“ DETERMINATION OF CODA WAVE ATTENUATION CHARACTERISTIC OF THE ARMUTLU PENINSULA AND ITS SURROUNDINGS (MIDDLE MARMARA REGION, TURKEY) „

The Armutlu Peninsula, is located in a geologically complex system on the North Anatolian Fault Zone and inside the high populated Marmara Region. Due to its location and importance, the Marmara Region has been subject to various studies in the past; now, this study is aimed to determine the local attenuation characteristic of the Armutlu Peninsula. In this study, 75 earthquake data that were recorded between 2013-2014, were analyzed by 9 seismic stations. The magnitudes (Ml) and the focal depths of the earthquakes vary from 1.5 to 3.3 and 0.9 to 16.9 km, respectively. A single back-scattering model was used for calculation of the coda wave quality factor. The lapse times were determined between 20 and 40 s at intervals of 5 s and they were filtered at central frequencies of 1.5, 3, 6, 9 and 12 Hz band-pass designed filter. Minimum 5 signal/noise ratio and 0.7 correlation coefficient data used to obtain reliable results. The CODAQ subroutine integrated in the SEISAN software was used for data processing. For each station, low values of the quality factor at 1 Hz (Q o ) and high values of the frequency dependent parameter (n) were determined. In the Armutlu Peninsula and its surroundings, Q c =51f 0.91 for 20 s and Q c =112f 0.72 for 40 s window lengths were calculated by using 9 stations. There is a tendency between increasing Q o and decreasing n parameters. Otherwise, for station TRML, located near the geothermal hotspot, these parameters are in direct proportion with each other, as Q c =46f 0.97 for 20 s and Q c =74f 1.06 for 40 s window lengths. These parameter changes are directly connected with the geothermal activity.


INTRODUCTION
The Armutlu Peninsula is positioned in the high populated and industrialized Middle Marmara Region, between northern and middle branches of the North Anatolian Fault Zone (NAFZ), one of the active fault zones in Turkey. The expected large Istanbul earthquake have led to researches in and around the peninsula. Investigating the seismic wave attenuation is important in order to determine the medium structure of the region.
The energy of seismic waves decreases while propagating through the anelastic and heterogenic medium due to intrinsic attenuation, geometrical spreading and multipathing [Stein and Wysession, 2009]. The dimensionless quality factor-Q characterizes the decay rate of the seismic waves [Knopoff and MacDonald, 1958]. After the arrival of direct-S wave phase, the tail of a seismogram can be identified as coda wave generated by scattered waves within the lithosphere. The frequency dependent coda wave quality factor, Q c , is a measure of

TECTONIC AND GEOLOGICAL STRUCTURE OF THE ARMUTLU PENINSULA
The Marmara Region, which has the highest population density and industrialization in the country, is located in northwestern Turkey. The Armutlu Peninsula is in the central part of the Marmara Region and it still continues to receive migration from other cities.
The dextral NAFZ is one of the significant fault system in Turkey. The tectonic structure of Armutlu Peninsula is controlled by the western realm of this fault zone. 24±1 mm/yr movement of the main fault system and its strands generate the crackled inner-structure and complex crustal regime [Reilinger et al., 1997[Reilinger et al., , 2006McClusky et al., 2003]. This feature produces micro-seismic events in and around the peninsula (Figure 1).
The August 17, 1999 Kocaeli (Mw=7.4) and the November 12, 1999 Düzce (Mw=7.2) earthquakes occurred very close to the eastern part of the peninsula and they disastrously affected east and middle parts of the Marmara Region [Erdik, 2001]. After these large right lateral strike-slip mechanism earthquakes, some moderate shakes (the 2016 Gemlik earthquake -Mw=5.2 is the largest one) were recorded on the peninsula and influenced the local people. A major earthquake has not been recorded in the last century along the southern segment of the peninsula [Caka, 2012].
The Armutlu Peninsula which is located between two branches of NAFZ, consists of Lower Paleozoic-Upper Cretaceous metamorphic rocks and Paleozoic aged slightly metamorphosed rocks as basement. Those rocks are overlaid by Upper Cretaceous-Eocene volcanic and sedimentary rocks. A granite emplacement occurred in Middle Eocene and from Miocene to present day aged sedimentary rocks covers the whole northern part of the peninsula [Yilmaz et al., 1995].
The geothermal activity is presumedly under controlled by NAFZ and geothermal sources are widely spread over the peninsula. The hottest point, Termal, is located on the northern realm of the peninsula and its average temperature is 60-70 o C; otherwise along the southern part, the thermal regions have 20-30 o C water temperature [Eisenlohr, 1997] ( Figure 1). The seismic activity of the region is associated with both fault systems and the geothermal activity. E.g. in 2014, the Termal geothermal reservoir caused a seismic swarm [KOERI press release, 2015;Yavuz et al., 2015].

METHOD
Coda waves that are generated by local earthquakes can be interpreted as backscattered S waves from the heterogeneities of Earth's crust and upper mantle. Aki and Chouet [1975] developed the theory of the single backscattering model and in this study this model was used for estimation of coda wave quality factor (Q c ). The coda wave amplitude decay can be defined as a function of time, t, and frequency, f, as where A(f,t) is the coda wave amplitude filtered at a specific central frequency (f); t is the lapse time, Q c is the coda wave quality factor, c is the source function at frequency (f) and α is the geometrical spreading factor that is here chosen equal to 1 for body waves [Sato and Fehler, 1998]. Considering α=1 and taking a logarithm of Equation 1, we obtain YAVUZ ET AL.
Regarding the ray paths that propagate from each scattered heterogeneity, the frequency dependent coda wave quality factor can be calculated from the slope π ⁄ The general dependence of Q c on f can be described as where, Q o is the value of Q c at 1 Hz, f o is the reference frequency, here assumed as 1 Hz and n is the frequency dependence parameter. The Q o and n values represent the general geological structure of the region. Low values of Q o (<200) characterize the region with high tectonic activity; otherwise, high values of Q o (>600) describe stable seismic activity [Aki and Chouet, 1975;Mak et al., 2004;Sertçelik, 2012]. At the same time, if n is higher than 0.8, the crust and upper mantle are tectonically active; while n values lower than 0.5, characterize an inactive region [Atkinson, 2004;Dasovic et al., 2012;Ranasinghe et al., 2014;Dobrynina et al., 2017;Banerjee and Kumar, 2017].

DATA AND PROCESS
The ARmutlu NETwork (ARNET) was installed at the end of 2005 with currently 27 active seismic stations operating by Kocaeli University Earth and Space Sciences Research Center and Helmholtz-Zentrum Potsdam DeutschesGeo-ForschungsZentrum (GFZ) [Tunç et al., 2011]. To determine attenuation features, 75 earthquake data recorded by 9 ARNET three components seismic stations (Table 1) between 2013-2014 are used ( Figure 2). The sampling rate for all stations is set as 100 samples per second. The local magnitude range of the events is from 1.5 to 3.3 and the depths are between 0.9-16.9 km (Appendix 1). The events 3 FIGURE 1. The tectonic features of the middle and eastern Marmara Region and its seismic activity in last two decades. Red lines indicate the North Anatolian Fault Zone, its branches and the micro−fractures [Eisenlohr 1997;Kuşçu et al. 2009;Pınar et al. 2003;Caka 2012].
are located within 0.6-75.6 km epicentral distances; however, the majority of the distances distributes within 40 km. The first locations of the earthquakes were determined by using the SEISAN software [Havskov and Ottemoller, 1999] with the HYPOCENTER algorithm. To this aim, all ARNET stations and the Özalaybey et al. [2002] 1-D velocity model are used (Table 2). Minimum five azimuthally well-distributed stations and up to 0.5 root mean square (RMS) values are the criteria used for obtaining first locations.
Higher than 5 signal-to-noise ratio (S/N) of the seismograms allow us to obtain reasonable Q c calculations [Ottemoller et al., 2014]. The elimination of direct S-wave phases is important to determine the beginning of the coda wave [Rautian and Khalturin, 1978]. The start of the coda (t start ) as measured from the origin time should be placed at 1.5, 2 or 2.5 time the S-wave travel time (t s ) [Havskov et al., 1989]. It is called the lapse time. 2t s is generally acceptable for local earthquakes [Spudich and Bostwick, 1987;Mukhopadhyay et al., 2010] and we decided to assume t start =2t s for all seismograms analyzed in the present study. Minimum 20 s coda window length is acceptable while larger windows could be served for more stabilized solutions Ottemoller et al., 2014]. According to the epicentral distances and magnitude of earthquakes, we have tried longer window lengths with 5 s step increases, but smaller than 40 s in order to analyze a sufficient number of data with ≥5 S/N ratio [Ot-temoller et al., 2014;personal communication with Lars Ottemoller, 2015].
The overall variation of the Q c in different frequencies is an important indication of the heterogeneity of the medium [Calvet and Margerin, 2013]. All coda records were band-pass filtered with 1.5, 3, 6, 9, 12 Hz central frequencies. The bandwidth of each central frequency f c is crucial to get an equal amount of energy. All designed filters should have the same value (f h -f l )/f c ; where f h and f l are the high corner and low corner frequencies, respectively [Aki and Richards, 2002;Ottemoller et al., 2014]. The filters are designed with the following central frequency (corresponding bandwidth) values: 1.5 Hz (1-2), 3 Hz (2-4), 6 Hz (4-8), 9 Hz (6-12) and 12 Hz (8-16). The purpose of using a central frequency as 1.5 Hz is to prevent the loss of long-period waves that can be recorded at distant stations; but, 12 Hz is also required to consider short-period waves that will be seen in the stations at close epicentral distances.
In order to calculate the quality factor, the CODAQ subroutine contained in the SEISAN software package was used [Havskov and Ottemoller, 1999]. Both the three components recorded at each station and their average were chosen for Q c calculation, assuming a minimum correlation coefficient of 0.7 [personal communication with Lars Ottemoller, 2015]. As an example, in Figure 3, the coda wave analysis obtained from the SEISAN software for a vertical component at station AVDN is shown.  [Özalaybey et al. 2002].

RESULTS AND DISCUSSION
In the computation of the quality factor for the investigated area, the stations' location and the distributions of the earthquakes are important parameters. These parameters refer to the ray path, followed by the earthquake waves that depend on the physical structure of the crust through which they travel and the energy loss due to friction, porosity, fluid content, etc. [Barton, 2007].    The stations located inside the peninsula, except the station TRML, reveal a proportional trend between Q o and lapse times. These values suggest the region is dominated by local faults and fractures due to the connectivity between the north and middle branches of the NAFZ. For instance, in the Kopili Fault Zone (India), the coda Q study performed by Bora et al. [2018] demonstrates that the seismic activity is in correlation with the low values of coda wave decay. Additionally, in NW Himalayan Region, Kumar et al. [2005] evaluated low Q o values through the whole main boundary thrusts and other fault segments.

CODA-Q CHARACTERISTIC OF THE ARMUTLU PENINSULA
The Termal district of Yalova, which has a geothermal spot inside the peninsula, houses the station TRML that works very close to this source. The obtained Q o values of station TRML are lower in comparison with the other stations. According to the 20 to 40 s time windows, the Q o values are increasing from 46 to 74 and the n values are also rising from 0.97 to 1.06. Naghavi et al. [2017] obtained high attenuation for the region with thermal springs and volcanic structure in North West of Iranian plateau. Canas et al. [1995] found low Q o values near the Canary Islands that have high magmatic activity.
Seismic waves attenuate due to the scatters caused by heterogeneities or an intrinsic phenomenon attributed to the anelastic behavior of the medium through which they propagate. Intrinsic attenuation is very sensitive to the physical condition of the underlying medium, while scattering attenuation indicates the degree of heterogeneity [Dainty, 1981;Sato and Sacks, 1989]. The scattering characteristic could express heterogeneities of the NAFZ and micro-fractures; otherwise, the intrinsic effect may be the dominant one for a geothermal region.
After determining the individual quality factor for each station, a single calculation was performed for all sta-tions. It identifies an average attenuation characteristic for the Armutlu Peninsula and its vicinity. The number of high-quality data is low on long-periodic waves, such as 1.5 Hz central frequency, and increasing lapse times owing to the localized area of study. Hence, more reliable and useful results could be obtained from lower lapse times and higher frequency content data. These two parameters affect the standard deviation value directly (Table 4).
From Table 5 results the coda wave attenuation can be estimated as Q c =(51±4)f (0.91±0.04) for 20 s lapse time through 2154 number of data. This function could be identified as the general attenuation law of the Armutlu Peninsula and its surroundings. Also, it is in a correlation with individual quality factor calculations for each station.   The attenuation rate difference between eastern and western side of the peninsula attracts the attention (Table  3). Each side is exposed to a different extension regime. It is known that both thermal sources and local fractures are dominant in the west [personal communication with A.M. Celal Şengör, 2018]. The sharp variation of Q o values is interpreted as the result of different inner structures which diversify according to the rarity of local micro-fractures and faults at the eastern side, on the contrary of the western side. According to increasing time window lengths, it is determined that the inner structures of the regions, except for the thermal area, resembles each other. We can interpret these results considering that structural deformation is dominant near the surface because of the local faults, while it is reduced as well as medium heterogeneity at larger depth.

CODA-Q CHARACTERISTIC OF THE ARMUTLU PENINSULA
Among the earlier studies on the neighboring areas of Turkey, Eck [1988] found Q o =65±5 and n=1.05±0.04 for up to 60 s lapse time in Dead Sea Region around Israel; Hatzidimitriou [1993]   at 25 s were calculated by Rahimi and Hamzehloo [2008]; the average frequency relation for Central Iran was determined as Q c =(79±2)f (1.07±0.08) by Rahimi et al. [2010]; a regional Q c =126f 1.05 relation was approximated by Meirova and Pinsky [2004] based on S wave coda decay rate for Israel. The discrepancies between the present results and those ones obtained in previous studies of Marmara Region reported in "Introduction" section and in Table 6 might be due to the differences in the considered window lengths and kind of quality factor vary from one author to another [Barış et al., 1992;Horasan et al., 1998;Kaşlılar-Özcan, 1999;Akyol et al., 2002;Horasan and Boztepe-Güney, 2004;Sertcelik and Guleroglu, 2017]. Generally, most of the studies demonstrate <200 of Q values and this suggest an active tectonic settings.

CONCLUSION
In and around the Armutlu Peninsula, nine seismic stations from the ARmutlu NETwork (ARNET) were used to estimate the quality factor (Q) through a single backscattering model with 75 earthquake data that were recorded between 2013-2014. The values of Q o and n expressing the frequency dependence of the quality factor vary from 51 to 112 and 0.91 to 0.72 for 20 and 40 s lapse times, respectively. These variations indicate a highly complex and tectonically active medium in agreement with the presence of the NAFZ and widespread micro-fractures in the study. In this localized area, the general coda wave quality factor relation results as Q c =(51±4)f (0.91±0.04) for 20 s lapse time. This function is indicative of the general attenuation characteristic of the region.
The geothermal activity around the station TRML is remarkably recognized in this study. The Q o and n values vary from 46 to 74 and 0.97 to 1.06 for 20 and 40 s lapse times, respectively. The direct proportion of Q o and n values is in agreement with the presence of thermal springs and thermal fluidity even depth.
The quality factor results obtained in this study suggest dominant and active tectonics for the region in agreement with the presence of the NAFZ, local fractures and thermal sources.
This study derives from the MSc. thesis of the first author (EY). The authors give many thanks to Dr. Deniz Çaka for recommendations and contributions. The data used in this study were recorded by the ARmutlu NETwork (ARNET); therefore, we are grateful to both Kocaeli Univer-sity and GFZ collaboration colleagues for sharing the data. We would like to thank Prof. Maria Elina Belardinelli (sector editor) and anonymous reviewers for their crucial and helpful comments.