Bull Earthquake Eng DOI 10.1007/s10518-017-0096-8 ORIGINAL RESEARCH PAPER
The 2014 Earthquake Model of the Middle East: seismogenic sources Laurentiu Danciu1 • Karin S¸ es¸ etyan2 • Mine Demircioglu2 • Levent Gu¨len3 • Mehdi Zare4 • Roberto Basili5 • Ata Elias6 • Shota Adamia7 • Nino Tsereteli7 • Hilal Yalc¸ın3 • Murat Utkucu3 • Muhammad Asif Khan8 • Mohammad Sayab9 • Khaled Hessami4 • Andrea N. Rovida10 • Massimiliano Stucchi11 • Jean-Pierre Burg1 • Arkady Karakhanian12 • Hektor Babayan13 • Mher Avanesyan12 • Tahir Mammadli14 • Mahmood Al-Qaryouti15 • Dog˘an Kalafat2 • Otar Varazanashvili7 • Mustafa Erdik2 • Domenico Giardini1 Received: 16 December 2015 / Accepted: 23 January 2017 Springer Science+Business Media Dordrecht 2017
Abstract The Earthquake Model of Middle East (EMME) project was carried out between 2010 and 2014 to provide a harmonized seismic hazard assessment without country border limitations. The result covers eleven countries: Afghanistan, Armenia, Azerbaijan, Cyprus,
Data and Resources: All datasets collected, compiled, produced and used within the EMME project are available online, open to access at the site of European Facilities for Earthquake Hazard and Risk (http:// www.efehr.org). Additional information about EMME project is available at http://www.emme-gem.org. Electronic supplementary material The online version of this article (doi:10.1007/s10518-017-0096-8) contains supplementary material, which is available to authorized users. & Laurentiu Danciu
[email protected] Karin S¸es¸ etyan
[email protected] Mine Demircioglu
[email protected] Levent Gu¨len
[email protected] Mehdi Zare
[email protected] Roberto Basili
[email protected] Ata Elias
[email protected] Shota Adamia
[email protected] Nino Tsereteli
[email protected]
123
Bull Earthquake Eng
Georgia, Iran, Jordan, Lebanon, Pakistan, Syria and Turkey, which span one of the seismically most active regions on Earth in response to complex interactions between four major tectonic plates i.e. Africa, Arabia, India and Eurasia. Destructive earthquakes with great loss of life and property are frequent within this region, as exemplified by the recent events of Izmit (Turkey, 1999), Bam (Iran, 2003), Kashmir (Pakistan, 2005), Van (Turkey, 2011), and Hindu Kush (Afghanistan, 2015). We summarize multidisciplinary data (seismicity, geology, and tectonics) compiled and used to characterize the spatial and temporal distribution of earthquakes over the investigated region. We describe the development process of the model including the delineation of seismogenic sources and the description of methods and parameters of earthquake recurrence models, all representing the current state of knowledge and practice in seismic hazard assessment. The resulting seismogenic source model includes seismic sources defined by geological evidence and active tectonic findings correlated with measured seismicity patterns. A total of 234 area sources fully cross-border-harmonized are combined with 778 seismically active faults along with background-smoothed seismicity. Recorded seismicity (both historical and instrumental) provides the input to estimate rates of earthquakes for area sources and background seismicity while geologic slip-rates are used to characterize fault-specific earthquake recurrences. Ultimately, alternative models of intrinsic uncertainties of data, procedures and models are considered when used for calculation of the seismic hazard. At variance to previous models of the EMME region, we provide a homogeneous seismic source model representing a consistent basis for the next generation of seismic hazard models within the region.
Hilal Yalc¸ın
[email protected] Murat Utkucu
[email protected] Muhammad Asif Khan
[email protected] Mohammad Sayab
[email protected] Khaled Hessami
[email protected] Andrea N. Rovida
[email protected] Massimiliano Stucchi
[email protected] Jean-Pierre Burg
[email protected] Arkady Karakhanian
[email protected] Hektor Babayan
[email protected] Mher Avanesyan
[email protected] Tahir Mammadli
[email protected]
123
Bull Earthquake Eng
Keywords Earthquakes Seismogenic sources Seismic source models Active faults Probabilistic seismic hazard assessment Earthquake Model of the Middle East—EMME
1 Introduction Probabilistic seismic hazard analysis (PSHA) as developed over decades relies on identified active faults, seismotectonic interpretations and earthquake catalogues to estimate the probability of seismic ground shaking of a given level (Petersen et al. 2015). The Earthquake Model for Middle East (EMME, Erdik et al. 2012) project aimed at developing a homogeneous seismic hazard model across the Middle East and Caucasus Regions. The covered area is one of the seismically most active regions on Earth, as the interaction of four tectonic plates (Africa, Arabia, India and Eurasia) frequently causes earthquakes, many of large magnitude (Allen et al. 2004). Within the EMME project, a probabilistic framework was adopted to describe the seismogenic potential of a region covering Afghanistan, Armenia, Azerbaijan, Cyprus, Georgia, Iran, Jordan, Lebanon, Pakistan, Syria and Turkey. Generally, the seismogenic potential of a region requires collecting evidence of seismicity, developing homogeneous earthquake catalogues, compiling geological information on the location of seismogenic fault zones, evaluating the mechanics of faulting and quantifying the state of stress in the Earth’s crust. This multidisciplinary process results in a forecast of seismicity specifying
Mahmood Al-Qaryouti
[email protected] Dog˘an Kalafat
[email protected] Otar Varazanashvili
[email protected] Mustafa Erdik
[email protected] Domenico Giardini
[email protected] 1
Department of Earth Science, ETH Zurich, Sonneggstrasse 5, 8092 Zurich, Switzerland
2
Kandilli Observatory and Earthquake Research Institute, Bog˘azic¸i University, Istanbul, Turkey
3
Department of Geophysical Engineering, Sakarya University, Serdivan, Sakarya, Turkey
4
International Institute of Earthquake Engineering and Seismology, Tehran, Iran
5
Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Roma 1, Rome, Italy
6
Faculte´ des Sciences, Universite´ Saint Joseph, Beirut, Lebanon
7
M. Nodia Institute of Geophysics, I. Javakhishvili Tbilisi State University, Tbilisi, Georgia
8
Karakoram International University, Gilgit, Pakistan
9
Geological Survey of Finland, Espoo, Finland
10
Istituto Nazionale di Geofisica e Vulcanologia, Milan, Italy
123
Bull Earthquake Eng
location, size and rate of future earthquakes for use with ground motion models in assessing the seismic hazard. From a practical point of view, seismogenic source models can vary in complexity, given the level of investigation, available data and resources and goals of seismic hazard analysis. Site-specific hazard analyses at critical infrastructure locations (e.g. dams, nuclear power plants, bridges) are focused on developing detailed data, investigations and interpretation. Indeed, a considerable level of investigation and judgment is required as sitespecific hazard analyses proceed from seismogenic sources to geological complexity because critical infrastructures are preferably built in regions of low seismic activity (PEGASOS-Project, Renault 2014). Regional seismic hazard analyses that encompass several countries pose challenges of data collection and harmonization, information sharing, constraints of political boundaries and technical limitations. Such models including large geographical domains without political boundary constraints were first established within the Global Seismic Hazard Assessment Program - GSHAP (Giardini 1999). Within GSHAP, the most common earthquake source representation was the seismic area source, frequently delineating the spatial distribution of earthquake epicentres (Cornell 1968). Nowadays, the current practice on developing seismogenic source models includes multiple seismic sources due to uncertainties of diverse data and limits. Such multiple seismic sources can delineate active faults, area sources or gridded seismicity for regions without identified active faults. Seismogenic sources are often combined for modeling the spatial and temporal distribution of earthquakes (Stirling et al. 2012; Moschetti et al. 2015; Adams et al. 2015). Recently, the 2013 European Seismic Hazard Model (hereafter ESHM13, Woessner et al. 2015) combined three seismic source models, e.g. (1) area source, (2) active faults and background area sources and (3) smoothed seismicity. ESHM13 included for the first time active faults over the Euro-Mediterranean region. The seismogenic source model of the Middle East region, hereafter EMME-SSM14, was developed from three major datasets and information regarding tectonics, seismicity and faulting characteristics of the region. In some places, e.g. North Anatolian Fault System (Turkey), Zagros (Iran), Dead Sea Fault (Lebanon), the seismicity patterns and the tectonic features are evident and a single seismogenic source model appears suitable. In other places, characterized by incomplete seismic data or lack of information, it is not possible to specify a single model with certainty. Thus, the final EMME-SSM14 combines two alternative earthquake source models in a simple logic tree structure to capture the implicit epistemic uncertainties of data and methods. The first seismogenic source model consists in area sources portraying the patterns of historical seismicity correlated with tectonic characteristics and identified geological structures. The second seismogenic model comprises active faults and off-faults background seismicity without the subjectivity of area source delineation. To avoid double counting of seismicity in fault proximity, the low-magnitude earthquakes are separated
11
European Centre for Training and Research in Earthquake Engineering (EUCENTRE), Pavia, Italy
12
Institute of Geological Sciences, Armenian Academy of Sciences, Yerevan, Republic of Armenia
13
Georisk Scientific Research Company, Yerevan, Republic of Armenia
14
Republican Seismic Survey Center of Azerbaijan National Academy of Sciences, Baku, Azerbaijan
15
Jordan Seismological Observatory, Ministry of Energy and Mineral Resources, Amman, Jordan
123
Bull Earthquake Eng
from the larger magnitude earthquakes associated with faults. All catalogue data, seismicity, active faults and resulting seismogenic sources are harmonized, so that they do not contain any limitations and/or inconsistencies across political boundaries. The earthquake catalogue is homogeneous for moment magnitude (Mw). Seismic activity rates of gridded sources (area and background sources) are based on a statistical analysis of the regional earthquake catalogue, both historical and instrumental. Characterization of fault sources relies upon conversion of geological slip-rates into faultspecific earthquake productivity. Moreover, the seismic activity rates of subduction interfaces are estimated from historical reports of past earthquakes and from converting the plate-boundary slip-rates (Reilinger et al. 2006). Hereafter, we summarize the development process of EMME-SSM14 as prepared and used for the earthquake hazard evaluation of the Middle East. Two companion papers describe the ground motion models (Danciu el al. 2016a) and the outcome of the seismic hazard estimates of the EMME region (S¸ es¸ etyan et al. 2017, this issue). It is beyond the scope of this contribution to describe all aspects and issues, e.g. compilation of the datasets, harmonization of earthquake magnitudes, style-of-faulting assessment, and validation of the datasets. However, we attempt describing most of the details that are essential for understanding the EMME-SSM14.
2 Regional datasets overview: tectonics, geology and seismicity Tectonics, active faults and seismicity represent the input data to start developing a seismogenic source model. Nonetheless, the relationship among these data sets is not straightforward. Some shortcomings are due to relatively short historical seismicity record, insufficient knowledge of blind/buried active faults and uncertain correlation between seismicity patterns and exposed faults. Compilation of regional datasets helps understanding the regional seismogenic profile and provides the support for developing a seismogenic source model. Several working groups were established and the summary presented below represents the work of local researchers and various international experts who contributed to certain parts and/or to the final seismogenic source model for the Middle East.
2.1 Tectonics The Middle East region, as defined within the EMME project (Fig. 1) spans a large geographical region (about 3 million km2) from the Mediterranean Sea to the Hindu Kush (Pakistan), from west to east, and from the Caucasus to the Arabian Peninsula, from north to south. On the tectonic point of view, the region is dominated by the interaction between and among the African (Nubia), Arabian, and Indian plates with the Eurasian tectonic plate. This complex area of continental lithosphere is one of the largest regions of convergent deformation on Earth (Allen et al. 2004). It is tectonically controlled by major plate-tectonic structures: Hellenic and Cyprian subduction zones, Anatolian block, Northern Anatolian Fault System, Eastern Anatolian Fault System, Dead Sea Fault, BitlisZagros Fold and Thrust Belt, Greater and Lesser Caucasus Thrust, Kopeth Dagh Fault, Herat Fault, Chaman Fault, Makran subduction zone, and Himalaya Main Boundary Thrust, all shortly described hereafter. The continental Anatolian Block is bounded by the North Anatolian and East Anatolian Faults. GPS measurements determined counter-clockwise rotation and westward escape towards the north-dipping Hellenic and Cyprian subduction zones in the eastern
123
Bull Earthquake Eng
Fig. 1 Sketch map of main faults (dashed lines) of EMME region: North Anatolian Fault (NAF) System, East Anatolian Fault (EAF) System, Dead Sea Fault (DSF), Bitlis-Zagros Fold and Thrust Belt (BZFTB), Main Caucasus Thrust (MCT), Lesser Caucasus Thrust (LCT), Kopeth Dagh Fault (KDF), Chaman Fault (CF), Herat Fault (HF), Pamir-Hindu Kush (PHK), Himalaya Main Boundary Trust (HMBT). The major tectonic plate boundaries are mapped in red, whereas the white arrows indicate direction of relative convergence of the four tectonic plates: Africa, Arabia, India and Eurasia (Reilinger et al. 2006)
Mediterranean Sea (McClusky et al. 2000; (Reilinger et al. 2006). Large earthquakes (Mw [ 8.0) are likely to occur in these zones as revealed by historical testimony (e.g. the 365 CE earthquake of Crete, Greece, followed by a tsunami across the entire southern Mediterranean Sea; Guidoboni et al. 1994). The North Anatolian Fault (NAF) system is continuous over 1500 km from east Turkey to the Aegean Sea. NAF delineates the northern boundary of the Anatolian Block and consists in numerous strike-slip fault segments characterized by an average right-lateral slip rate of *24 mm yr-1 (McClusky et al. 2000; Reilinger et al. 2006). Accordingly, the high displacement rate converts into high seismicity rate as more than 25 earthquakes of Mw [ 6.5 occurred during the last century along the entire length of NAF. The 1999 Izmit (Turkey) earthquake, Mw 7.4, is one of the recent examples (Gu¨len et al. 2002). The *580-km-long left-lateral strike-slip Eastern Anatolian Fault (EAF) is the southeastern boundary of the Anatolian block. The active displacement rate of EAF is about 9 mm yr-1 (McClusky et al. 2000; Reilinger et al. 2006). Large earthquakes occurred along this fault system, however less frequently than along the NAF; notably the 1893, Mw 7.1 Malatya (Turkey), the 1971, Mw 6.9 and 2003, Mw 6.4, Bingol (Turkey) earthquakes, and more recently the 2010, Mw 6.1, Elazıg˘ earthquake (Turkey). The Dead Sea Fault (DSF) system is the 1000 km long tectonic boundary between the African and Arabian plates (Fig. 1). DSF includes left-lateral fault segments subdivided in three structural sections; the southern and northern segments are separated by the restrained bend in Lebanon. A left-lateral slip-rate of 2–8 mm yr-1 is estimated within the DSF (Allen et al. 2004). Although seismically active, the DST is characterized by long inactive periods with sudden large magnitude earthquakes (e.g. 1995, Mw 7.3, Gulf of Aqaba). The DSF has one of the longest paleoseismic records, spanning over thousands of years (Marco et al. 1996). The northward motion of Arabia towards Eurasia produces continental collision along the about 1500 km long Bitlis-Zagros Fold and Thrust Belt (BZFTB) from eastern Turkey to southern Iran (Fig. 1). The effects of the on-going convergence extend from the BZFTB all the way north to the Caucasus and the Kopeth Dagh mountains. The 2011, Mw 7.2, Van
123
Bull Earthquake Eng
(Turkey) earthquake is the most recent, large magnitude earthquake that has shaken this collisional region. The Alborz Mountain Range (Iran), to the north, delimits the Iranian plateau, which includes prominent strike-slip faults. The Alborz Mountain Range stretches for 600 km along the southern shore of the Caspian Sea and connects northward with the Talesh Mountains of northwestern Iran and Azerbaijan. The shortening rate within the Alborz Mountain Range and the South Caspian Basin is about 10 mm yr-1, assuming that ca 8–9 mm yr-1 of the *22 mm yr-1 overall Arabia-Eurasia convergence are accommodated by the BZFTB (Sella et al. 2002). Hence, due to its narrowness, the Alborz Mountain Range might deform more than the BZFTB (Tatar et al. 2002; Allen et al. 2004). Moderate to large magnitude earthquakes are frequent in Alborz Mountain Range, e.g. 1957, Mw 6.8 Mazandaran (Iran); 1962, Mw 7.1 Booin-Zahra (Iran); 1990, Mw 7.3 Rudbar (Iran); 2004, Mw 6.2 Baladeh (Iran) are among the most destructive events known in this region. Two large magnitude earthquakes (M [ 7.0) in 1721 and 1780 are documented for the Tabriz surroundings (Iran). The 2012 moderate earthquakes of Mw 6.3 and Mw 6.4 occurred northeast of Tabriz, causing extensive damages in the rural areas of the East Azerbaijan Province, Iran (Copley et al. 2014). In the north, the Caucasus Mountain Range is subdivided into the Greater (GC) and the Lesser Caucasus (LC, Fig. 1). The former is surrounded by north- and south-dipping thrust faults, whereas strike-slip faults predominate in the latter. Greater Caucasus continues eastward across the Caspian Sea to the Kopeth Dagh faults on the border between Turkmenistan and Iran (Jackson et al. 2002). Destructive earthquakes occurred in the Greater Caucasus region during the last century, e.g. the 1991, Mw 7.0, Racha earthquake (Georgia), the largest event ever recorded there. In the Lesser Caucasus, damaging earthquakes are also frequent: for example the 1988, Mw 6.9, Spitak (Armenia) and the 1976, Mw 7.3, C¸aldiran (Turkey) earthquakes. The Kopeth Dagh Fault (KDF) system extends for about 700 km from the Caspian Sea to the Afghanistan border through northeast Iran, following the north side of the Kopeth Dagh Mountains. The KDF includes both right-lateral strike-slip faults and reverse faults. Recent GPS measurements did not identify rapid displacement albeit frequent earthquakes of high magnitude (Yeats 2012). One of such earthquakes was the 1948, Mw 7.2, event that severely affected Ashgabat (Turkmenistan, Berberian and Yeats 2001). Farther south, convergence between the Arabian and Eurasian plates is responsible for the Zagros Fold and Thrust Belt (ZFTB) and the Makran oceanic subduction, offshore south-eastern Iran and southwestern Pakistan. The Makran subduction zone displays two seismicity patterns: an apparently aseismic or currently locked western section and a much more active eastern section, where the 1945, Mw 8.1, Belochistan earthquake (Pakistan) generated a tsunami within the Gulf of Oman and Arabian Sea (Yeats 2012). Based on GPS measurement, the convergence rate across the Makran subduction zone is 32–35 mm yr-1 (Reilinger et al. 2006). Central Iran is a continental block delimited by the Alborz Mountains and KDF to the north and by BZFTB and Makran subduction zone to the southwest. The region embeds two zones of north–south and north-northwest striking active faults bounding the Lut Block (Iran). The Helmand (or Afghan) Block is the next stable continental block extending eastward to the left-lateral, 850 km long transform Chaman fault (CF) in Afghanistan (Yeats 2012). The north–south CF intersects the east–west Herat fault system (Afghanistan), which separates the stable Turan platform (Turkmenistan) to the north from the Helmand block, to the south.
123
Bull Earthquake Eng
Pamir and Hindu-Kush (NW Pakistan and NE Afghanistan) mountain range systems are an expression of the India-Eurasia collision. The western syntaxis of the Himalayas is characterized by a high concentration of intermediate depth seismicity (*70–300 km) confined in a north-dipping western zone underneath Hindu Kush Mountains and a south-dipping eastern zone beneath the Pamir Mountains. Shallow earthquakes also occur in the Pamir Mountains as both strike-slip and thrust mechanisms. The 1911, Mw 7.4, Sarez earthquake (Rushon District of eastern Tajikistan) ruptured in the Central Pamir Mountains, triggering a landside that blocked the Murghab Valley, so creating the largest known natural dam, the Usoi Dam (Schuster and Alford 2004). Yet, deep seismicity dominates the Pamir and Hindu-Kush region, with an average of four earthquakes of M [ 5.0 per year. The 2002, Mw 7.4, Hindu-Kush earthquake at *226 km depth struck northern Afghanistan and the 2015, Mw 7.4, earthquake at *212 km depth severely affected the major cities of northern Pakistan, with casualties reported in Afghanistan and India. The described complex tectonic environment forms the basis for analysis of the seismogenic potential; however, a simplified tectonic regionalization was derived for assisting the selection of ground motion equations (GMPEs). Five seismo-tectonic regions were defined: active regions in shallow crust, stable regions, deep seismicity, subduction interface and inslab. The simplified tectonic regionalization to pair seismogenic sources and ground motion characteristic models is presented in Fig. 1 of Danciu et al. (2016a).
2.2 Geological faults Local researchers and experts have placed great effort to compile the first datasets of active faults for the Middle East. Main contributors to this task are listed in the companion paper (S¸ es¸ etyan et al. 2017, this issue). The geological fault source database (EMME_fsdb) was compiled and revised in sequential steps between 2010 and 2012. Several workshops and meetings were organized to define data format and harmonization procedures. Compilation of the active faults started in Turkey as a contribution to the European Database of Seismogenic Faults (EDSF; Basili et al. 2013a). Progressively, the active fault data extended to other countries with contributions from Iran, Georgia, Lebanon, Jordan and Pakistan. Seismogenic faults of Afghanistan were retrieved from Ruleman et al. (2007). EMME_fsdb extends the database format of Wills et al. (2008) and contains information of fault sections, segments and subduction zones. A fault section as defined in EMME_sfdb, includes fault parameter information about earthquake causative faults, without reference to earthquake recurrence, whereas a fault segment is associated with the occurrence of earthquakes limited by segment tips. Thus, we use fault segment with earthquake recurrence models while defining a seismically active fault (Gu¨len et al. 2014). Fault segments may have more than one fault section and an automatic technique implementing geological rules of fault association was applied to obtain seismogenic faults. Each record of the EMME_fsdb is represented by [ 30 parameters, some of the most important being fault section name, fault section identity, fault trace (pairs of latitudes and longitudes, and depth of the upper fault edge), fault type (i.e. thrust, reverse, strike-slip, extensional), average strike, dip, and rake values, fault section length, lower and upper seismogenic depth, horizontal and vertical slip rates and average aseismic slip factor. The slip rate information, as provided by regional compilers, includes values inferred from GPS measurements, geodetic observations across the region and geological estimations. The
123
Bull Earthquake Eng
latter represent the ratio between the total displacements of dated geological structures. Given the age of the used information, ranging from 10,000 to 150,000 years, the slip-rates have to be considered as long-term slip rates. Whenever available, both the lower and upper bound values of slip rates are retained. The quality level of information on active faults varies from country to country. To account for this variability, we subdivided the EMME_fsdb in classes of confidence that reflect the accuracy and reliability of fault information: • Class A (well-constrained) includes entries of robust information of fault geometry and long-term slip rates; information supported by historical and instrumental observations and geological field investigations. Faults in such class cover Turkey, Armenia, Iran, and Caucasus. • Class B (satisfactory) contains records with partial or satisfactory information available; the entries are jointly constrained by historical ruptures and geological evidence in Central Iran and Eastern Mediterranean. • Class C (limited) comprises few faults with incomplete information in Afghanistan and Pakistan. These faults have historical evidence but limited/debatable geological information. Class C faults are approximately located along clearly identified largescale features (e.g. Chaman fault system). The resulting EMME_fsdb consists of about 3397 fault segments totalling about 91,551 km length. Subduction zones are also part of the EMME_fsdb, with data of the Hellenic and the Cyprian subduction zones obtained from ESHM13 (Giardini et al. 2013), whereas data on the Makran subduction zone are compiled within the EMME project and discussed in Sect. 4.4: Subduction Zones.
2.3 Seismicity A new earthquake catalogue was built within the project framework using historical (pre1900) and early and modern instrumental events until 2006. Zare et al. (2014) summarized the procedural steps to obtain a harmonized earthquake catalogue for the Middle East. In this primary catalogue, each entry has assigned information of origin time, longitude, latitude, magnitude, and hypocentral depth. A total of 30,218 earthquakes were compiled and harmonized to a uniform moment magnitude (Mw) scale. The magnitude range of the compiled catalogue is 4.0 \ Mw \ 8.1 with the largest magnitude event being the 1945 Makran event. There are 125 events between 7 \ Mw \ 8,444 in 6 \ Mw \ 7 range and 1897 between 5 and 6 Mw. This high rate of magnitudes characterizes the region as one of the seismically most active on Earth. The earthquake catalogue, complemented with parts of the ESHM13 catalogues (Stucchi et al. 2012; Gru¨nthal et al. 2012) is presented in Fig. 2 as a function of the hypocentral depth. Depth variability is large, with shallow seismicity in the range of 5–35 km with an average value of about 15 km for the entire region. Intermediate depth seismicity, i.e. 60–120 km, is recorded in Northern Caucasus, Zagros Fold and Thrust Belt, and the Hellenic and Cyprian subduction zones. A cluster of deep seismicity, deeper than 100 km and as deep as 300 km, is present in the Pamir-Hindu Kush region. Beneath Pamir, the intermediate depth seismicity is separated from shallow crustal activity by an apparently largely aseismic lower crust between 25 and 70 km. Bellow this aseismic layer, earthquakes occur down to 170 km in the eastern part and to 250 km in the western part (Fig. 2). Earthquakes in Hindu-Kush are confined to 180–220 km depth in a steeply southdipping cluster. These earthquakes seem to be separated into an upper and lower part by a
123
Bull Earthquake Eng
Fig. 2 Spatial distribution of the harmonized earthquake catalogue as a function of the assigned focal depth (km)
vertical gap at approximately 150 km depth. The upper part of the Hindu Kush seismicity is sparsely sampled while the lower part consists of a number of highly active clusters (Sippl et al. 2013). Two different declustering algorithms, Reasenberg (1985) and Gru¨nthal (1985), were used to identify and remove dependent events such as foreshocks, aftershocks and swarm events. Identification of dependent events is a prerequisite to ensure the Poisson assumption of earthquake recurrence (Knopoff 1964). Recurrence statistics estimated from an un-declustered catalogue are incorrect, since the occurrence rates of large magnitudes are affected by many small events that are not main shocks. Basically, an increased number of small events tends to increase the slope of the magnitude-frequency curve, which decreases the frequency of larger events and ultimately underestimates the hazard (e.g. Musson 1999). A sensitivity analysis revealed that the Reasenberg algorithm did not identify all the dependent events in the EMME catalogue. A visual inspection of the 1999 Kocaeli and 1995 Aqaba sequences shows that many of the aftershocks remained in the catalogue; Hence the algorithm of Gru¨nthal (1985) was preferred. The final sub-catalogue consisting of 7395 main events was kept to describe the seismicity of the region.
3 Catalogue completeness, maximum magnitudes and regional b-value estimations The concept of macrozones (large regions used to define the regional distribution of a hazard modelling parameter) was first introduced in ESHM13 where specific superzones were developed for catalogue completeness, maximum magnitude (Mmax) and regional seismicity parameters e.g. the b-value (Gutenberg and Richter 1944), as discussed later. All investigated parameters (e.g. the maximum magnitude, b-value and completeness intervals) are deemed to be homogenous over the delineated macrozones. Seismic sources overlying a macrozone get the parameters of that macrozone. Thus, a macrozone provides a simplification in large-scale seismic hazard modelling and applies only for parameters that may be modelled alike over a number of neighboring seismic sources. We have delineated 20 macrozones in the EMME region, taking into consideration possible trends in catalogue completeness and provided as Electronic Supplement to this manuscript (see Fig. S1). Unlike the ESHM13, where different superzones were used for
123
Bull Earthquake Eng
different parameters, the EMME macrozones are used to evaluate three key source parameters, namely catalogue completeness (considering regional history of earthquake record), maximum magnitude (considering the main regional tectonic features), which, in a combined way were also used for the computation of regional b-values. The use of regional b-values was twofold, once in connection with the fault source and second with the background seismicity model, as explained later in this paper. It should be noted that the macrozones are not considered as an input for hazard calculation.
3.1 Catalogue development and completeness Completeness time intervals of different magnitude ranges were estimated for each macrozone considering regional differences in history record of earthquakes both in early and recent (i.e. instrumental) periods. The historical earthquake catalogue is sensitive to availability and quality of reported evidence (e.g. Ambraseys et al. 2002), whilst the advance of seismic networks e.g. installation time and number of modern instruments affects the recent times. Completeness corresponding to each macrozone is evaluated using the Stepp (1972) procedure for recent periods (i.e. the last 30 years) and by historical considerations and expert judgement for earlier periods. Alternative branches of completeness were proposed, also based on historical considerations, but given the high demand of computational time in hazard calculation, only one set of completeness intervals was considered. The resulting completeness time intervals are listed in Table S1 of the Electronic Supplement. Completeness time intervals are further used when estimating the recurrence parameters of the seismogenic sources.
3.2 Assessing Mmax Mmax is an influential seismogenic source parameter that represents the upper limits of the earthquakes size, interpreted as the largest possible earthquake of an active fault within the considered volume of Earth’s crust. Generally, Mmax earthquakes have long recurrence intervals and the assigned Mmax affects the low annual probabilities, which are of particular importance for critical infrastructures such as bridges and nuclear power-plants (Mueller 2010). In tectonically active regions, where most fault systems (i.e. capable of hosting large magnitude earthquakes) are known, Mmax can be directly estimated from fault characteristics: total rupture area, total length, rupture length, co-seismic slip or seismic moment (Mignan et al. 2015). In regions with poorly identified seismogenic structures, assessing Mmax is particularly difficult because the most significant physical constraint to assess the fault dimension is essentially not known. As a result, assigning the Mmax to these regions relies either on the historical record and analogies to similar tectonic features (Wheeler 2009) or on the use of statistical ‘‘maximum possible’’ magnitude (Kijko and Singh 2011). For a seismically active region as considered in EMME project, albeit abundance of large earthquake ([100 earthquakes of Mw [ 7.0), the earthquake record spans over *1500 years only. This time interval is much smaller than the recurrence time of very large earthquakes e.g. 8.0 Mw. The use of active faults information (e.g. converting Mmax from fault dimension) may extend this duration to geological times, but the fault information is also either incomplete or uncertain. In this context, we based assessment of Mmax on different definitions of the maximum earthquake: (1) estimated from maximum observed (Mobs) in the historical seismicity record, (2) from analogue tectonic features and (3) from fault size conversion. Initially, the maximum historical earthquake (Mobs) is assigned to each macrozone by considering the location uncertainty of the historical
123
Bull Earthquake Eng
events, i.e. resolving possible floating earthquakes located close to the boundary between two zones. Furthermore, when active faults were known, we compared the Mmax inferred from fault dimensions with the Mmax from historical earthquakes in each area-source zone. Finally, the initial Mmax assigned to the macrozones was adjusted to be the largest value obtained. To account for epistemic uncertainties on Mmax in absence of better qualitative and quantitative data coverage, we extended the Mmax values with uniform magnitude increments of ?0.3 and ?0.6. These values correspond to the magnitude uncertainty reported in the regional earthquake catalogue (Zare et al. 2014). Following Wheeler (2009), the two added increments imply extension of the historical record by a factor of 1.35–1.7, assuming a regional b-value of 0.9–1.00. With the two factors, the historical record is extended to *4000 and *5000 years to provide conservatism against earthquakes larger than the measured Mmax. The spatial distribution of Mmax is displayed in Fig. 3a for area source model and fault sources (Fig. 3b). The largest Mmax values (8.0–8.8 Mw) are assigned to the Hellenic Subduction Zones and deep sources of Pamir-Hindu-Kush, whereas the Mmax values for Makran Subduction are from 7.8–8.6 Mw. The assigned Mmax for active shallow crust regions is between 5.8 (North Armenia) and 8.4 Mw (North Anatolian Fault). For stable continental regions the Mmax ranges from 6.8 to 7.4 Mw. The lowest values attributed to sources in North Armenia reflect the opinion of local experts (Karakhanyan A. 2013,
Fig. 3 (a) Spatial distribution of the Mmax assigned to the to area sources (grey polygons); blue polygons are the subduction interface, green polygons are the subduction inslab, light red polygons are deep seismicity zones. The uncertainties associated to these Mmax values are 0.3 and 0.6 Mw increment, as quoted in Sect. 3.2: Assessing Mmax. (b) Spatial distribution of the Mmax corresponding to the seismically active faults; These Mmax values were used for crosschecking the Mmax within each area source. No magnitude increment added to these values of Mmax when used for fault characterization
123
Bull Earthquake Eng
personal communication). For the Mediterranean subduction zones, Mmax values were inherited from ESHM13. Mmax is further used as upper bound of the earthquake recurrence of various magnitudes specific to each seismogenic source.
4 EMME-SSM14: elements and development process The initial steps in generating a seismogenic source model are the delineation of the seismogenic sources and the estimate of the size, rates and associated uncertainties of future earthquakes within the region. Specifically, we delineated two alternative earthquake source models organized in a logic tree structure to capture the modelling uncertainty. We discuss the main elements of EMME-SSM14 in the next sections, starting with the area sources, followed by active faults, gridded background seismicity and the subduction sources.
4.1 Seismic area sources Area sources represent regions of homogenous seismicity, often used for modelling earthquake patterns with or without tectonic evidence. In the investigated region, most of the previously proposed seismic hazard models (e.g. Erdik et al. 1999; Tavakoli and Ghafory-Ashtiany 1999) were built upon area sources mainly due to GSHAP legacy (Danciu and Giardini 2015). Within the first year of the project, country-specific seismogenic source models were collected. Next, the area source models were revisited and modified during several workshops to ensure cross-border harmonization by (1) removing duplicated area-sources, (the same area sources defined within countries), (2) simplifying unnecessary or artificial complex area-source boundaries (i.e. area-boundaries with inner spaces) and (3) correcting the shape of individual area sources e.g. removing duplicated vertexes or multiple segments and/or points. Although area sources are widely used, little is available on how to define such polygons (Meletti et al. 2008; Musson et al. 2009; Schmid and Slejko 2009; Wiemer et al. 2009; Vilanova et al. 2014). To this extent, criteria for guiding the delineation of area sources were defined for uniformity and transparency. We started with the criteria of Erdik et al. (1999) used for Turkey and neighboring regions in GSHAP (Giardini 1999). These criteria describe the correlation between the seismicity and major tectonic features, but no reference on how to integrate active faults is given. Thus, we updated these criteria to account not only for the seismicity patterns but also its correlation with the newly compiled fault dataset. The resulting criteria to guide delineation of area sources are: 1.
2.
Seismicity-based area sources are preferable only if data/info are insufficient to associate the seismicity with seismogenic crustal structures; Beware that some of the active faults have no surface expression and the historical evidence of earthquake occurrence might be the only information to delineate such area sources. Area source boundaries should be adjusted to follow the surface projection of identified faults; an area-source should not interrupt a fault system unless major differences are observed (changes in style-of-faulting/stress orientation, changes in crustal depth).
123
Bull Earthquake Eng
3.
4.
5.
6. 7. 8.
9.
10.
11.
12.
An area source should not span over two different tectonic units, e.g. active shallow crust and a stable continental region. Each area-source shall represent only a single geological unit or tectonic feature. Area source geometry should be consistent with the seismicity patterns of the regional earthquake catalogue, if they are delineated following the local earthquake catalogue only. Area sources delineating a pattern of high seismicity should be small, otherwise the high rate of seismicity is distributed over a larger zone and hence reduced; For example, the seismic hazard estimates may be different if the same seismicity is assigned to two sources of different size; The estimates will be reduced for the source of a larger size. In contrast, the smaller the source area for the same seismicity, the greater will be the resulting hazard estimates. This effect has been referred to as spatial smearing (NRC 1988). Area source geometry can be adjusted to include an earthquake, but a single earthquake cannot create an area source. A single earthquake can be moved inside an area source if the location is uncertain. Area source should be volumetric, accounting for earthquake depth variation. Hypocentre depth distribution should be obtained from interpretation of the earthquake catalogue. Area-sources can overlap only on the volume domain; The crustal area sources can overlap the subduction inslab sources, or deep seismicity sources; however, the earthquake size and earthquake occurrence rates has to be carefully evaluated to avoid double-counting; A minimal number of earthquakes has to be present within each area-source; areasources with no known significant seismicity or known tectonic/geologic features should be treated as background area-source; A minimum activity rate should be assigned because available data are in general too poor to exclude earthquake occurrence with confidence. The lower seismogenic depth (maximum down-dip crust extend) should be evaluated from crust thickness, Moho-discontinuity location and/or lower depths of seismicity. Focal mechanisms on major faults should be associated to individual area sources. In absence of identified faults, analyses of the earthquake catalogue shall be performed to identify the predominant style of faulting.
To finalize the area source model, we overlapped multiple datasets (e.g. active faults and earthquake catalogue) and we applied adjustments following the above-mentioned criteria. Further, we coordinated our effort with the two sibling regional initiatives to obtain an inter-regional harmonization. The area sources of ESHM13 spanning the western Greece-Turkey border were integrated with no modification. Similarly, on the northeastern part of the region, the area sources of Earthquake Model of Central Asia (Ullah et al. 2015), hereafter EMCA15, were also considered. Area sources as inherited from ESHM13 and EMCA15 depict complete information of geometry, style-of-faulting, upper magnitude and earthquake recurrence parameters, thus no need of adjustments or modification. Hitherto, the inter-regional harmonization is incomplete as no regional models were available on the northwest, south and southeast parts of the EMME region. Figure 4 illustrates the geographical distribution of the fully cross-border harmonized, area source model, consisting of 224 crustal area sources (i.e. 213 active shallow area sources and 11 area sources of stable continental regions), 6 area sources describing the deep seismicity
123
Bull Earthquake Eng
(Northern Caucasus, Zagros Fold and Thrust Belt, Pamir and Hindu Kush) and 4 area sources modelling the seismicity of the subduction in slab (e.g. Hellenic, Cyprian and Makran subduction zones). The subduction interface sources are discussed in Sect. 4.4: Subduction Zones. Next section describes our procedures to estimate the earthquake activity rates for all area sources.
4.1.1 Recurrence rates for area-sources To forecast the number of events of different magnitudes within individual area sources, the earthquake magnitude-recurrence rates must be estimated. The latter, expresses the annual frequency of earthquakes of various magnitudes equally binned between a minimum (Mmin) and a maximum magnitude (Mmax). This expression relies greatly on the catalogue. We assume a time independent model for earthquake occurrence and the common exponential magnitude distribution based on the Gutenberg–Richter (1944) formulation Log N(M) = a - b * M; where N(M) is the cumulative number of earthquakes per unit time equal or larger than magnitude M. The a- and b-values are constants. The activity rate (a-value) represents the total seismic productivity of a given source [=log N(M), when M = 0], or the log of number of events (M [ M0); the b-value is the negative slope of the recurrence curve expressing average ratio of exponentially distributed small and large magnitude earthquakes. Earthquake recurrence parameters of each area source are derived from the declustered catalogue using the completeness intervals as defined for each macrozone and applying a maximum likelihood procedure (Weichert 1980). For all area sources, i.e. crustal shallow, deep and subduction, the resulting recurrence parameters are depth dependent. Thus, only earthquakes shallower than 40 km are considered for shallow crustal sources. Earthquakes between 40 and 100 km depth are used to characterize in-slab sources, whilst those deeper than 100 km describe the recurrence of the deep seismicity sources (e.g. Hindu Kush). Further, a minimum of 10 earthquakes was imposed to flag the about 20 area sources without enough data to derive stable recurrence parameters. The seismicity of these particular area sources is assigned a b-value of 0.85 and an a-value of 3.35, which are the lowest values estimated within the stable continental regions (Table S2 of the Electronic
Fig. 4 Geographical distribution of the cross-country harmonized area source model; yellow polygons are area sources inherited from ESHM13 model; blue polygons are of EMCA15 model; green polygons depict area sources of subduction inslab; magenta polygons are the surface projection of subduction interface; polygons with red boundaries are the stable shallow sources, whereas the polygons in red delineate area sources of deep seismicity
123
Bull Earthquake Eng
Supplement). In other words, these values represent the water level of earthquake potential over the entire region for the time interval of interest. Examples of earthquake recurrence rates of four seismogenic sources are presented in Fig. 5. Each plot displays the rates of earthquakes (solid circles) and the best-fit curve truncated at the corresponding maximum magnitude values, flanked by curves representing
123
Bull Earthquake Eng b Fig. 5 Recurrence curves of six seismic sources representative for active shallow crust (a), stable regions (b), deep seismicity (c), subduction interface (d) and subduction inslab (e). Cumulative number of events per year are plotted as circles, the black curves indicate frequency distribution corresponding to different Mmax values; the uncertainty range illustrated in blue area between the lower and upper parameters uncertainties curves, as resulting from applying the algorithm of Weichert (1980). (f) Example of an expert fitting (in red) for an area source without suitable data for automatic fitting. In this particular case, the adjusted curve favours the observed recurrence rates of moderate to large magnitude events, as the instrumental data might be incomplete
lower and upper error bounds. However, these are neat examples of recurrence fitting. In practice, a large statistical variability in the recurrence parameters is observed, which is generally attributed to the quality and quantity of the earthquake catalogue. On the one hand, the catalogue consists in an incomplete report of small magnitude earthquakes in early periods, which introduces errors on the recurrence fit. On the other hand, lack of large magnitude earthquakes within a seismic source also substantially increases the uncertainty in establishing recurrence rates. Besides, the number of earthquakes greatly varies within area sources, which increases the statistical variability in using the maximum likelihood method. The latter is controlled by the number of earthquakes observed and not by their magnitude. This tends to predominantly fit smaller, more frequent events. Thus, a direct attempt for an automated fit to sparse seismicity data results in biased recurrence parameters. Woessner et al. (2015) identified similar issues and adopted a fitting procedure based on expert opinions. We also adopted an expert fitting procedure, implying a sourceby-source revision to adjust the automated fittings for area sources with inconsistent earthquake data. Figure 5e offers an example of adjusted fitting to match recurrence rates of magnitudes 5.5–6.0 Mw. The fitting red line in Fig. 5e overestimates the recurrence rates of magnitudes \ 5.5 Mw when compared with the black line of automatic fitting (Fig. 5e). The higher recurrence rates of low magnitudes reflects several factors: limited number of events in seismic zones, lack of entries in the earthquake catalogue, bias in low magnitude conversion, errors in the estimation of the completeness intervals. However, for area sources with sparse seismicity, the expert fitting remains within the uncertainty range of the automatic fitting algorithm. In summary, area sources are characterized by a specific earthquake recurrence (a- and b-values) fitted by experts where necessary, a distribution of Mmax and the uncertainties in the recurrence parameters of the model. The minimum magnitude (Mmin) is 4.0 Mw. The forecasted seismic productivity is assumed to be uniform within each polygon. Additionally, area sources are characterized by alternative values of source specific parameters that were not treated as epistemic but rather aleatory. These parameters represent the depth distribution and the predominant style-of-faulting.
4.2 Active faults The compilation of the EMME_fsdb is a step forward in developing active faults specific source models for seismically active regions. We note that not all the mapped faults are seismogenic as the key information to characterize the fault-specific earthquake recurrence, i.e. the slip rate, was not available. Thus, active faults were selected from the entries of the EMME_fsdb if they met the following conditions: • Identified Holocene active faults that generated earthquakes during the last 10,000 years, associated with high values of slip rates (i.e. Northern Anatolian Faults, Marmara Faults, Zagros Faults).
123
Bull Earthquake Eng
• Active during the second half of the Quaternary (roughly *1 million years) with an average slip-rate of at least 0.1 mm yr-1. • Late Quaternary active faults (observed or assigned fault movement during the last 130,000 years). • Geological features (section, segment or faults) having a slip rate of at least 0.1 mm yr-1 corresponding to about 1 m of displacement during the Holocene (*10,000 years). • Maximum moment magnitude (Mw) of at least 6.20. • Available data of both geometry and geological slip rates. In total 778 active faults were retained to form the fault source model. This number is significantly smaller that total entries of the EMME_fsdb. Albeit preferable, the lack of accuracy or information prevents using all records of EMME_fsdb. The next section provides insights about the estimation of the fault-specific earthquake recurrence parameters.
4.2.1 Recurrence rates of active faults Seismic activity of faults is estimated from geological slip rates linked to earthquake recurrence through the use of seismic moment (e.g. Anderson 1979; Molnar 1979; Anderson and Luco 1983; Youngs and Coppersmith 1985; Stirling et al. 2012) as applied with other regional PSHA models (Field et al. 2014; Moschetti et al. 2015; Woessner et al. 2015). Geological slip-rates integrate several seismic cycles of large-magnitude earthquakes on a fault, offering advantages over observed (instrumental or historical) seismicity that the latter is normally too short to forecast the large magnitude events on any given fault. Characterizing active faults with seismic moments requires knowledge of average rate at which faults are slipping, the relative rate at which tectonic strain takes place, Mmax and crustal rigidity. Once a seismic moment is estimated for a fault, it must be translated into magnitude domain according to a selected magnitude recurrence model. In this analysis, exponential magnitude distribution is preferred to characteristic earthquake model (Schwartz and Coppersmith (1984). Wesnousky et al. (1983) have shown that the seismicity on a fault may not obey the GR relation because few or no moderate sized earthquakes may occur during the cycle of a maximum magnitude earthquake expected on a fault, thus resulting in an increase of small to moderate earthquakes. From a different perspective, Hofmann (1996) also argued that individual faults cannot produce an exponential earthquake recurrence and that the combination of regional background with faults results in an exaggerated number of moderateto-small magnitude events. A way to alleviate double-counting seismicity is to combine recurrence of small magnitude events, obtained from seismicity data, with recurrence intervals for largemagnitude events on an individual or a system of faults. A threshold magnitude separates the two recurrence models and limits the minimum magnitude on individual faults to a slightly higher value than the maximum considered for the background zone. For instance, a threshold of 6.5 Mw was used in ESHM13 to separate the background seismicity recurrence from the recurrence due to faults. The slope of the fault-specific recurrence curve was controlled by b-values statistically estimated in each background zone and the embedded faults were assumed to be spatially completed within each background zone. Here, a slightly different approach than that in ESHM13 was followed. We derived asymmetric buffer zones around each mapped active fault instead of background area
123
Bull Earthquake Eng
sources. A buffer zone results from combining the fault-width projection (the hanging-wall side) to the Earth surface with an arbitrary polygon that extends up to 5 km on the footwall side, along the fault trace. Buffers are allowed to merge and generate larger buffer zones. Within the fault buffer, the threshold magnitude was set to 5.5 Mw as it is difficult to assign low magnitude events to identified faults. Earthquakes with magnitudes below 5.5 Mw are assumed to occur anywhere in the buffer zones, whereas the magnitude values above 5.5 Mw are linked to individual faults. In this way, double-counting seismicity in the fault proximity is avoided. In regions off buffer, seismicity is regarded as background. Sketch maps to illustrate the outcome of these procedural steps are given as Electronic Supplement to this manuscript. Details of the characterization of the in-buffer and background seismicity are discussed in Sect. 4.3. Following an evaluation process, we considered the Model 2 of Anderson and Luco (1983) and a truncated exponential magnitude-frequency distribution to characterize the seismic activity of active faults. The functional form of the recurrence relationships is provided in Anderson and Luco (1983), and not repeated here. To assess the recurrence models, we distinguish between two fault types present in our geo-dataset: individual and composite faults. Individual faults have a trace length corresponding to its assigned maximum magnitude (i.e. segments along Northern Anatolian Fault system). To characterize an individual fault (or segment) we employ the equations II.1 and II.8 from Table 3 (p. 480) of Anderson and Luco (1983). Composite faults (Basili et al. 2008, 2013b) have significant uncertainty about their length apparently inconsistent with the given maximum magnitude. Recurrence of such composite faults (i.e. faults in Caucasus Region) was computed with equations II.1 and II.4 in Table 3 (p. 480) of Anderson and Luco (1983). For all faults, the following values were considered as default: crust shear modulus (l) of 3.0 9 1011 dyne/cm2; typical values (c = 16.05 and d = 1.50) for the magnitude-moment scaling coefficients as originally proposed by Kanamori and Anderson (1975); an average value of fault-slip to fault length ratio (a = 1.25 9 10-5) as recommended by Anderson and Luco (1983). The remaining key input parameters are b-values and fault slip rate. Regional b-values statistically estimated from seismicity for each macrozone are listed in Table S2 of the Electronic Supplement. To use the fault specific slip-rates in estimating the average earthquake frequency, the following assumptions were adapted from Schwartz and Coppersmith (1986): (1) the slip rate is representative of the future time period of interest; (2) all slip measured across the fault is seismic with the coupling factor set to unity; (3) slip is constant across the fault area—the surface slip measured at the top trace is representative of slip at seismogenic depth; (4) the slip rate is a geometric mean. A final consideration in assessing the seismic activity from slip rates is the choice of the Mmax. For a constant slip rate, increasing the Mmax decreases the recurrence rate of low-tomoderate magnitude events. This is because the large magnitude events absorbs much of the total seismic moment rate, while increasing the magnitude requires subtracting many smaller earthquakes to preserve the same seismic moment budget (Youngs and Coppersmith 1985). Notwithstanding the importance of Mmax, we chose not to alter the Mmax values reported for each fault mainly because of their correlation with the fault size. No additional increment was added to the Mmax values of the faults. Moreover, a sanity-check investigation revealed discrepancies between some fault dimensions and the assigned Mmax by compilers of the EMME_fsdb. Specifically, there are small-sized faults with high Mmax value; as such, the rupture corresponding to this Mmax might extend beyond the fault geometry limits. Only for these individual faults, the Mmax
123
Bull Earthquake Eng
was empirically estimated from their geometry using empirical equations (Wells and Coppersmith 1994). Figure 6 compares the annual earthquake recurrence rates of the in-buffer zones together with those generated by both the embedded faults and the corresponding macrozone. The annual earthquake rates of the buffer zones are double truncated within Mmin = 4.0 and the threshold magnitude (Mthr = 5.5) whilst the exponential recurrence curves of individual faults is defined by the Mthr and the fault specific Mmax. Annual earthquake rates of the macrozones depict the regional seismicity represented by the cumulative number of events from Mmin to Mmax. The two regions considered for comparison are the Marmara and the Dead Sea. Figure 6 indicates a large variability of earthquake recurrence rates due to faults when compared with the seismicity counterpart for both regions. The difference between the seismic productivity of the faults and the background seismicity is evident when comparing the recurrence rates above Mthr. Total productivity of the faults (black line in Fig. 6) is significantly larger than the seismic productivity of the background zones (blue line in Fig. 6). The difference may indicate either incomplete catalogues (both instrumental and historical) or overestimated slip-rate values. Moreover, the seismic productivity is statistically obtained from analysis of the earthquake catalogue spanning over 1500 years, which may not describe the recurrence cycle of large magnitude earthquakes. Thus, the seismic productivity of the faults may be an indication of a much longer geological time (e.g. 10,000 years) than the time span of the catalogue. Over the entire region, the total seismic moment based on fault slip-rates is greater than the seismic moment from seismicity. Similarly, recurrence rates for crustal faults in the Dead Sea region exceed seismicity, although dispersion is lower. Overall, it is obvious that the seismicity is well constrained for large regions, but the seismic moment budget may be unbalanced when seismogenic faults are added.
Fig. 6 Individual recurrence curves of crustal shallow faults, in-buffer, background seismicity and macrozones. The blue circles indicate the annual cumulative number of events; the red-dashed curves are the magnitude frequency distributions (mfd) of individual faults, the grey line indicates the mfd of in-buffer seismicity, light grey-dashed line marks the background seismicity, whereas the blue-line represents the mfd of the background macrozone. The black line marks the summed mfds of all faults, in-buffer seismicity and background seismicity. When compared with the macro zones mfd, the differences are obvious above threshold magnitude (5.5 Mw) as the contribution from faults is dominant
123
Bull Earthquake Eng
4.3 Background smoothed seismicity We used both historical and instrumental seismicity to describe (1) the earthquake pattern of regions with no identified faults and (2) the small magnitude events assumed to occur off-faults. We refer to the former as background seismicity and to the latter as in-buffer seismicity. Spatially smoothing seismicity over appropriate intervals of catalogue completeness defines the background seismicity. It can be seen as an alternative to area-sources removing their subjective delineation. When counting and smoothing seismicity, the spatial stationary implies that the pattern of past seismicity is predictor of the pattern of future seismicity in regions without known faults (Kafka 2007). However, the link between the spatial small-to-moderate earthquakes and the pattern of large-magnitude earthquakes is debatable as the duration of the seismic record is too short to ascertain such a hypothesis (Hiemer et al. 2014). In addition, the choice of the smoothing kernel is arbitrary; Therefore, smoothing seismicity models introduce subjectivity too. The in-buffer seismicity characterizes seismicity in faults’ proximity. As discussed in Sect. 4.2.1, the in-buffer seismicity accounts for smoothing-specific earthquakes rates up to 5.5 Mw. Recurrence rates of earthquakes of magnitudes [ 5.5 Mw are fault specific. This separation is required to avoid double-counting of earthquakes linked to active faults. Background and in-buffer sources spatially complement the identified seismogenic faults, hence resulting into an independent seismogenic model. Each seismogenic point source is characterized by grid specific recurrence parameters (a- and b-values) estimated on the declustered and complete seismicity samples using the smoothing algorithm of Frankel (1995). Spatial variation of earthquakes within each cell is modelled applying a Gaussian kernel to smooth seismicity between neighboring cells. Results of a sensitivity analysis to specific kernel bandwidths (e.g. 5, 10, 15, 20, and 25 km) indicate 20 km as suitable. Each point source is represented by its geometry properties, i.e. depth distribution, styleof-faulting and corresponding recurrence parameters i.e. activity, b-value, lower and upper bound magnitude values. Mmax of background seismicity is geographically variable and inherited from the corresponding area sources, whereas the Mmax is 5.5 Mw for all point sources in buffer. Activity rates are reported in the middle of each grid cell of 10 9 10 km. Background and in-buffer seismicity complement the active faults and are presented in Fig. 7. Deep seismicity sources and subduction zones are included. Details of the subduction source modelling are discussed next.
4.4 Subduction zones Two types of seismogenic sources are selected to model subduction zones: complex faults and area sources. The former are preferred to describe the subduction interface whereas the latter are used to delineate the deep slab. Subduction sources cover the southeastern part of the Hellenic, Cyprian and Makran subduction zones. The subduction sources (both interface and in-slab zones) of the Hellenic and Cyprian subduction zones were inherited from the sibling ESHM13, thus not described here. An illustration of all subduction models is given as Electronic Supplement to this manuscript. The Makran subduction zone (Figs. 1, 7) is characterized by low level of seismicity and few large earthquakes. This subduction zone is delineated by an offshore deformation front located 150 km away from the coastline. It is mostly aseismic and there is marked
123
Bull Earthquake Eng
Fig. 7 Geographical distribution of the active faults and background seismicity model; black lines are active faults; green polygons depict the subduction inslab sources; magenta polygons are the surface projection of subduction interface; polygons with blue delineate the deep seismicity; background seismicity is represented as smoothed activity rates (a-values)
seismicity differences between the east and west of near the Iran-Pakistan border (Yeats 2012). Eastern Makran, in which large earthquakes (1765, 1851, 1945) have occurred, hosts frequent small to moderate thrust earthquakes. The slap dip angle of about 10N extends for about 400–500 km. Western Makran is a wide quiet region (Byrne et al. 1992). However, few inland earthquakes in western Makran occur at intermediate depths, likely within the down going plate. They are thus modelled as in-slab events. The slab was represented as area sources located at depths ranging 50–150 km. Two alternative models delineating the subduction interfaces as complex faults were used. The first model consists in two complex faults dipping arcward from and along the plate boundary (see Fig S3a, in the Electronic Supplement to this manuscript). The second model depicts a clear separation between the eastern and western zones of the coastal, inner and northern part of the Makran subduction interface zone (see Fig S3b, Electronic Supplement to this manuscript). Earthquake recurrence parameters of the first model are estimated from a declustered subset of the earthquake catalogue valid for the Makran region and the corresponding completeness is given in Table S1 of the Electronic Supplement. For the second model, we introduced an alternative recurrence model to account for the uncertainty in associating ill-documented historical earthquakes with the subduction zones. The alternative recurrence model converts the plate boundary slip-rates to earthquake productivity of the subduction interface by using the slip to moment conversion (Anderson and Luco 1983). The average values of subduction slip-rates were about 33 mm yr-1 for Hellenic and 18 mm yr-1 for the Cyprian subduction zones. The annual rate of subduction in Makran is between 32 and 35 mm yr-1 (McClusky et al. 2003). Inferred GPS observations show convergence of 19.5 mm yr-1 between Oman and Iran (Musson 2009; El-Hussain et al. 2012; Yeats 2012). Hence, we attributed a slip-rate value of 20 mm yr-1 for the Makran interface. Makran subduction interface scales with depth of 10–45 km. A particular case of subduction slab is the deep seismicity concentrated between 80 and 200 km depth beneath the Hindu Kush Mountains (Pegler and Das 1998; Negredo et al. 2007; Yadav et al. 2013). A single area source describing seismicity at different intermediate depths was used. Additionally, five area sources were delineated to account for earthquakes deeper than 50 km in the Zagros Fold and Thrust Belt and Mid-Caspian regions. Note that the EMME seismic catalogue has questionable deep events in the Zagros region (Engdahl et al. 2006).
123
Bull Earthquake Eng
In summary, two alternative branches describe the non-crustal earthquakes. The first branch describes subduction sources of the Mediterranean, deep seismicity of Mid Caspian, ZFTB, Pamir, Hindu Kush and Makran subduction in-slab and interface (see Model 1 in Fig. S3a, Electronic Supplement to this manuscript). Similarly, the second branch comprises subduction interface zones of Mediterranean and Makran (see Model 2 in Fig. S3b, Electronic Supplement), deep seismicity. The first branch is complementary to the area source model (Fig. 4) and the second branch complements the active faults seismogenic source model (Fig. 7). All subduction interface sources are characterized by interfacespecific earthquake recurrence rates obtained from the catalogue when used with area source model. In contrast, when used with the active faults plus background seismicity, subduction interface sources are characterized by earthquake recurrence rates converted from subduction slip-rates. Inslab and deep seismicity are modelled as area sources and recurrence rates as described in Sect. 4.1.1 (Fig. 5).
5 Uncertainties and logic tree Development of a seismic hazard model requires decisions and wide interpretations at every step, including efforts to separate uncertainty types. Uncertainties are generally distinguished as epistemic and aleatory. The epistemic part resides on the imperfect representation and understanding of on-going geological processes, hence reflects incomplete knowledge and data limitation to make predictive models. Aleatory variability accounts for randomness associated with the prediction of a parameter from a specific model, assuming that the model is correct. In PSHA, aleatory uncertainty is carried out throughout the integration procedure, whereas epistemic uncertainties are expressed by incorporating multiple hypotheses, models or parameter values in a logic tree. As originally proposed for PSHA (Kulkarni et al. 1984), the logic tree implies that branches are mutually exclusive and that all may be collectively exhaustive at a ‘‘node’’ of the tree (Bommer and Scherbaum 2008). However, in practice, these criteria are difficult to be fulfilled due to
Fig. 8 Master logic tree consisting of one branching level of two alternative seismogenic source models. Horizontal segments enumerate independent, not alternative models. First branch consists of area sources specific to shallow crust (both active and stable), deep seismicity, subduction inslab and complex faults of subduction interface. The recurrence parameters of this branch are statistically estimated from regional earthquake catalogue. Second branch depicts the active faults combined with background seismicity, area sources of deep seismicity and subduction inslab and complex faults for subduction interface. The recurrence parameters for faults and subduction interface are estimated from seismic moment balance and slip-rate conversion to seismic activity
123
Bull Earthquake Eng
correlation of multidisciplinary data, cognitive bias of experts, expert roles, and peer review processes. To account for these uncertainties, we built alternative interpretations of the key elements and employ a logic tree structure. The logic tree, as shown in Fig. 8, consists of a simple branching level accounting for two alternative seismic source models: the area sources model and the active faults plus background seismicity model. Slight preference (0.6/0.4) for the area model over the fault model reflects the fact that the area sources correlate tectonic and geological information with a pattern of observed seismicity. The weights reflect this degree of correlation, where the area sources account spatially for the active faults because they were delineated according to fault location. Also, area sources account for future earthquakes that are not associated with mapped faults. On the other hand, the earthquake-recurrence of the area source model is solely based on a short observational time, the about 1500 year length of the earthquake catalogue. Within this time interval, a complete earthquake cycle may have not finished for a given area source. Active faults may be more appropriate to account for large earthquakes as the slip rates may cover one or more seismic cycles with complete information dating back to hundreds years. At the same time, the fault-specific earthquake recurrence is also correlated with the regional seismicity through the regional b-values. Thus, a degree of redundancy in seismological and geological data prevents stating which seismogenic model is more suitable. Summary statistics of earthquake recurrence parameters of seismogenic sources are listed in Table 1. The b-values across the seismogenic area sources are remarkably stable with median values of 0.92 for active crust, 0.93 for stable regions, 0.94 for deep sources and 0.89 for subduction interface. Seismic productivity of active shallow area sources quantified by median a-values normalized to km2 is almost double that corresponding to stable shallow sources (-0.59 and -1.22, respectively). An increase of productivity is observed for subduction interfaces (normalized median a-value about -0.18). Seismic activity per km2 of the fault sources (0.18) is much higher than the area source counterpart. The complementary statistics for Mmax together with the corresponding mean seismic moment are listed in Table 1. Further, the Mmax uncertainty is considered for all seismic sources except active faults. The reasoning was that active faults have Mmax equivalent to the fault dimensions; Hence, we did not apply further constraints on this parameter. It is worth noting that we decided to treat the epistemic uncertainty on Mmax as an aleatory uncertainty for computational optimization. In a similar way, the uncertainties of hypocentral depth and focal mechanisms of area and background seismicity sources are considered aleatory. Fig S4 (in the Electronic Supplement) presents the alternative options to the source specific parameters of all source typologies as parameterized for use with the hazard library of OpenQuake (Pagani et al. 2014). Thus, the epistemic uncertainty of the seismogenic source model does not represent the full distribution of all source parameters, but instead the central part of the distribution, as described by weighted mean values. The trade-off is that the ground motion models control the distribution of the hazard estimates. A similar strategy was adopted in the ESHM13.
6 Summary and remarks An earthquake source model accounting for seismogenic patterns of Middle East was developed in the EMME project (http://www.emme-gem.org/). The model covers a complex tectonic region, from the Hellenic subduction zone in the west to the Hindu Kush
123
Bull Earthquake Eng Table 1 Summary statistics of the earthquake recurrence parameters per seismic sources aggregated per tectonic features Recurrence parameters
Mean
Minimum
Median
Maximum
Seismic moment rate (dyne-cm/yr)
Active shallow crust (213) a-value a-value normalized (km2)
3.67
1.64
-0.63
-3.76
3.7
5.31
-0.59
0.92
b-value
0.92
0.6
0.95
1.13
M01 max
7.29
5.2
7.4
8
1.41E?27
M02 max
7.57
5.4
7.7
8.2
3.98E?27
M03 max
7.85
5.8
8
8.4
1.12E?28
Stable regions (11) a-value
3.57
3.35
-1.15
-1.38
-1.22
b-value
0.93
0.85
0.95
1
M01 max
7.07
6.8
6.8
7.4
1.78E?26
M02 max
7.37
7.1
7.1
7.7
5.01E?26
M03 max
7.67
7.4
7.4
8
1.41E?27
a-value normalized (km2)
3.5
4.35 -0.86
Deep seismicity/subduction inslab (10) a-value
4.22
3.84
4.15
-0.88
-1.68
-0.84
-0.61
b-value
0.94
0.72
0.99
1.07
M01 max
7.74
7
7.7
8.2
3.98E?27
M02 max
8.04
7.3
8
8.5
1.12E?28
M03 max
8.34
7.6
8.3
8.8
3.16E?28
a-value normalized (km2)
4.8
Subduction interface (6) a-value
3.89
3.65
3.95
4.03
-0.16
-0.46
-0.18
0.19
b-value
0.89
0.85
0.9
0.9
M01 max
8
7.4
8.2
8.2
2.24E?28
M02 max
8.3
7.7
8.5
8.5
6.31E?28
M03 max
8.6
8
8.8
8.8
1.78E?29
a-value normalized (km2)
Active faults (778) a-value
3.28
0.85
3.3626
5.14
a-value normalized (km2)
0.14
-1.66
0.18
1.78
b-value
0.95
0.7
0.95
1.13
Mmax
7.14
6.4
7.14
8.33
5.75E?26
The minimum, mean, median, maximum values of the key earthquake recurrence parameters (a-value, bvalue and Mmax) are statistics of all seismic sources within a tectonic feature. Mmax corresponds to the uncertainty values discussed in Sect. 3.2: Assessing Mmax, and the other source specific parameters are the a-value and b-value. The seismic moment is computed from the mean value of the Mmax. The recurrence values are not normalized, thus they are statistics of the source distribution within each tectonic feature
in the east. The main tectonic features, geophysical information and reported seismicity form the basis of the earthquake source (seismogenic) model. We used multidisciplinary procedures to analyze data with particular attention on harmonization of the input datasets. A regional earthquake catalogue homogeneous in terms of moment magnitude (Mw) was
123
Bull Earthquake Eng
developed for the probabilistic seismic hazard analysis. The catalogue covers a time period of 1500 years, which may be too short to include a full seismic cycle everywhere in the EMME region, but it is likely near complete for low magnitudes. Compilation of a database of active faults throughout the studied region bears significant importance. Recent, large magnitude earthquakes in regions of relatively low seismic hazard (e.g. the 7.7 Mw, 2001, Gujarat, India, the 7.9 Mw, 2008, Wenchuan, China, the 7.0 Mw, 2010, Port–auPrince, Haiti, the 6.3 Mw, 2011, Christchurch, New Zealand) raised the importance of mapping faults for seismic hazard assessment. Coincidentally, the 2013 Balochistan earthquake happened while regional experts were gathered in Istanbul to discuss the use of active faults in the PSHA model, in particular the class C faults in our classification scheme in Afghanistan and Pakistan. The 7.7 Mw, 2013, Balochistan hypocenter was located at a depth of 15 km on a newly identified fault. There was an instant consensus that active faults are key features, and efforts were intensified to complete the dataset of active faults in the investigated regions. The newly compiled EMME_fsdb allows a better understanding of the seismogenic patterns. However, how area sources should be prepared is not much discussed in the literature (Musson et al. 2009). We proposed a set of criteria to help delineating area sources, and those were applied to unify country-based source models without cross-border variation. These criteria are only generic rules listed for transparency of the model. Further homogeneity was achieved by inter-regional integration with sibling area source models of ESHM13 and EMCA15. When estimating the magnitude recurrence parameters, the dominant ingredient remains the seismicity, although we additionally used slip-rates, where those are known. Yet, earthquake recurrence rates estimated from geological slip-rate are still dependent on the regional seismicity pattern through the use of regional b-values. In summary, the EMME-SSM14 consists of three typologies of seismogenic sources: classical area-sources, gridded point sources and faults—crustal or subduction interface. We have delineated 234-area sources representative of all tectonics e.g. active and stable shallow crust, subduction inslab and deep seismicity (Fig. 4). The delineation of area sources follows regional patterns of seismicity and active faults. The rate of seismicity in each area source is statistically estimated from the newly compiled earthquake catalogue. The area sources encompassing North and East Anatolian Fault Systems, Dead Sea Fault System, Zagros Folds and Thrust Belt, Greater Caucasus, Tabriz, Chaman Faults, Pamir, Hindu Kush, Hellenic and Cyprian subduction zones and Eastern Makran exhibit the highest rates of seismic activity. The annual recurrence rates of Mw C 7.0 is 0.0020–0.0010, which converts to mean return periods of 500–1000 years. Furthermore, 778 crustal faults and 3 subduction interface sources (i.e. Mediterranean Sea and Makran) combined with gridded point sources were distributed through the entire region. Mmax is estimated on large background macrozones based on the observed seismicity and then transferred to all sources but faults. Among these active faults, 122 forecast Mw C 7.0 with mean return periods of about 500 years. In fact, these seismically active faults coincide with the area sources of high rate of seismic activity, both following known plate boundaries, active tectonic processes and historical seismicity. The characterization of seismicity patterns and off-faults rates is based on a smoothing algorithm applied to equally distributed grid cells. Background off-faults seismicity is alternative to area sources in regions without faults, hence removing the subjectivity of source delineation. In contrast, the gridded seismicity in proximity of faults, referred to as in-buffer seismicity, is used to complement the earthquake rates of faults with small to moderate earthquakes.
123
Bull Earthquake Eng
Mmax is estimated from historical seismicity within large background zones and transferred to all sources excepted active faults. The latter preserve their Mmax provided by the compliers. We combined all sources into a logic tree structure to quantitatively account for their spatial incompleteness and alternative values of model parameters. The logic tree of the seismogenic model comprises one branching level, with area sources weighted slightly higher (0.6/0.4) than the active faults penalizing the degree of correlation between data of the two seismic source models. The earthquake source model presented herein provides a regional model with a certain degree of simplification. For site-specific analyses, we recommend models to be subjected to refinement of data and information. Moreover, the seismogenic source model represents a step forward on data collection and harmonization, a model to overcome cross-border problems. We strongly believe that the availability of the standardized input files (Danciu et al. 2016b) for OpenQuake, throughout the site of the European Facilities of Earthquake Hazard and Risk (EFEHR, www.efehr.org) will provide a unique opportunity to repeat the calculation, perform sensitivity analyses, benchmark and calibrate other models. Comparison at local and regional scales is encouraged as well as further academic discussion and eventual improvement of the seismic hazard model of the region. We wish to express confidence that this study represents the starting point of more detailed studies in the region, focusing on data collection and use of homogeneous techniques and tools for data analysis. Acknowledgements We would like to acknowledge the collaborative efforts of various local and regional researchers throughout the project. The following individuals have contributed to EMME-SSM14 in a major way by providing data, specific source models, feedback and comments: Sinan Akkar, Arif Axhundov, Avetis Arakelyan, Tamaz Chelidze, Raffi Durgaryan, Mohsen Ghafory-Ashtiany, Rasheed Jaradat, Sepideh Karimi, Ozkan Kale, Saud Quraan, Dinc¸er Ko¨ksal, Yig˘it Ince, Gianluca Valensise, Alexandre Gventcadze, Nino Gaguadze, Mohammad Reza Zolfaghari and M. Tolga Yilmaz. We thank Marco Pagani and Graeme Weatherill at Global Earthquake Model for their help and guidance throughout the project. We also thank Jochen Woessner (SHARE-Project), Stefano Parolai, Dino Bindi and Shahid Ullah (EMCA-Project) for their efforts on cross-border harmonization. Further, we would like to express our gratitude to the OpenQuake IT development team, which provided constant and steady support during the EMME project. More specifically, the support was granted by: Michele Simionato, Daniele Vigano, Lars Butler and Paul Henshaw. M. Sayab acknowledges his former organization, NCE in Geology, Peshawar University, for EMME-related research facilities. Finally, we thank Celine Beauval and an anonymous reviewer for their constructive comments and review of the manuscript.
Appendix: OpenQuake: tailored to fit OpenQuake was used for calculating the seismic hazard over the entire region (S¸ es¸ etyan et al. 2017, this issue). OpenQuake is built on open-source and open-standards and freely available (www.globalearthquakemodel.org). The key features of OpenQuake (here we refer to the hazard library as OQ-hazard engine) are the state-of-the art seismic source representation, advance treatment of uncertainties and various options for hazard calculators (Pagani et al. 2014). We used the default seismic source definitions of OQ-hazard engine (version 1.5) as the blueprints to design our source models; Hence, individual sources were parameterized according to the User’s Manual (Crowley et al. 2015). According to the software manual, geometry parameters and seismicity occurrence models represent each seismic source. The
123
Bull Earthquake Eng
geometry implies definition of source location, style-of-faulting, and depth. In particular, for the area and point sources, the style of faulting is important. The software allows defining extensive ruptures linked to the magnitude distribution; hence, a seismic source is not anymore a point source. Generation of extensive ruptures, as tuned by style-of-faulting parameters and magnitude, allows correctly computing the distance definitions used by new ground motion models (Bommer and Akkar 2012). The impact of using extensive ruptures on the hazard estimates regarding the point-rupture approximation, leads to a significant increase in the probabilities of exceedence for specific level of motion (Monelli et al. 2014). In our model, the area sources and gridded smoothed seismicity models share the same attributes for style-of-faulting and depth distribution. Specifically, there are three depth values and three style-of-faulting (i.e. normal, thrust, strike-slip) assigned to each individual seismic source. Style-of-faulting of future earthquake ruptures is assessed source by source based on various data sets, including earthquake focal mechanisms, stress indicators, stress orientation and geological structure. Results of the assessment are relative frequency of strikeslip versus normal and reverse faulting averaged across each seismic source. Style-offaulting frequency values are treated as aleatory variability and converted to probabilistic weights for seismic hazard integration (see Fig. S4 in the Electronic Supplement of this manuscript). Additional parameters are the lower and upper seismogenic depth describing the region where source specific extensive ruptures are allowed to propagate. These parameters were obtained mainly from seismicity focal depths, location of top and bottom edges of the faults and the crustal model CRUST 2.0 (Bassin et al. 2000). Crustal faults are modelled as simple faults, and the subduction interface zones are represented as complex faults. A simple fault describes a fault surface projected along strike and dip. A complex fault does not require a dip angle because the geometry can be described by combinations of fault edges to describe top, mid or bottom of a fault surface. Common to all sources is the magnitude scaling relationship (Wells and Coppersmith 1994); the scaling relationship controls the size of floating ruptures as a function of magnitude. A truncated GR (Gutenberg and Richter 1944) magnitude frequency distribution, defined by the activity parameters (a- and b-value), lower and upper magnitude is used to characterize all seismic sources. Minimum magnitude used in the probabilistic hazard calculation is 4.5 Mw. whereas the upper bounds vary accordingly to the Mmax logic tree (see Fig. S4 in the Electronic Supplement of this manuscript). Mmax is treated as aleatory to overcome the computational difficulties arising from multiple factors including complex seismogenic source model, extensive ruptures generation and software optimization.
References Adams J, Halchuck S, Allen T, Rogers G (2015) Canada’s 5th Generation seismic hazard model for the 2015 National Building Code of Canada. In: 11th Canadian Conference on Earthquake Engineering, 21–24 July, 2015, Victoria, Canada, Paper 93775 Allen M, Jackson J, Walker R (2004) Late Cenozoic reorganization of the Arabia-Eurasia collision and the comparison of short-term and long-term deformation rates. Tectonics. doi:10.1029/2003TC001530 Ambraseys NN, Jackson JA, Melville CP (2002) Historical seismicity and tectonics: the case of the eastern mediterranean and the middle east, international handbook of earthquake and engineering seismology, V.81A, ISBN: 0-12-440652-1 Anderson J (1979) Estimating the seismicity from geological structure for seismic-risk studies. Bull Seismol Soc Am 69:135–158
123
Bull Earthquake Eng Anderson JG, Luco JE (1983) Consequences of slip rate constraints on earthquake occurrence relations. Bull Seismol Soc Am 73:471–496 Basili R, Valensise G, Vannoli P et al (2008) The Database of Individual Seismogenic Sources (DISS), version 3: summarizing 20 years of research on Italy’s earthquake geology. Tectonophysics 453:20–43. doi:10.1016/j.tecto.2007.04.014 Basili R, Kastelic V, Demircioglu MB, Garcia Moreno D, Nemser ES, Petricca P, Sboras SP, BesanaOstman GM, Cabral J, Camelbeeck T, Caputo R, Danciu L, Domac H, Fonseca J, Garcı´a-Mayordomo J, Giardini D, Glavatovic B, Gu¨len L, Ince Y, Pavlides S, S¸ es¸ etyan K, Tarabusi G, Tiberti MM, Utkucu M, Valensise G, Vanneste K, Vilanova S, Wo¨ssner J (2013a) The European Database of Seismogenic Faults (EDSF) compiled in the framework of the Project SHARE. http://diss.rm.ingv.it/share-edsf/. doi:10.6092/INGV.IT-SHARE-EDSF Basili R, Tiberti MM, Kastelic V et al (2013b) Integrating geologic fault data into tsunami hazard studies. Nat Hazards Earth Syst Sci 13:1025–1050. doi:10.5194/nhess-13-1025-2013 Bassin C, Laske G, Masters G (2000) The current limits of resolution for surface wave tomography in North America. EOS Trans AGU 81:F897 Berberian M, Yeats RS (2001) Contribution of archaeological data to studies of earthquake history in the Iranian Plateau. J Struct Geol 23(2):563–584. doi:10.1016/S0191-8141(00)00115-2 Bommer JJ, Akkar S (2012) Consistent source-to-site distance metrics in ground-motion prediction equations and seismic source models for PSHA. Earthq Spectra 28(1):1–15 Bommer JJ, Scherbaum F (2008) The use and misuse of logic trees in probabilistic seismic hazard analysis. Earthq Spectra 24:997–1009. doi:10.1193/1.2977755 Byrne DE, Sykes LR, Davis DM (1992) Great thrust earthquakes and aseismic slip along the plate boundary of the Makran subduction zone. J Geophys Res Solid Earth 97(B1):449–478. doi:10.1029/91JB02165 Copley A, Faridi M, Ghorashi M, Hollingsworth J, Jackson J, Nazari H, Oveisi B, Talebian M (2014) The 2012 August 11 Ahar earthquakes: consequences for tectonics and earthquake hazard in the TurkishIranian Plateau. Geophys J Int 196(1):15–21. doi:10.1093/gji/ggt379 Cornell CA (1968) Engineering seismic risk analysis. Bull Seismol Soc Am 58:1583–1606 Crowley H, Monelli D, Pagani M, Silva V, Weatherill G, Rao A (2015) The OpenQuake-engine user manual, global earthquake model (GEM) technical report 2015-09 Danciu L, Giardini D (2015) Global Seismic Hazard Assessment Program—GSHAP legacy. Ann Geophys 58(1):S0109. doi:10.4401/ag-6734 ¨ , Akkar S (2016a) The 2014 Earthquake Model of the Middle East: ground motion model Danciu L, Kale O and uncertainties. Bull Earthq Eng (2016). doi:10.1007/s10518-016-9989-1 Danciu, L, S¸ es¸ etyan K, Demircioglu M, Erdik M, Giardini D (2016b) Input files for OpenQuake used to compute the seismic hazard of the Middle East region within the Earthquake Hazard Assessment of Middle East (EMME) Project. Available at: https://doi.org/10.12686/A3 El-Hussain I, Deif A, Al-Jabri K, Toksoz N et al (2012) Probabilistic seismic hazard maps for the sultanate of Oman. Nat Hazards 64(1):173–210 Engdahl ER, Jackson JA, Myers SC, Bergman EA, Priestley K (2006) Relocation and assessment of seismicity in Iran region. Geophys J Int 167:761–778 Erdik M, Alpay BY, Onur T, S¸ es¸ etyan K, Birgoren G (1999) Assessment of earthquake hazard in Turkey and neighboring regions. Ann Geofis 42:1125–1138 Erdik M, Sesetyan K, Demircioglu MB, Tuzun C, Giardini D, Gu¨len L et al (2012) Assessment of seismic hazard in the Middle East and Caucasus: EMME (Earthquake Model of Middle East) project. In: Proceedings of 15th world conference on earthquake engineering Field EH, Arrowsmith RJ, Biasi GP, Bird P, Dawson TE, Felzer KR et al (2014) Uniform California earthquake rupture forecast, version 3 (UCERF3)—the time-independent model. Bull Seismol Soc Am 104(3):1122–1180 Frankel A (1995) Mapping seismic hazard in the central and eastern United States. Seismol Res Lett 66:8 Giardini D (1999) The global seismic hazard assessment program (GSHAP)—1992/1999. Ann Geophys 42(6). ISSN 2037-416X Giardini D, Woessner J, Danciu L, Crowley H, Cotton F, Gru¨nthal G et al. (2013) Seismic Hazard Harmonization in Europe (SHARE): Online Data Resource. doi:10.12686/SED-00000001-SHARE Gru¨nthal G (1985) The up-dated earthquake catalogue for the German Democratic republic and adjacent areas—statistical data characteristics and conclusions for hazard assessment. In: 3rd International symposium on the Analysis of seismicity and seismic risk, Liblice/Czechoslovakia, 17–22 June 1985 (proceedings vol I, pp 19–25) Gru¨nthal G, Wahlstro¨m R (2012) The European-Mediterranean Earthquake Catalogue (EMEC) for the last millennium. J Seismol 16(3):535–570. doi:10.1007/s10950-012-9302-y
123
Bull Earthquake Eng Guidoboni E, Comastri A, Traina G (1994) Catalogue of ancient earthquakes in the mediterranean area up to the 10th century, SGA ¨ zel N, Horasan G, Yılmazer M, Is¸ ıkara AM (2002) Surface fault breaks, Gu¨len L, Pınar A, Kalafat D, O aftershock distribution, and rupture process of the August 17, 1999 Izmit, Turkey Earthquake. Bull Seismol Soc Am 92:230–244 Gu¨len L, S¸ es¸ etyan K, Adamia S, Sadradze N, Gvencadze A, Karakhanyan A et al (2014) Earthquake model of the middle east (Emme) project: active faults and seismic sources second European conference on earthquake engineering and seismology, 2ECEES, 24–29 August 2014, Istanbul, Turkey Gutenberg B, Richter CF (1944) Frequency of earthquakes in California. Bull Seismol Soc Am 34:184–188 Hiemer S, Woessner J, Basili R, Danciu L et al (2014) A kernel-smoothed stochastic earthquake rate model considering seismicity and fault moment release for Europe. Geophys J Int. doi:10.1093/gji/ggu186 Hofmann R (1996) Individual faults can’t produce a Gutenberg-Richter earthquake recurrence. Eng Geol 43:5–9. doi:10.1016/0013-7952(95)00085-2 Jackson J, Priestley K, Allen M, Berberian M (2002) Active tectonics of the South Caspian Basin. Geophys J Int 148:214–245 Kafka AL (2007) Does seismicity delineate zones where future large earthquakes are likely to occur in intraplate environments? Geol Soc Am Spec Papers 425:35–48. doi:10.1130/2007.2425(03) Kanamori HD, Anderson DL (1975) Theoretical basis for some empirical relations in seismology. Bull Seismol Soc Am 65:1073–1096 Kijko A, Singh M (2011) Statistical tools for maximum possible earthquake magnitude estimation. Acta Geophys 59(4):674–700 Knopoff L (1964) The statistics of earthquakes in Sourthern California. Bull Seismol Soc Am 54(6):1871–1873 Kulkarni RB, Youngs RR, Coppersmith KJ (1984) Assessment of confidence intervals for results of seismic hazard analysis. In: Proceedings of the eighth world conference on earthquake engineering, San Francisco, pp 263–270 Marco S, Stein M, Agnon A, Ron H (1996) Long-term earthquake clustering: a 50,000-year paleoseismic record in the Dead Sea Graben. J Geophys Res 101(B3):6179–6191. doi:10.1029/95JB01587 McClusky S, Balassanian S, Barka A, Demir C, Ergintav S, Georgiev I, Gu¨rkan O, Hamburger M, Hurst K, Kahle H, Kastens K, Kekelidze G, King R, Kotzev V et al. (2000) Global Positioning System constraints on plate kinematics and dynamics in the eastern Mediterranean and Caucasus. J Geophys Res 105(B3):5695–5719 McClusky S, Reilinger R, Mahmoud S, Ben Sari D, Tealeb A (2003) GPS constraints on Africa (Nubia) and Arabia plate motions. Geophys J Int 155(1):126–138. doi:10.1046/j.1365-246X.2003.02023.x Meletti C, Galadini F, Valensise G, Stucchi M, Basili R et al (2008) A seismic source zone model for the seismic hazard assessment of the Italian territory. Tectonophysics 450(1):85–108 Mignan A, Danciu L, Giardini D (2015) Reassessment of the maximum fault rupture length of strike-slip earthquakes and inference on Mmax in the Anatolian Peninsula, Turkey. Seismol Res Lett 86(3):890–900 Molnar P (1979) Earthquake recurrence intervals and plate tectonics. Bull Seismol Soc Am 69:115–133 Monelli D, Pagani M, Weatherill G, Danciu L, Garcia J (2014) Modeling distributed seismicity for probabilistic seismic-hazard analysis: implementation and insights with the OpenQuake engine. Bull Seismol Soc Am 104(4):1636–1649. doi:10.1785/0120130309 Moschetti MP, Powers PM, Petersen MD, Boyd OS, Chen R, Field EH, Frankel AD, Haller KM, Harmsen SC, Mueller CS, Wheeler RL (2015) Seismic source characterization for the 2014 update of the US national seismic hazard model. Earthq Spectra 31(S1):31–57 Mueller CS (2010) The influence of maximum magnitude on seismic-hazard estimates in the central and eastern United States. Bull Seismol Soc Am 100(2):699–711. doi:10.1785/0120090114 Musson RMW (1999) Probabilistic seismic hazard maps for the North Balkan Region. Ann Geofis 42(6):1109–1124 Musson RMW (2009) Subduction in the Western Makran: the historian’s contribution. J Geol Soc Lond 166:387–391 Musson RM, Sellami S, Bru¨stle W (2009) Preparing a seismic hazard model for Switzerland: the view from PEGASOS Expert Group 3 (EG1c). Swiss J Geosci 102(1):107–120 National Research Council (US) (1988) Panel on Seismic Hazard Analysis, Keiiti Aki, National Research Council (US). Committee on Seismology, National Research Council (US). Board on Earth Sciences, National Research Council (US). Commission on Physical Sciences, Mathematics and Resources. Probabilistic Seismic Hazard Analysis. National Academies Negredo AM, Replumaz A, Villasenor A, Guillot S (2007) Modeling the evolution of continental subduction processes in the Pamir-Hindu Kush region. Earth Planet Sci Lett 259:212–225
123
Bull Earthquake Eng Pagani M, Monelli D, Weatherill G, Danciu L, Crowley H, Silva V et al (2014) OpenQuake-engine: an open hazard (and risk) software for the Global Earthquake Model. Seismol Res Lett 85:692–702. doi:10. 1785/0220130087 Pegler G, Das S (1998) An enhanced image of the Pamir-Hindu Kush seismic zone from relocated earthquake hypocentres. Geophys J Int 134:573–595. doi:10.1046/j.1365-246x.1998.00582.x Petersen MD, Moschetti MP, Powers PM, Mueller CS, Haller KM, Frankel AD, Zeng Y, Rezaeian S, Harmsen SC, Boyd OS, Field N (2015) The 2014 United States national seismic hazard model. Earthq Spectra 31(S1):S1–S30 Reasenberg P (1985) Second-order moment of central California seismicity, 1969–82. J Geophys Res 90:5479–5495 Reilinger R et al (2006) GPS constraints on continental deformation in the Africa-Arabia-Eurasia continental collision zone and implications for the dynamics of plate interactions. J Geophys Res 111(BO5411):1–26. doi:10.1929/2005JBOO4051 Renault P (2014) Approach and challenges for the seismic hazard assessment of nuclear power plants: the Swiss Experience. Boll di Geof Teorica ed Applicata 55(1):149–164. doi:10.4430/bgta0089 Ruleman CA, Crone AJ, Machette MN, Haller KM, Rukstales KS (2007) Map and database of probable and possible Quaternary faults in Afghanistan: U.S. Geological Survey Open-File Report 2007-1103, 1 plate Schmid SM, Slejko D (2009) Seismic source characterization of the Alpine foreland in the context of a probabilistic seismic hazard analysis by PEGASOS Expert Group 1 (EG1a). Swiss J Geosci 102(1):121–148. doi:10.1007/s00015-008-1300-2 Schuster RL, Alford D (2004) Usoi landslide dam and lake Sarez, Pamir Mountains, Tajikistan. Environ Eng Geosci 10(2):151–168 Schwartz DP, Coppersmith KJ (1984) Fault behavior and characteristic earthquakes: examples from the Wasatch and San Andreas fault zones. J Geophys Res. doi:10.1029/JB089iB07p05681 Schwartz DP, Coppersmith KJ (1986) Seismic hazards: new trends in analysis using geologic data. National Research Council, Washington, DC (USA). Geophysics Study Committee; 266, pp 215–230; DOE/ER/ 12018–T10 Sella GF, Dixon TH, Mao A (2002) REVEL: a model for recent plate velocities from space geodesy. J Geophys Res Solid Earth. doi:10.1029/2000JB000033 S¸ es¸ etyan K, Danciu L, Demircioglu M, Giardini D, Erdik M, Akkar S, Gu¨len L, Zare M, et al. (2017) The 2014 Earthquake Model of the Middle East: overview and results (this issue) Sippl C, Schurr B, Yuan X et al (2013) Geometry of the Pamir Hindu Kush intermediate depth earthquake zone from local seismic data. J Geophys Res 118(4):1438–1457. doi:10.1002/jgrb.50128 Stepp JC (1972) Analysis of completeness of earthquake sample in the Puget Sound area and its effect on statistical estimates of earthquake hazard. National Oceanic and Atmospheric Administration Environmental Research Laboratories, Boulder Colorado, p 80302 Stirling M, McVerry G, Gerstenberger M et al (2012) National seismic hazard model for New Zealand: 2010 update. Bull Seismol Soc Am 102:1514–1542. doi:10.1785/0120110170 Stucchi M, Rovida A, Gomez Capera AA et al (2012) The SHARE European Earthquake Catalogue (SHEEC) 1000–1899. J Seismol 17:523–544. doi:10.1007/s10950-012-9335-2 Tatar M, Hatzfeld D, Martinod J, Walpersdorf A, Ghafori-Ashtiany M, Che´ry J (2002) The present-day deformation of the central Zagros from GPS measurements. Geophys Res Lett. doi:10.1029/ 2002GL015427 Tavakoli B, Ghafory-Ashtiany M (1999) Seismic hazard assessment of Iran. Ann Geophys. doi:10.4401/ag3781 Ullah S, Bindi D, Pilz M, Danciu L, Weatherill G et al (2015) Probabilistic seismic hazard assessment for Central Asia. Ann Geophys. doi:10.4401/ag-6687 Vilanova SP, Nemser ES, Besana-Ostman GM, Bezzeghoud M et al (2014) Incorporating descriptive metadata into seismic source zone models for seismic-hazard assessment: a case study of the AzoresWest Iberian Region. Bull Seismol Soc Am 104(3):1212–1229. doi:10.1785/0120130210 Weichert DH (1980) Estimation of the earthquake recurrence parameters for unequal observation periods for different magnitudes. Bull Seismol Soc Am 70(4):1337–1346 Wells DL, Coppersmith KJ (1994) New empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement. Bull Seismol Soc Am 84:974–1002 Wesnousky SG, Scholz CH, Shimazaki K, Matsuda T (1983) Earthquake frequency distribution and the mechanics of faulting. J Geophys Res 88:9331–9340 Wheeler RL (2009) Methods of Mmax estimation east of the Rocky Mountains: U.S. Geological Survey Open-File Report 2009–1018. http://pubs.usgs.gov/of/2009/1018/. Accessed 15 Nov 2015
123
Bull Earthquake Eng Wiemer S, Garcı´a-Ferna´ndez M, Burg JP (2009) Development of a seismic source model for probabilistic seismic hazard assessment of nuclear power plant sites in Switzerland: the view from PEGASOS Expert Group 4 (EG1d). Swiss J Geol 102(1):189–209. doi:10.1007/s00015-009-1311-7 Wills CJ, Ray J, Weldon RJ, Bryant WA (2008) California Fault Parameters for the National Seismic Hazard Maps and Working Group on California Earthquake Probabilities, US Geological Survey Open File Report 2007-1437A Woessner J, Danciu L, Giardini D, Crowley H, Cotton F, Gru¨nthal G et al (2015) The 2013 European seismic hazard model: key components and results. Bull Earthq Eng 13(12):3553–3596 Yadav RBS, Tsapanos TM, Yusuf Bayrak, Koravos G Ch (2013) Probabilistic Appraisal of Earthquake Hazard Parameters Deduced from a Bayesian Approach in the Northwest Frontier of the Himalayas. Pure Appl Geophys 170(3):283–297. doi:10.1007/s00024-012-0488-2 Yeats RS (2012) Active faults of the world. Cambridge University Press, Cambridge Youngs RR, Coppersmith KJ (1985) Implications of fault slip rates and earthquake recurrence models to probabilistic seismic hazard estimates. Bull Seismol Soc Am 75:939–964 Zare M, Amini H, Yazdi P, S¸ es¸ etyan K, Demircioglu MB et al. (2014) Recent developments of the Middle East catalog. J Seismol 18(4):749–772
123