KNR.

5.3.2 Energy and duration of motion

While the PGV maps present an absolute maximum value of ground velocity throughout the simulation domain, they give no indication of the energy distribution in time, or the duration of ground motion. The subsurface structure of the Qishon and Hilazon Grabens causes ground motion amplification due to the impedance ratio between the basin boundaries and the sedimentary filling, as well as the impedance ratio between the layers of the sedimentary column comprising the basin (for example the unconsolidated Kurkar sands or the soft Yafo clays lying atop the hard Mavki’im Fm. In the QG). The reflection of the waves from the walls of the basin (edge effect), and the interference of waves trapped inside sedimentary layers amplify the waves amplitude and prolong the duration of the ground motion. Arias intensity plots allow to examine the energetical distribution in time, or the amount of energy attributed to the basin structure.

Arias intensity (IA) is defined:

휋 푇 (19) 퐼 = ∫ 푑[푎(푡)]2푑푡 퐴 2푔 0

Where g is gravity, and Td is the duration of the seismic signal.

Figure 23 presents four IA plots, three for the QG stations PBZN, NVHB and KMKB and one for the HG EHM station, derived from the KNR-DSM simulation results. An immediate observation is that the X component produces significantly higher Arias intensity than the Y component. This supports the observation that the dominating phase in the valley are Love waves, propagating from the eastern source westwards. Two important observations at the PBZN station (Figure 23a) are: 1. The horizontal component ratio (X/Y) is the lowest amongst the QG stations (Figure 23c,e). The X component contributes a total amount of 1.2 m/s to the motion and the Y component contributes 0.3 m/s (ratio = 4). In the NVHB and KMKB stations (Figure 23c, e) the ratio is 8 and 14 correspondingly, meaning that the Y component contributes less energy, moving from the deepest parts of the graben outwards. Also, the time to 5% and 95% of the energy (D595) in the station for both horizontal components is very close together (39 seconds for the X component and 44 seconds to Y). This might imply, as was suggested earlier, interference of waves near the surface of the basin, contributing to the motion in the Y component. 2. The accumulation of 95% of the energy happens in a very short time, for the X component the D595 is 8 sec compared to 11 sec in the Y component. This second observation of strong amplification over a short time might implicate that waves are trapped inside a sedimentary layer, then slowly weaken and migrate outside the trapping layer, and no new motion is detected afterwards in this station.

The HG station EHM demonstrates two episodes of amplified ground motion. First amplified motion dominated by the X phase, similarly to the other stations discussed is probably a result of the Love waves traveling E-W from the source to the valley. The second amplified motion consists of both X and Y phases, leading to a possible conclusion that some of the energy trapped and amplified in the QG migrated northwards towards the HG, thus creating the motion in the Y axis.

Figure 23 - KNR-DSM results: (a) PBZN Arias intensity plot (b) PBZN Velocity seismogram, (c) NVHB Arias intensity plot (d) NVHB Velocity seismogram, (e) KMKB Arias intensity plot (f) KMKB Velocity seismogram, (g) EHM Arias intensity plot (h) EHM Velocity seismogram.

The wave propagation inferred from the Arias intensity plots can be further supported by the velocity magnitude maps obtained from the DSM-KNR 3D simulation. Figure 24 shows the velocity magnitude and propagation pattern through time. The results at sea were not masked as they show the propagation pattern of the waves more clearly. As can be seen in the Figure, the energy propagates from the southern source, reaching the valley from east (26 sec to 32 sec frames). The waves are then trapped in the

52 southern graben (40 sec frame), amplified and migrate northwards to the HG (45 sec and 50 sec frames). The waves are clearly reflected from the southern boundary, the Yagur fault, into the QG which creates an edge effect that clearly contributes to the amplification of waves traveling northward.

Figure 24 - Velocity magnitude propagation with time in the KNR-DSM model.

Similar behavior is observed in the northern source simulation. The waves are traveling to the ZV from north-east, reaching first the HG, therefore the EHM station records the highest ground motion velocities and cumulative energy (Figure 25c,d). Part of the waves are trapped inside the HG, then migrating southwards to the QG, prolonging the motion in the Y direction. This can be observed in the Arias intensity plot of the PBZN station (Figure 25a) that shows accumulated motion in the X direction, closely followed by motion in the Y direction, both with similar IA values.

Figure 25 - JGF-DSM results: (a) PBZN Arias intensity plot (b) PBZN Velocity seismogram, (c) EHM Arias intensity plot (d) EHM Velocity seismogram. 5.3.3 Spectral analysis

Spectral amplification is observed throughout the entire modeled frequency spectrum. Typically, two amplification peaks are evident for each horizontal component. The next part distinguishes the material related amplification from structure related amplification, by associating certain frequencies at a station to its subsurface geological configuration at depth. The modal frequency, fn, for vertically incident waves are: 푉 (20) 푓 = (2푛 − 1) 푠 푛 4퐻 where Vs is shear velocity [m/sec], H is the thickness of the layer and n is the mode.

Correlation between the amplified frequencies and possible seismic reflectors

The range of frequencies explored in the simulations is relatively narrow with maximum modeled frequency of 0.75 Hz (low frequency zone). However, deep basins show amplification in the low frequency zone, therefore the analysis presented here is pertinent. In this section, the main results from the spectral analysis are presented for

54 the three structures of the ZV. Spectral analysis provides information about the frequencies dominating the amplification at each station. The H/H and the H/V calculations do not always produce the same results. Often, resembling frequencies are amplified, yet the H/V analysis produces different amplification values than the H/H (Bonilla et al. 1997, Lermo & Chavez-Garcia 1993, Kuo et al. 2015, Rong et al. 2018). The H/V method assumes that when a shear-wave travels through soft sediments its horizontal component is amplified whilst its vertical component does not (Nakamura 1989). However, the results show some amplification at the reference YGR station as well (Figure 12-Figure 21, b). These differences could also be the result of minor errors that were made during the construction of the model and or the result of numerical effects that occurred during the running of the simulations. The following correlation between amplified frequencies and seismic reflectors refers to H/H analysis results;

Qishon Graben

The deepest section of the QG is characterized by the presence of the Saqiye Group; PBZN and NVHB stations are located atop a full section as such, where the hard Mavki’im/Patish Formations of the Saqiye Group overlay the soft clays and marls of the Yafo Formation, with an impedance ratio of ~ 2.5. This material transition was marked as a potential important seismic reflector in previous sections. The H/H analysis for the PBZN station (Figure 12d and Figure 17d) shows a ~ 0.5 Hz peak that corresponds to the Top Mavki’im reflector. Though the NVHB subsurface has a similar geological structure as the PBZN, the Top Mavki’im reflector appears to be too shallow (140 m deep compared to 250 m) to be detected within the simulated frequency range and should cause amplification approximately at 0.8 Hz. Another notable peak is ~ 0.18 Hz for the PBZN station and ~ 0.2 Hz for the NVHB station (Figure 12, Figure 13, Figure 17, Figure 18, d), both consistent with the Top Judea reflector, located more than 1 km below the surface.

The farther station of the graben, KMKB (Figure 14d and Figure 19d) demonstrates a ~ 0.55 Hz peak, corresponding to the Top Judea reflector, which lies only a few hundred meters deep.

Afek Horst and Hilazon Graben

The unconsolidated sands of the Kurkar Group located atop the AH form a very thin layer of ~ 30 m, underlain by a shallow Top Judea reflector. This reflector cannot be detected within the limited simulated frequency range of this work. The estimated frequency to be amplified due to the Top Judea reflector in the horst is ~ 2.7 Hz. Gvirtzman & Louie (2010) found that ~ 2 Hz frequencies are amplified in the horst and Shani-Kadmiel et al. (2018) calculated amplification at ~ 3 Hz. both studies support the estimation made above. The HG station displays a ~ 0.3 Hz peak (Figure 16d and Figure 21d), corresponding to the Top Judea reflector at depth of 860 m. Additionally, another peak has been detected at ~ 0.38 Hz and may be related to the Base Saqiye reflector, at depth of 180 m, underlying the softer Bet Guvrin Fm. with an impedance ratio of ~ 1.4.

Summary of the amplified frequencies correlating to a reflector representing the synthetic stations is presented in Figure 26.

Figure 26 – Amplified frequencies correlating to a reflector at the PBZN, NVHB, KMKB and EHM stations.

6 Discussion 6.1 Amplification maps

The results presented in this work are the first of their kind synthetic ground motion amplification values obtained using 3D modeling for the ZV. Using the 3D numerical method, one can estimate the seismic hazard for this high risk, vulnerable area before the next strong motion earthquake occurs. One of the main contributions of this work are the amplification maps created based on the basin to no-basin PGV ratios obtained from the simulations. In a previous work, Zaslavsky et al. (2008) used HVSR measurements of ambient vibration to determine local site effects in the QG in Haifa Bay. The results were presented in the form of amplification ratio maps for the first and second fundamental resonant frequencies. Figure 27a,b shows the first resonance frequency peak range (0.4-4 Hz) and amplification map from the work of Zaslavsky et al

(2008). The second resonance peak exceeds this work’s frequency limit (fmax3D = 0.75 Hz, fZaslavsky et al. (2008) = 1-10 Hz) and therefore is not presented here. Figure 27c refers to the basin to no basin ratio results obtained from the Kinnarot DSM 3D model. The location of six seismic stations YGR, PBZN, NVHB, KMKB, KHSD and AFK, synthetic stations used in the 3D analysis, is marked on both figures, thus allowing to discuss the results from both maps for chosen locations.

The results obtained in this work are partially in agreement with Zaslavsky et al. (2008). Similar values of amplification are observed in the shallow parts of the QG (KMKB, KHSD) and at the YGR reference station, with amplification values of 2-3. However, the 3D modeling yields higher amplification values in the deeper parts of the QG (PBZN and NVHB stations) with values 1.5 to 2 times higher than shown in the previous work. For example, Zaslavsky et al. (2008) shows amplification value of 4 for the PBZN station location, while the present analysis suggests a higher value of 7. Similarly, the NVHB station in the previous study shows 2-2.5 amplification ratios, compared to 4 in the present study.

Figure 27 - (a) first resonance frequency peak range from Zaslavsky et al. (2008). (b) Amplification ratios map for the first resonance frequency peak obtained by Zaslavsky et al. (2008). (b) Amplification ratios map based on the present work’s DSM-KNR model results.

The differences in values may be explained by the fact Zaslavsky et al. (2008) research was conducted using ambient noise recordings while at the present work 3D modeling was used to simulate an elaborate subsurface structure of the research area. Additionally, Zaslavsky et al. (2018) used the HVSR spectral method to obtain the amplification ratios for the Valley. At the present work, the values in the amplification maps are based on the basin to no-basin ratio. These values correlate well to the ratios obtained using the H/H method (Figure 17-Figure 19, d) due to the narrow frequency band modeled.

The results of the H/V analysis (Figure 17-Figure 19, e) suggest much higher amplification values than those obtained using the two other methods. The HVSR Nakamura 1989 method assumes that the vertical component of motion is not amplified. This method, although simplified and is largely used in site response studies, is still debated in the scientific literature. Rong et al. (2017) among others, suggested Nakamura’s method correctly depicts the amplified frequencies but not the amplification factor, which might explain the discrepancies between the presented maps. The H/V results in the present work are not validated using the results from Zaslavsky et al. (2008) for the following reasons: first, In Figure 12 - Figure 21, some amplification is detected in the vertical component (Nakamura 1989 basic assumption), therefore the HVSR method may not yield reliable results. 2. H/V is a local, point-specific method and the synthetic stations do not exactly match the locations of Zaslavsky et al. (2008), thus the comparison is ineffective.

Shani Kadmiel et al. (2018) summarized their results in a table, comparing the frequency and amplification they obtained in the PBZN, NVHB, KMKB and KHSD stations, to the frequency and amplification values derived from Zaslavsky et al. (2008) at the same locations. In Table 6, the HVSR method results from Zaslavasky 2008 are reported alongside the H/V results from this study. For the deepest parts of the ZV there is a good correlation between the amplified frequencies and the amplification values as well. For example, at the PBZN station, frequency of 0.6 Hz and amplification value of 4 (Zaslavsky

60 et al. 2008) correlate well to frequency of 0.55 Hz and amplification value of 4 (present study). Similarly, at the NVHB station, frequency of 0.5 Hz and amplification value of 2 in the previous study are in good agreement with frequency of 0.55 Hz and amplification value of 3 in this study. For the KMKB station no correlation was found, both for the amplified frequency and the amplification value (Table 6). 3. Because of the limited amount of synthetic stations in our model, a map based on our H/V results is physically impractical. Moreover, the main intention in Figure 27 is not comparison between this work’s and Zaslavsky et al. (2008) results; our aim is to observe the patterns of regional amplification values, and not the focus on point-vise values. We mention specific values as part of the discussion, but the main impetus of this work are the regional patterns, which are clearly obvious from both Figure 27b and Figure 27c: There is a deep structure related amplification atop the deep parts of the QG, with typical amplification at low frequencies (Figure 27).

Gvirtzman & Louie (2010) performed 2D numerical analysis of wave propagation within the frequency range of 0.2-6 Hz in the ZV, for a chosen cross section. Both the 2010 model and the new 3D model use the same velocity model (see ‘Model Setup’ chapter), which makes the comparison between them more relevant. Shani-Kadmiel et al. (2018) presented amplification factors and frequencies in the ZV based on instrumental measurements collected for 16 months period (over the years 2014-2015) using transportable network consisting of six stations in the ZV, mentioned earlier in this chapter. Shani-Kadmiel et al. (2018) provided a table of the first and second frequency peak and their associated amplification factors for the stations PBZN, NVHB and KMKB from three independent methods: the HVSR method (Zaslavsky et al. 2008), 2D numerical method (Gvirtzman & Louie, 2010) and the Spectral Ratios method (arithmetic mean of the horizontal components) applied in their research. Here the table for the first frequency peak is complemented by the 3D numerical model results (Table 6).

Table 6. Comparison of spectral analysis of ZV conducted by 4 different studies: Zaslavsky et al. (2008) - HVSR, Gvirtzman & Louie (2010) – 2D, Shani-Kadmiel et al. (2018) – 3D and the present

3D numerical study – Hstation/HYGR, H/V, basin to no basin PGV ratio.

HVSR 2D SR 3D H/H 3D H/V 3D PGV

f0 Af0 f0 Af0 f0 Af0 f0 Af0 f0 Af0 A

PBZN 0.6 4 0.4 5 0.5 7 0.5 6.5 0.55 4 7

NVHB 0.5 2 0.5 5 0.4 4 0.2 3.5 0.55 3 4

KMKB 0.8 2 N/A N/A 0.8 3 0.55 2 0.6 8 3

As discussed by Shani-Kadmiel et al. (2018) in their paper, the SR method produces higher amplitudes than those of the HVSR and 2D studies. The 3D amplification values agree well with the SR results, though are not assigned to a frequency spectrum. However, the spectral analysis using the H/H method gave inconsistent results. Whereas, the deep section station PBZN showed good agreement with previous studies, other stations do not correlate well with the other frequency and amplification values. For instance, previously the NVHB station yielded amplification values of up to 5 within the frequency range of 0.4-0.5 Hz, while the present work detected a 0.2 Hz frequency peak and an amplification value of 3.5. Yet, one should recall that the YGR reference station showed some amplification, thus lowering the amplification ratio of the basin stations. This means that the amplification factors in the H/H analysis may be underestimated. The previous studies on the ZV site response had focused mainly on the amplification atop the QG, that represents the widest and deepest geological column in the valley, resulting from two main seismic reflectors, the top Judea and top Mavki’im, the latter being absent outside the QG. This study expands the research area to the AH and the HG, providing first insights into ground motion amplification expected in the northern

62 parts of the Valley using 3D modeling. In the northern graben, besides the amplification related to the top Judea reflector (0.26 Hz, 3-4 amplification values). Another amplification was detected at ~ 0.4 Hz, possibly due to the Base Saqiye reflector. Although the Bet Guvrin Fm. Situated on top of the Avedat Gr. represent low impedance ratio (1.4), the narrow graben might cause a trapping effect, prolonging and amplifying the motion to an amplification value of 6 (Figure 16d).

While this research focuses on the subsurface geological structures of the ZV, it can also provide information on ground motion amplification in the cities located within the research area; Figure 28 presents the amplification ratios map overlain by polygons representing the major municipal and industrial boundaries in the research area. The map shows similar amplification values for most of the populated Krayot cities, including Kiryat Yam, Motzkin, Bialik and Ata, with amplification values of up to 3. The south- western part of Kiryat Ata and the Kiryat Haim neighborhood (part of Haifa municipality) are situated on the margins of the QG and demonstrate amplification ratios between 3 and 4. The industrial area of the Haifa Bay and the Haifa Port facilities are situated atop of the deepest part of the QG, with amplification values of up to 10.

An observation worth mentioning is the 3 to 4 amplification ratios observed for the city of Acco, and even a higher value of 5 was detected in the southern part of the city. Although there was no initial goal of modeling areas outside the boundaries of the valley, the elevation surfaces used for constructing the 3D model (see ‘Model Setup’ chapter) had the subsurface data of Acco city stored on them, based on borehole data, structural maps, etc. Since there was no intention of modeling the outside of the valley, no synthetic recording stations were placed atop of Acco and no further results will be shown.

2 3

6 5

Figure 28 - Amplification ratios map based on the KNR-DSM simulation, combined with polygons of 1. Kiryat Yam 2. Kiryat Motzkin 3. Kiryat Bialik 4. Kiryat Haim 5. Kiryat Ata 6. ZV industrial area. 7. Acco

The amplification values obtained in the simulations, although high, are not unprecedented. In Mexico City, following the 1985 Mw earthquake and later the 2017 earthquake, spectral amplification ratios larger than 7 were observed (Romo & Seed 1986; Seed et al. 1987; Sahakian et al. 2018). In Takai et al. (2016), spectral amplification ratios larger than 10 are discussed for the Kathmandu valley during the 2015 Mw 7/8 Gorkha earthquake.

6.2 Amplification in the Qishon Graben

High amplification values were observed atop the deepest parts of the QG; The PGV maps showed amplification ratios of up to 12 (Figure 9c, d and Figure 10c, d) and the H/H analysis for the PBZN station displayed spectral amplification values of 9 at ~ 0.18 Hz (Figure 12d and Figure 17d). The amplified frequency can be correlated to the top Judea reflector, the main reflector considered in this work, located at the depth of 1400 m at this station (Figure 11). However, the impedance ratio between the Judea Gr. and the overlaying Mt. Scopus Gr. is only 2.4 and therefore cannot explain the high amplification values received.

Another plausible cause for the high amplification might be the presence of Rayleigh waves. When observing the vertical component on the PBZN station recordings (Figure 12a, Figure 17a), some amount of amplification can be detected, which supports the presence of Rayleigh waves. In Figure 29, H/H and Z/Z analysis of the PBZN and YGR reference station reveal amplification in both X and Z directions, at similar frequencies. This supports the hypothesis that Rayleigh waves enter the basin and possibly cause the high amplification observed.

Previous studies on site response suggest that the steep structure of the sediment- bedrock interface at the edge of a basin lead to phenomena like focusing and scattering of seismic waves (for example high mountains surrounding basins). The complex interaction of incident body waves and diffracted waves can generate secondary surface waves capable of causing significant ground motion (Davis et al. 2000, Yoshimoto & Takemura 2014, Pilz et al. 2018). It has been shown that not only the border itself but every point from the free surface to the deepest point of the basin along its wall will play a role in the generation of the surface waves (Narayan 2015). Figure 17a shows that amplification of ground motions in X and Z components begin 30 seconds after source initiation and lasts for approximately 8 seconds. During this period waves can be observed propagating along the south-western border of the valley, on both sides - the ZV and the Carmel mountain (Figure 24, 26 to 40 seconds frames). In these frames the

65 velocity magnitude inside the valley, at the adjacent QG is higher than on the other side of the Carmel fault. During this period 90% of the Arias intensity in the X component is attained (Figure 23a). Next, the waves are reflected from the fault wall northwards into the QG and their amplitudes decay gradually (Figure 17a and Figure 24, 45 and 50 second frames). As discussed, these waves might be secondary Rayleigh waves, originated at the south-western border of the valley and not necessarily from waves arriving from the source.

The challenge with waveforms recorded at sedimentary basin sites is they are often characterized by a complex superposition of various arrivals, which makes it difficult to identify the different phases and their attribute to the ground motion. To fully understand the wave propagation inside sedimentary basins, complex mathematical analysis must be conducted to separate waves arriving at a single site from different directions. In this way, secondary surface waves originating at various sites along the edge of the basin can be identified (Pilz et al. 2018).

Figure 29 - Component ratio analysis for the PBZN station for KNR-DSM model

6.3 Simulated frequency content

The simulated range of frequencies in this work is narrower than in other studies on this subject; The most promising approaches utilize a hybrid of comprehensive and simplified physics-based models at low frequencies and high frequencies, respectively (Somerville et al. 1996). Yet, this work was performed using the deterministic low-frequency approach only, with frequencies lower than 1 Hz (fmax = 0.75), resulting from computational limitations that were listed previously in the ‘Model Setup’ section of this work. From a geological point of view, amplification of motion in deep basins is caused at low frequencies, according to the 1D resonance frequencies equation (Eq. 15 in the ‘Results’ section). From an engineering point of view, Haifa Bay area is a densely populated area, with interlaced residential areas and industrial facilities. When the natural period of an engineering structure approaches that of the ground it will experience resonance, meaning in most cases prolonged and amplified motion leading to excessive damage. The main factor that affects the natural period of the building is the height of the building, namely the number of stories. Generally, taller buildings have longer periods, although buildings of the same height might have different periods due to their mass and rigidity. The analysis presented in this work, from a structural point of view, is relevant for structures with a low fundamental period residential building, tall industrial facilities such as oil refineries (e.g. > 1 S) and also long bridges.

Figure 30 presents the spectral amplification results in frequency range of 0.1-5 Hz obtained by Shani-Kadmiel et al. (2018), for stations PBZN, NVHB and KMKB, which were also simulated in this study. The results are within a frequency range of 0.1-5 Hz. They observed strong low frequency (<1 Hz) amplification factors (7 in average), and mostly lower amplification factors in the higher-frequency range (values up to 3). Most buildings have their fundamental frequency between 1-5 Hz (intermediate frequencies), and short-story buildings with high rigidity have a high fundamental frequency, therefore it is important to simulate and discuss the high frequency range as well as the low frequencies.

Figure 30 - Spectral amplification at the PBZN, NVHB and KMKB stations, after Shani-Kadmiel et al. (2018). 6.4 Source magnitude

The simulations in this work were modeled with magnitude Mw = 6. Archeoseismic studies specify two large earthquakes that ruptured the Jordan gorge segment, north of the sea of Galilee. First, the M>7 earthquake in 1202 displaced the Ateret crusader fortress by 1.5 m. Second, in 1759 the fortress structure was displaced by additional 0.5 m [e.g., Ellenblum et al. 1998, 2015]. For the “Kinnarot” segment (south of the sea of Galilee), known as the Jordan Valley fault northern segment (north of the intersection with the Carmel Gilboa Faria fault, Hamiel et al. 2016), historical records suggest two large earthquakes occurred in 749 and 1033, severely damaging adjacent populated areas. Based on these events the minimum slip deficit in the modeled segment of the DST is 2.5 m since 1202 or ~ 0.9 m since 1759. Sadeh et al. (2012) suggested that such slip deficits correspond to earthquake magnitudes between 7 and 7.4, based on the work of Wells & Coppersmith (1994).

Accordingly, the magnitude M 6 earthquake modeled in this work is the lowest possible considering the above described estimations, and although the simulated magnitude is not expected to affect the amplification values (under linearity assumption), the modeling of the higher estimated magnitudes in this section of the DST is crucial to comprehend the high PGV values they would cause, since PGV translates to intensity and consequently, damage.

To model an Mw =7 the rupture area must be considered. The seismic moment of an earthquake is given by

(15) 푀0 = 휇퐴퐷̅ where μ is the rupture shear modulus, A is the rupture area and 퐷̅ is the average amount of slip. Based on the next equation (Hanks & Kanamori 1977)

2 (16) 푀 = [log(푀 ) − 9.1] 푤 3 0

19 The moment equivalent to Mw = 7 is M0 = 3.98x10 . The approximate crustal rigidity index is µ = 3x1010 N/m. Average amount of slip for Mw = 7 is 5.2 m based on the relations of Wells & Coppersmith (1994). This requires the rupture area of minimum 1000 km2 within the simulation. This requirement cannot be effectively satisfied within the computational limitations of this research (see ‘Methods’ chapter), but on the expense of grid size, frequency limit and minimum shear velocity.

The simulated synthetic sources in this study were located along the Dead Sea Transform. These sources represent actual segments of the DST. The last recorded strong motion earthquake along these segments was over 90 years ago (1927 Jericho earthquake). Thus, seismic moment has been accumulating on the DST for almost a century with no significant release, therefore a strong motion earthquake can be expected in our lifetime. The DST sources simulate wave field in the far field, yet there is another probable seismic source in the near field, the Carmel Fault, bounding the ZV from south-west. The only earthquake attributed to this segment is the 1984 Mw=5.3

69 earthquake (Hofstetter et al. 1996), but since no recordings exist from within the ZV for the event, the relation between this earthquake and the Carmel Fault is unclear. In addition, it is more difficult to model the near field than the far field because there are more parameters to consider while modeling the effects on an area adjacent to the source, specifically the near fault displacements. Therefore, the Carmel fault was not simulated as a source in our work, although it might have an even stronger effect on the Valley following an earthquake, with extremely high PGV values and very destructive consequences.

6.5 Model Limitations

3D modeling of northern Israel sedimentary structures - The focus of this work was ground motion amplification atop the ZV. While the valley structures were modeled in high resolution, the rest of the simulated domain was constructed under the assumption of a unified Israeli layered model. Although the unified model GITT05 is an established model that represents the region, it may not be accurate specifically for the modeled area, thus contributing to the inaccuracy of amplification values received in the analysis. Moreover, it does not reflect the additional structures forming the subsurface of northern Israel, such as the Jordan Rift Valley (including the Hula basin and the Kinneret basin), also the Jezreel Valley extending from the Jordan Valley to the ZV. These structures would have an important effect on the partitioning of seismic energy during a strong motion seismic event and may cause amplification of seismic waves traveling from the source towards the ZV. There are several obstacles preventing the accurate modeling of the northern sedimentary structures. First, 3D modeling of subsurface structures in Israel is relatively new and still in development; Volk et al. (2017), created a 3D model of the subsurface of the Israeli coastal plain, considering structural scenarios that would cause amplification. The 3D modeling of the ZV area is a first of its kind in Israel. Second, additional obstacle for accurate modeling of sedimentary structures is data availability. The subsurface geology and material properties are not rigorously established and require more field work, geophysical and borehole data for sufficiently

70 accurate modeling. Third, computational limitations – the ZV simulations were confined to the size of a 104 x 85 x 19 km block and limited to a 0.75 Hz frequency because of computational memory restrictions. To expand the model size and frequency range more available memory must be acquired.

7 Conclusions

The present work studies the ground motion amplification atop the sedimentary structures underlying the Zevulun Valley: the Qishon Graben (QG), the Afek Horst (AH) and the Hilazon Graben (HG). 3D numerical modeling was performed to simulate earthquakes generated from two source locations: the Jordan Gorge Fault (JGF) and the Kinnarot source (KNR). Based on observations discussed in this work the following conclusions were made:

1. Ground motion amplification is highest atop the deep parts of a sedimentary basin: up to values of 10 atop the QG.

2. The amplification inside the QG is most probably the result of the combination of edge effect caused by the south-western boundary (Carmel Fault) and the material properties of the sedimentary column (impedance ratio).

3. The significant amplification of motion at the deepest parts of the QG cannot be explained solely by its 1D subsurface structure. While the top Judea and top Mavki’im reflectors may partially contribute to the amplification observed, the substantial amplification is attributed to the edge generated secondary Rayleigh waves, initiated due to the Carmel Fault confining the valley from south-west. In the HG the top Judea reflector, together with the top Saqiye reflector may be responsible for amplification ratios of up to 4.

4. In the AH the base Kurkar reflector is very shallow (few tens of meters) and therefore cannot be detected within the narrow range of frequencies modeled in this work (f  0.75)

5. The results of this study show clearly the attribute of the deep sedimentary structure to the ground motion amplification values, and the relevance of a strong reflector causing a sudden change in material properties between layers. Thus, challenging the current Israeli construction standards, relying only on the properties of the upper-most layer (for example Vs30) for calculations of amplification caused by seismic events.

6. The 3D model presented in this work is the first 3D numerical model of the Zevulun Valley subsurface structure. The inaccuracies of the model should be further studied to develop in more elaborate and accurate model.

7. Although the simulations in this work are within a limited frequency range (<0.75) and the most significant amplification at the deepest parts of the Zevulun Valley subsurface structure is within the low-frequency range, it is imperative to simulate higher frequencies in order to attain better understanding for various engineering structures that might be damaged due to diverse geological subsurface structures in a strong motion earthquake.

8. The Haifa bay area is a densely-populated area, with interlaced residential and industrial land uses. As the deep structure of the Zevulun Basin is clearly reflected in the ground motions the development and improvement of numerical simulation are essential for mitigation of risk in future seismic events.

9. In this research we have applied numerical modeling of deep subsurface structure based on available geological and geotechnical data, on a local scale. This method should be spatially expanded to a regional scale in order obtain a better comprehension of ground motions and amplifications expected in northern Israel, where subsurface sedimentary basins are of common occurrence. On the practical side, the results of the

72 numerical simulation must be tested against more recordings of seismic events with similar back-azimuth range to validate the results obtained in this work.

8 References

Agnon, A. (2014). Pre-instrumental earthquakes along the Dead Sea rift, in Dead Sea Transform Fault System: Reviews, Z. Garfunkel, Z. Ben-Avraham, and E. Kagan (Editors), Springer, The Netherlands, 207–261.

Aguirre, J., and Irikura, K. (1997). Nonlinearity, liquifaction, and velocity variation of soft soil layers in Port Island, Kobe, during the Hyogo-ken Nanby earthquake, Bulletin of the Seismological Society of America 87, 1244-1258.

Almagor, G., Hall, J. K. (1980). Morphology of the continental margin of northern Israel and southern Lebanon. Israel Journal of Earth-Sciences, 29, 245-252.

Almagor, G. (1993). Continental slope processes off northern Israel and southernmost Lebanon and their relation to onshore tectonics. Marine Geology, Volume 112, Issues 1–4, June 1993, Pages 151-169. Doi: 10.1016/0025-3227(93)90166-S

Bar Yossef, Y., Michaeli, A., Zah-Dvori, N., and Wolman, S. (2003). Hydrological model (prediction) for testing producing scenarios in Naaman and Kurdani Aquifers, Natural Resorces Department LTD, Report NR/389/03.

Bonilla, L.F., Steidl, J.H., Lindley, G.T., Tumarkin, A.G., Archuleta, R.J. (1997). Site amplification in the San Fernando Valley, California: Variability of site-effect estimation using the S-wave, coda, and H/V methods. Bulletin of the Seismological Society of America; 87 (3): 710–730.

Bonilla, L.F, LIU, P.C, NIELSEN, S. (2006). 1D and 2D linear and nonlinear site response in the Grenoble area. Third International Symposium on the Effects of Surface Geology

73 on Seismic Motion Grenoble, France, 30 August - 1 September 2006 Paper Number: 082/S02.

Boore, D.M. (2004). Can site response be predicted? Journal of Earthquake Engineering, 8, 1-41.

Davis, P.M., Rubinstein, J.L., Liu, K.H., Gao, S.S. & Knopoff, L., (2000). Northridge earthquake damage caused by geologic focusing of seismic waves, Science, 289, 1746– 1750.

Ellenblum, R., Marco, S., Kool, R., Davidovitch, U., Porat, R., Agnon, A. (2015), Archaeological record of earthquake ruptures in Tell Ateret, the Dead Sea Fault, Tectonics, 34, 2105–2117, doi:10.1002/2014TC003815.

Ellenblum, R., Marco, S., Agnon, A., Rockwell, T., X. Boas (1998), Crusader castle torn apart by earthquake at dawn, 20 May 1202, Geology, 26, 303–306.

Eytam, Y., Ben-Avraham, Z. (1992). The morphology and sediments of the inner shelf off northern Israel, Israel Journal of Earth-Sciences, 41, 27-44.

Fleischer, L., Gafsou, R., (2003). Top Judea Group – digital structural map of Israel. The Geophysical Institute of Israel, Report 753/312/03.

Gitterman, Y., Pinsky, V., Shapira, A., Ergin, M., & Kalafat, D., Gurbuz, C., Solomi, K. (2002). Improvement in Detection, Location, And Identification Of Small Events Through Joint Data Analysis By Seismic Stations In The Middle East/eastern Mediterranean Region.

Graves,R. W., Pitarka A., Somerville P. G. (1998). Ground-motion amplification in the Santa Monica area: Effects of shallow basin-edge structure, Bulletin of the Seismological Society of America. 88. 1224-1242.

Gvirtzman, Z., Don, A. (2005). Ground motion amplification during an earthquake

74 in the Zevulun Valley and western Galilee. Geological Survey of Israel, Report No. GSI/15/05.

Gvirtzman, Z., Zaslavsly, Y. (2009). Map of areas suspected of ground motion amplification. GSI/15/2009.

Gvirtzman, Z., Louie, J.N. (2010). 2D analysis of earthquake ground motion in Haifa Bay, Israel.

Bulletin of the Seismological Society of America. DOI: 10.1785/0120090019.

Gvirtzman, Z., Makovsky, Y., Sagy, Y. (2011). Reprocessing and geological interpretation of old seismic lines in the Haifa Bay. Geological Survey of Israel, Report GSI/27/2011.

Hamiel, Y., Amit, R., Begin, B.Z., Marco, S., Katz, O., Salamon, A., Ezra, Z., Porat, N. (2009). The Seismicity along the Dead Sea Fault during the Last 60,000 Years. Bulletin of The Seismological Society of America - Bulletin of the Seismological Society of America. 99. 2020-2026. 10.1785/0120080218.

Hamiel, Y., Piatibratova, O., Mizrahi, Y. (2016). Creep along the northern Jordan Valley section of the Dead Sea Fault. Geophysical Research Letters, 43, 2494–2501, doi:10.1002/2016GL067913.

Hanks, T.C., and Kanamori, H. (1979). A moment magnitude scale. Journal of Geophysical Research, 84, 5, 2348 – 2350, 9B0059, doi:10.1029/JB084Ib05p02348.

Hofstetter, A., Eck, T., Shapira, A. (1996), Seismic activity along the fault branches of the Dead Sea-Jordan transform system: The Carmel-Tirza fault system, Tectonophysics, 267, 317-330.

Holmes, W.T., Somers, P. (1996). Northridge earthquake of January 17, 1994, Reconnaissance Report, Vol. 2, Earthquake Spectra, Supplement C to vol 11.

Kafri, U., Ecker, A. (1964). Neogene and Quaternary subsurface geology and hydrogeology of the Zevulun Plain. Geological Survey of Israel, Bulletin No. 37.

Kawase, Hiroshi. (1996). The Cause of the Damage Belt in Kobe: "The Basin-Edge Effect," Constructive Interference of the Direct S-Wave with the Basin-Induced Diffracted/Rayleigh Waves. Seismological Research Letters. 67. 25-34. 10.1785/gssrl.67.5.25.

Kramer, S.L., (1996), Geotechnical Earthquake Engineering, Prentice Hall, New Jersey, 653 pp.

Kuo, Chun-Hsiang & L. Wen, K & M. Lin, C & Wen, Strong & Huang, Jyunyan. (2015). Investigating Near Surface S-Wave Velocity Properties Using Ambient Noise in Southwestern Taiwan. Terrestrial Atmospheric and Oceanic Sciences. 26. 205-211. 10.3319/TAO.2014.12.02.05(EOSI).

Lermo, J., Chávez-García, F.J. (1993). Site effect evaluation using spectral ratios with only one station. Bulletin of the Seismological Society of America; 83 (5): 1574–1594.

Madariaga, R. (2007). Seismic source theory, in Treatise on Geophysics, G. Schubert (Editor), Vol.4, H. Kanamori (Series Editor), Elsevier Science.

Marco, S., Stein, M., Agnon, A., Ron, H. (1996). Long‐term earthquake clustering: A 50,000‐year paleoseismic record in the Dead Sea Graben. Journal of Geophysical Research: Solid Earth/ Volume 101, Issue B3.

Maruyama, T. (1963). On the force equivalents of dynamical elastic dislocations with reference to the earthquake mechanism, Bulletin of Earthquake Engineering. 41, 467– 486.

Mero, D., (1983). Subsurface geology of Western Galilee and Zevulun Plain. TAHAL, TelAviv, Internal Report No. 04/83/48: 36 pp.

Milne, J. (1898). Seismology, 2014 Ed., Cambridge University Press, Cambridge, United Kingdom.

Moczo, P., Kristek, J., Bard, PY. et al. Bull Earthquake Eng (2018) 16: 2421. https://doi.org/10.1007/s10518-018-0345-5

Nakamura, Yutaka. (1989). A method for dynamic characteristics estimation of subsurface using microtremor on the ground surface. Quarterly Report of RTRI (Railway Technical Research Institute) (Japan). 30.

Nakano. H. (1923). Notes on the Nature of the Forces which give Rise to the Earthquake Motion. Seismol. Bull. Cent. Met. Obs. Tokyo. 1. 92-120.

Narayan, J.P., (2015). 3D basin-shape ratio effects on frequency content and spectral amplitudes of basin-generated surface waves and associated spatial ground motion ampliﬁcation and differential ground motion, Journal of Seismology., 19, 293–316.

Olsen, K.B., Pechmann, J.C., Schuster, G.T. (1996). An analysis of simulated and observed last records in the Salt Lake basin. Bulletin of the Seismological Society of America (1996) 86 (4): 1061-1076.

Petersson, N.A.; Sjogreen, B. (2017), SW4, version 2.0 [software], Computational Infrastructure of Geodynamics, doi: 10.5281/zenodo.1045297, url: https://doi.org/10.5281/zenodo.1045297

Pilz, M., Parolai, S., Petrovic, B., Silacheva, N., Abakanov, T., Orunbaev, S., Moldobekov, B. (2018). Basin-edge generated Rayleigh waves in the Almaty basin and corresponding consequences for ground motion amplification. Geophysical Journal International, 213, 301-316. doi: 10.1093/gii/ggx555.

Romo, M.P. and Seed, H.B. (1986). Analytical Modeling of Dynamic Soil Response in Mexico Earthquake of September 19th, 1985. Proceedings of International Conference on the 1985 Mexico Earthquake, Mexico City, 19-21 September 1986, 148-162.

Rong, M., Fu L•Y, Wang Z., Li X., Carpenter NS., Woolery EW. Lyu Y. (2017). On the amplitude discrepancy of HVSR and site amplification from strongmotion obser vations. Bulletin of the Seismological Society of America, 107:2873–2884. https://doi.org/10.1785/0120170118

Sadeh, M., Hamiel, Y., Ziv, A., Bock, Y., Fang, P. (2011). Crustal Deformation along the Dead Sea Transform and Carmel Fault Inferred from 12 years of GPS Measurements in Israel. AGU Fall Meeting Abstracts. 117. 0844-. 10.1029/2012JB009241.

Sagy, A., Rosensaft, M., Bartov, Y., Sneh, A. (2017). Map of active faults for Israeli standart 413, Jerusalem: Geological Survey of Israel.

Sahakian, V., Baltay, A., Hanks, T., Buehler, J., Vernon, F., Abrahamson, N. (2018). Decomposing leftovers: Event, path, and site residuals for a small-magnitude Anza region GMPE. Bulletin of the Seismological Society of America, 108(5A): 2478-2492, doi: 10.1785/0120170376.

Salamon, A., Aksinenko, T., Zaslavsky, Y., Kalmanovich, M., Shvartsburg, A., Giller, V., Dan, H., Giller, D., Zviely, D., Ankori, E., Medvedev, B., Frieslander, U., Zilberman, E., Baruch, O. (2013). The location of the Carmel Fault between Yagur and Haifa Port as inferred from subsurface data. Report GSI/01/2013.

Seed, H.B., Romo, M.P., Sun, J., Jaime, A., Lysmer, J. (1987) Relationships between Soil Conditions and Earthquake ground motions in Mexico City in the earthquake of September 19, 1985. UCB/EERC-87/15, University of California, Berkeley, p 125.

Shani-Kadmiel, S., Tsesarsky, M., Louie, J.N., Gvirtzman, Z. (2014). Geometrical focusing as a mechanism for significant amplification ground motion in sedimentary basins: Analytical and numerical study. Bulletin of Earthquake Engineering. 12, No. 2, 607-625, doi: 10.1007/s10518-013-9526-4.

Shani-Kadmiel, S., Tsesarsky, M., Gvirtzman, Z. (2016) Distributed slip model for forward modeling strong earthquakes. Bulletin of the Seismological society of America, pp. 93-103.

Shani-Kadmiel, S., Volk, O., Gvirtzman, Z., Tsesarsky, M. (2018). Ground motion amplification atop the complex sedimentary basin of Haifa Bay (Israel). Bulletin of Earthquake Engineering.

Silva, W. (1988). Soil Response to Earthquake Ground Motion, Electric Power Research Institute, Palo Alto, California.

Singh, S.K., Lermo, J., Dominguez, T., Ordaz, M., Espinosa, J.M., Mena, E., and Quass, R. (1988). The Mexico earthquake of September 19, 1985 - a study of amplification of seismic waves in the valley of Mexico with respect to hill zone site, Earthquake Spectra 4, 653-673.

Somerville, P.G., and Saikia, C.K., Wald, D., Graves, R. (1996). Implications of the Northridge earthquake for strong ground motions from thrust faults. Bulleting of the Seismological Society of America, 86, S115-S125.

Takai, N., Shigefuji, M., Rajaure, S., Bijukchhen, S., Ichiyanagi, M., Dhital, M.R., Sasatani, T. (2016). Strong ground motion in the Kathmandu Valley during the 2015 Gorkha, Nepal, earthquake. Earth, Planets and Space, 68, doi:10.1186/s40623-016-0383-7.

Tucker, B.E., King, J.L., Hatzfeld, D., and Nersesov, I.L. (1984). Observations of hard-rock site effects. Bulletin of the Seismological Society of America, Vol. 74, No. 1, pp. 121-136

Volk, O., Shani-Kadmiel, S., Gvirtzman, Z., Tsesarsky, M. (2017). 3D Effects of Sedimentary Wedges and Subsurface Canyons: Ground‐Motion Amplification in the Israeli Coastal Plain. Bulletin of the Seismological Society of America, 107, doi:10.1785/0120160349.

Wei, S., Chen, M., Wang, X., Graves, R., Lindsey, E., Wang, T., Karakaş, C., Helmberger, D. (2017). The 2015 Gorkha (Nepal) earthquake sequence: I. Source modeling and deterministic 3D ground shaking. Tectonophysics. 722. 10.1016/j.tecto.2017.11.024.

Wells, D.L., Coppersmith, K.J. (1994). New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement. Bulletin of the Seismological Society of America. 84, 974–1002.

Yoshimoto, K., Takemura, S., (2014). A study on the predominant period of long-period ground motions in the Kanto Basin, Japan, Earth Planets Space, 66, 100, doi:10.1186/1880-5981-66-100.

Zaslavsky, Y. (2006). Empirical determinations of local site effect using ambient vibration measurements for the earthquake hazard and risk assessment to Qrayot- Haifa Bay areas. Geophysical Institute of Israel. Report No. 595/064/06.

Zaslavsky Y. (2007). Use of ambient vibration measurements for reconstruction of subsurface structure along three profiles in the Zevulun Plain. Geophysical Institute of Israel. Report No. 571/240/07

Zaslavsky, Y. (2008). Empirical determination of site effects for seismic hazard assessment in the Kishon graben area near the Carmel fault. Geophysical Institute of Israel. Report No. 510/389/08.

Zilberman, E., Ron, H., Sa’ar, R. (2011). Evaluating the potential seismic hazard of the Ahihud ridge fault system by paleomagnetic and morphological analyses of calcretes, s.l: GSI/15/2011.

תקציר

אגנים סדימנטריים הם מבנה גאולוגי תת קרקעי נפוץ באזור הצפוני של ישראל. מבנים סדימנטריים מאופיינים בחומר מילוי צעיר ורך יחסית, בעל מהירויות סיסמיות נמוכות אשר מגבירות את תנודת הקרקע בזמן רעידת אדמה.

עמק זבולון השוכן במפרץ חיפה הוא אזור עירוני מאוכלס ובו קרוב למיליון תושבים, כמו כן מתקני תעשיה מפותחים כגון נמל ימי, בתי זיקוק לנפט ומפעלים כימיים נוספים המכילים חומרים מסוכנים. מתחת לפני השטח, לעמק מבנה סדימנטרי מורכב של שני גרבנים והורסט. מסיבות אלו, האזור נחקר בצורה נרחבת למטרות הערכת הסיכונים הסיסמיים הטמונים בו, ובפרט נחקרת הגברת תנודות הקרקע הצפויה על גבי המבנים הסדימנטריים התת קרקעיים מהם הוא מורכב בעת רעש אדמה חזק.

בישראל, בשל פרישה מצומצמת של הרשת הסיסמית הארצית ושכיחות נמוכה של אירועים סיסמיים גדולים, המחקר האמפירי אינו מספק היות והוא דורש כמות גדולה של נתונים מאירועים שנרשמו בעבר. לפיכך, מודלים נומריים חיוניים לחקר האזורים הקריטיים העומדים בפני סיכונים סיסמיים עתידיים.

בעבודה זו, מודל נומרי תלת ממדי נבנה בשיטת ההפרשים הסופיים בעזרת תוכנת הSW4, על מנת לדמות את התקדמות הגלים הסיסמיים במבנה התת קרקעי המורכב של עמק זבולון במהלך רעידת

אדמה חזקה )Mw = 6(, המגיעה משני מקורות שונים לאורך טרנספורם ים המלח. מפות יחסי ההגברה בין מודל עם אגן למודל ללא אגן מציגות יחסי הגברה מקסימליים של 12.7 עבור המקורות הנקודתיים, יחסי הגברה של 11 עבור המקור הדרומי ו10 עבור המקור הצפוני עבור מודל ההחלקה המתפשט )DSM(, מה שמוביל להבנה שיחסי הגברה נותרים דומים עבור מקורות שונים ומהירויות קרקע שונות. ערכי ההגברה הגבוהים ביותר אותרו ישירות מעל האזורים העמוקים ביותר בגרבן קישון. בגרבן חילזון נמצאו ערכי הגברה של כ4 ולא הובחנה הגברה מעל ההורסט המרכזי.

הניתוח הספקטרלי, המוגבל בעבודה זו ל- 0.75 הרץ, בוצע בשיטת H/H ומצא שני גבולות גאולוגיים המשמשים כרפלקטורים סיסמיים חזקים בעמק; בשני הגרבנים, בתדרים הנמוכים )> 0.3 הרץ( מוגברים הגלים בשל נוכחותו של גג חבורת יהודה, עם יחסי הגברה ספקטרליים של כ9 בתחנה הסמוכה לאזור העמוק של גרבן קישון ו1.5- בתחנה הממוקמת בשוליו. בנוסף, נמצא יחס הגברה של כ- 3.5 בגרבן חילזון. תדר נוסף בגרבן קישון בו אותרה הגברה הוא כ0.45 הרץ והוא מיוחס לגג תצורת מבקיעים/פטיש המגבירות ביחס של כ3. בהורסט המרכזי לא אותרה הגברה ספקטרלית בשל העומק הרדוד של גג יהודה. ניתוח התוצאות בשיטת H/V ניפק יחסי הגברה ספקטרליים שאינם עולים בקנה אחד עם שתי השיטות האחרות, אלא ערכים גבוהים הרבה יותר ברוב המקרים.

בעבודה זו הוצגה מפת ערכי הגברה בכל עמק זבולון, שקוראת תיגר על הערכים שהועלו בעבודה קודמת בהתבסס על מדידות רעשי רקע שעובדו בעזרת מודלים חד ממדיים.

תנודות קרקע בעמק זבולון )מפרץ חיפה( השפעת המבנה העמוק

חיבור לשם קבלת התואר "מגיסטר" בפקולטה למדעי הטבע

מאת אלינה גולדברג 317518090

תאריך עברי: טבת ה'תש"פ תאריך לועזי: ינואר 2020

חיבור זה מהווה חלק מהדרישות לקבלת התואר "מוסמך למדעי הטבע" )M.Sc(

תנודות קרקע בעמק זבולון )מפרץ חיפה( השפעת המבנה העמוק

מאת אלינה גולדברג 317518090

שמות המנחים: פרופ' מיכאל צסרסקי, פרופ' זוהר גבירצמן

חתימת המחבר ______תאריך 14.01.2020

אישור המנחה פרופ' מיכאל צסרסקי ______תאריך 14.01.2020

אישור המנחה פרופ' זוהר גבירצמן ______תאריך 14.01.2020

אישור יושב ראש ועדת מוסמכים מחלקתית ______תאריך ______