Slip distribution and source parameters of the 20 July 2017 Bodrum-Kos earthquake (Mw6.6) from GPS observations

Abstract Greek-Turkish boundary near the cities Kos and Bodrum has been shaken on July 20, 2017 by a Mw6.6 earthquake. The mainshock is located offshore and did not generate an on-land surface rupture. Analyzing pre- and post-earthquake continuous/survey-type static GPS observations, we investigated co-seismic surface displacements at 20 sites to characterize source parameters and slip-distribution of the mainshock. Fault plane solutions as well as co-seismic slip distribution have been acquired through the inversion of co-seismic GPS displacements modeling the event as elastic dislocations in a half space. Fault plane solution shows a southward dipping normal-type fault segment extending a depth down to ~12 km, which remains within the brittle upper crust. Results from the distributed slip inversion show that the mainshock activated a ~65 km fault section, which has three high slip patches, namely western, central and eastern patches, where the coseismic slips reach up to 13, 26, and 5 cm, respectively. This slip pattern indicates that the pre-earthquake coupling, which is storing the slip deficit, occurred on these three patches.

20 July 2017 Bodrum-Kos earthquake (Mw6.6 -01:31 UTC) occurred along the western section of the Gökova Gulf, between the cities of Bodrum and Kos. The rupture zone cannot be seen directly since the earthquake is located offshore and did not generate any on-land rupture. Therefore, resolving the nodal plane from auxiliary plane is an ambiguity in seismological solutions. In this case, GPS-derived coseismic displacements are of great importance for modeling the source parameters and the slip distribution of the mainshock. In this study, co-seismic surface displacements are converged to characterize source parameters and the slip distribution of the 2017 Bodrum-Kos Earthquake (Mw 6.6) assuming a finite dislocation plane in an elastic half-space.
We used 15 continuous GPS and 5 campaign-surveyed GPS stations to capture coseismic displacement during the 2017 Bodrum-Kos Earthquake (Mw6.6). The cGPS data covers three days before and three days after the mainshock. Obtained coseismic surface displacements at 20 sites are analyzed to characterize source parameters and slip distribution of the 2017 Bodrum-Kos Earthquake (Mw6.6).
The Gulf of Gökova is one of E-W trending asymmetric depression with 120 km long and widens westward from 5 to 30 km. It is developed on the Lycian nappes, filled with Plio-Quaternary terrestrial and marine sediments (Gurer, Sangu, Ozburan, Gurbuz, & Sarica-Filoreau, 2013) and is bordered by Datça Peninsula to the south, the island of Kos to the west, and Bodrum Peninsula to the North. The northern margin of the Gulf is controlled by the Gökova Fault Zone (GFZ), is seismically one of the most active structures in SW Anatolia. It is a E-W to NE-SW trending broad arc-shaped zone that formed the northern breakaway fault of the Gökova Gulf ( Figure 1). Along much of its onshore lengths, GFZ has well-developed surface geomorphological expressions and geological features such as fresh fault surface that observed on triangular facets. Field-based studies suggest that GFZ is defined by south-dipping, high angle (70°-85°) normal faults (Gurer et al., 2013). Several marine-based

Seismotectonic setting
The Aegean region, is seismically one of the most active and rapidly extending regions on the Earth, has been deformed under the control of an N-S extensional tectonic regime at a rate reaching up to 30/40 mm/yr since the Pliocene (Bozkurt, 2001;Brun et al., 2016;Dewey & Şengor, 1979;Jackson & Mckenzie, 1984;Jolivet & Brun, 2010;Kaymakci, 2006;Kocyigit & Deveci, 2007;Le Pichon, Chamot-Rooke, Lallemant, Noomen, & Veis, 1995;Sozbilir, Sari, Uzel, Sumer, & Akkiraz, 2011;Yilmaz et al., 2000). There are two basic mechanisms explaining the extension of the Aegean region. The first explanation is built on the southwestern escape of the Anatolian plate (McKenzie, 1972) in response to convergent movement of the Arabia-Africa and Eurasian plates along the Bitlis collision zone (Dewey & Şengor, 1979). An alternative explanation is based on the framework of the northward-moving Arabian plate and the rollback of the Hellenic subduction zone where the African lithosphere is subducted below the Aegean Sea (McClusky et al., 2000). This extensional field resulted in approximately geophysical surveys have been carried out in the Gulf of Gökova in order to delineate fault systems of this region (Iscan, Tur, & Gokasan, 2013;Kurt, Demirbag, & Kuscu, 1999;Tur, Yaltirak, Elitez, & Sarikavak, 2015;Ulug, Duman, Ersoy, Ozel, & Avci, 2005). According to the results obtained from these surveys (Figure 1), the GFZ runs E-W parallel to the coast in the eastern part of Gökova Gulf from Akyaka to Ören, where it makes a left bend and continues west-southwestward that splay into several submarine fault strands displaying an anastomosing pattern, where there is a number of second order faults between the seismic segments ( Figure 2). Thus, in the Gulf of Gökova, the GFZ begins to lose its single fault line character and splays into a complex fault pattern. The complexity of the GFZ in the Gulf may actually be related to the interaction between deep-seated strikeslip faults and shallow-seated normal faults which characterize the SW Anatolia neotectonic regime. According to geological markers observed in this study, the total offset of GFZ is about 1000 m since the Plio-Quaternary. This suggests a slip rate of 0.2 mm/yr.   (Kalafat & Horasan, 2012). The last destructive earthquake struck at 1:31 am local time near the Kos (Greece), and Bodrum (Turkey) with the magnitude of 6.6 (KOERI, 2017).

The 20th July 2017 Bodrum-Kos earthquake
20 July 2017 Bodrum-Kos earthquake (Mw6.6) occurred at 01:31 (UTC) and ruptured a fault section along the western Gökova Gulf, between the cities of Bodrum and Kos ( Figure 2, Table 2). The mainshock resulted in two fatalities in the Kos island. It caused also some damage due to the tsunami along the southern coast of Bodrum Peninsula as well as the northern coast of Kos island. 13 cm wave-height was measured at Bodrum sea-level station. After the earthquake, 30-40 cm wave-height was observed on the Bodrum shores. The tsunami caused floods in some areas and reached heights of up to 1.9 m.
After that mainshock, about a dozen of aftershocks were recorded throughout the Gökova Gulf ( Figure 2, Table 3). The global focal mechanism solutions suggest that an approximately E-W trending fault generated the earthquake with a pre-dominant normal-type slip. The northern border of the Gökova Gulf, where the epicenters of earthquakes are aligned, is seismotectonically controlled by GFZ.
After the 2017 Bodrum-Kos mainshock (20 July 2017, Mw6.6, KOERI, 2017), the data dated from 19 to 23 July 2017 by 20 stations of the networks near the Bodrum-Kos earthquake epicenter was obtained immediately. More than 3000 aftershocks were recorded by the KOERI in the first 3 days ranging between magnitudes of 2.0 and 5.0 ( Figure 3).
2-day data including pre-seismic and coseismic deformation from 20 stations were recorded at 30s intervals in RINEX format. The precise coordinates of the stations were estimated using GAMIT/GLOBK software. To process the data with GAMIT software, the rapid orbit information, earth rotations parameters and antenna information were obtained from Scripps Orbit and Permanent Array Centre (SOPAC). Moreover, the antenna phase center was derived according to the height-dependent model. During the analysis, LC (L3), which is the ionosphere-independent linear combination of the L1 and L2 carrier waves and the
The combined geodetic network consists of 15 continuous and 5 campaign-surveyed GPS stations ( Figure  3 Tiryakioglu et al., 2013). In this process, the daily coordinates were estimated from 15 IGS   Significant displacements have been observed, in particular for the stations near the epicenters (e.g. stations YALI, ORTA, TRKB, TGRT, DATC, BODR and DIDI). Uncertainties in coordinate differences were determined using standard error propagation. The observed horizontal and vertical displacements and uncertainties are illustrated in Figure 5 and Table 3.
As can be seen from the detected coseismic displacement in Table 3, the North components of these stations had more coseismic displacement than the East component. The coseismic displacements in the North component lie in the range of 19-160.2 mm, with uncertainties of 2.7-3.8 mm. Significant displacements in the East component lie in the range from 9.2 to 39.4 mm, with uncertainties 2.4-3.5 mm. There was no significant

Coseismic displacements
Coseismic displacements are tightly correlated with the time series models. Short-term and long-term solution for the displacement are clearly exposed using continuous GPS data (Aktug et al., 2010;Tiryakioglu et al., 2017). The long-term evaluation is able to inform about the station movements which can be linear, periodic, irregular or episodic. On the contrary, short-term analysis is very conservative and stinted but it has more valuable information.
In this study, we analyzed the short-term daily solutions of continuous GPS in terms of the static coseismic displacements based on a scale in the Bodrum-Kos earthquake.
Because the earthquake occurred at 2017-07-20 01:31:12 (UTC), the estimation of the daily solution of the GPS sites on that day (DoY 201) did not include the data after the time of the earthquake. GPS data of MUMC station, one day before and after the earthquake, was not available. The MUMC station had a linear trend before the earthquake, which continued for approximately 1.5 years. The missing coordinates of MUMC station were estimated using the linear functions.
By comparing to the station position of the GPS sites from daily solutions before and after the Bodrum-Kos earthquake (DoY 201 and DoY 202), we obtained the coseismic displacement (Table 4).  direction, respectively and with respect to the reference point at 36.956n 28.370e. the depth is given from the surface. the moment is included as an auxiliary parameter since as the slip is fixed, the moment is not an independent parameter but instead a function of length and width.
geometry and slip rates, we adopted a hybrid optimization scheme which involves both global and local optimization. The details of the optimization strategy can be found in (Aktug et al., 2010). The objective function for the optimization was defined as the weighted residual sum of squares between the observed and the modeled displacements. While the relation between the surface displacements and the fault geometry is nonlinear, the slips on the dislocation are linearly related to the surface displacements. Therefore, we followed a two-step approach in the inversion. First, we inverted the coseismic displacements for the fault geometry assuming a uniform slip over the initial fault geometry. In the second step, we estimated the slip components fixing the fault geometry found in the first step. The inversion method should be able to escape from the local minima and should be efficient. Our hybrid approach employs a global optimization scheme which employs simulated annealing to avoid local minima (Kirkpatrick, Gelatt, & Vecchi, 1983). However, as with all global optimization methods, the simulated annealing is not as efficient as quasi-Newton methods near global minimum. Therefore, the results were refined using a BFGS (Broyden-Fletcher-Goldfarb-Shanno) algorithm (Fletcher, 1987). The observed and modeled displacements are shown in Figure 5.
As opposed to the focal mechanisms obtained through seismometers, GPS derived fault mechanisms are unambiguous. Furthermore, the length and the width of the ruptured fault can be determined directly along with the slip components on the fault surface. coseismic displacement in the Up components of each station. From the results, we found that the maximum coseismic displacement of 160.2 mm in the North components occurred at station BODR. The estimated displacement in the East components of station BODR coseismic displacement of -38.3 mm. Station YALI had a coseismic displacement of 153.0 mm in the North components and in the East components of station YALI lie within its uncertainty.
Station ORTA had the coseismic displacement of 100.3, -39.4 mm at the North and East components, respectively. Stations MUMC and TRKB had the coseismic displacement of 69.1 mm, 23.3 mm and 64.9 mm, -24.9 mm at the North and East components, respectively. Stations TGRT and DATC had the coseismic displacement of 24.8 mm, -9.2 mm and -33.7 mm, 12.2 mm at the North and East components, respectively. Station DIDI had the least coseismic displacement of 18.9 mm at the North component. Furthermore, no significant coseismic displacement was observed in the remaining stations (MUG1, FETH, AYD1, CESM, IZMI, RODO, ROD2, KALY, CAMK, MARM, KYCZ and KNID).

Inversion modeling for the fault parameters
The GPS derived coseismic displacements were modeled as the surface displacements of a finite dislocation in an elastic half-space following Okada (1985). The relation between the surface displacements and the fault geometry parameters is nonlinear and involves many local minima. To invert the displacements for the fault The mainshock hypocenter is located at the transition between the western and the central high-slip patches. Referring to the hypocenter, which represents the nucleation point of the rupture, the activated fault hosts an asymmetrical coseismic slip distribution as it terminated at ~20 km to the west while it extended ~45 km to the east. Its highest slip patches are concentrated in the center as well as in the western section of the rupture, which is located between the longitudes 27.2°E and 27.6°E. This verifies the shake maps, where the highest ground shake is observed in the east of Kos and the south of Bodrum (Figure 7 bottom, source: https://earthquake.usgs.gov).

Conclusion
Combined GPS network consisting of GPS stations is used to capture coseismic displacements associated with the Bodrum-Kos earthquake (Mw6.6). Since the earthquake occurred offshore, the determination of source parameters using geodetic data is indispensable. Significant displacements have been observed, in particular, for the stations near the epicenter. The stations at a distance range of 13.5-50 km to the mainshock epicenter recorded coseismic displacements at ranges of 19-153 mm and 9.2-39.4 mm for the North and East components, respectively. Results showed that the mainshock generated more in N-S directed movement compared to the E-W direction. This is well correlated with a N-S extending and EW striking normal-type focal mechanism . We have not observed a significant coseismic displacement in the Up components. Furthermore, no significant coseismic displacement have been observed in the stations that are more than 50 km away from the epicenter.
The mainshock activated a ~65 km long and 25 km wide inclined fault area which dips down to 13 km depth. The rupture has three high-slip patches, namely western, central and eastern patches, where the co-seismic slips reach up to 13, 26 and 5 cm, respectively. This slip pattern indicates that the pre-earthquake coupling, which is storing the slip deficit, occurred on these three patches. The highest pre-earthquake coupling has probably occured in central patch. Mainshock nucleation point remains between the western and central patches. The ruptured has an asymmetrical distribution of coseismic slips referring to the hypocenter as it terminated at ~20 km in the west while it extended for ~45 km to the east.
As opposed to the large on-land earthquakes, where surface observations are also available, earthquakes of moderate size are, in particular offshore requires a good observation coverage. KOERI and AFAD, which operates national seismic networks in Turkey, revised their magnitudes up to 0.5, which presents a large uncertainty for source parameters of such a large size earthquake. Accordingly, coseismic geodetic displacements are relatively small to be determined in survey-type measurements and a continuous monitoring is needed. In this The obtained fault geometry and slip mechanism are given in Table 1. In general, the vertical offsets obtained from GPS measurements are noisier than horizontal components. There are various reasons for this including the constellation of GPS orbits and the seasonal effects which are the largest in the vertical component. Therefore, the vertical offsets are often well defined and their uncertainties are optimistic. To account for the possible contamination of the vertical component, we made the separate inversions, one with 2D displacements and one with 3D displacements.
Depending on the benchmark geometry, the slips are mostly correlated with the length and width of the fault. In this respect, in poor coverages with few sites, narrower and shorter faults with larger slips could give very similar results with those of wider and longer faults with smaller slips in the inversion (Aktug et al., 2010). The slip parameters are linearly related to the surface displacements as opposed to the geometry parameters displacements. In this respect, they can be estimated separately in a second step after all geometry parameters are resolved. The trade-offs between the geometry parameters are given in Figure 6.

Inversion for coseismic slip
'Steepest Decent/Gradient' inversion method is used to investigate co-seismic slip distribution along the rupture plane (Wang et al., 2009). The method employs Okada's semi-infinite space model to simulate elastic Green's functions in order to converge to the observed co-seismic displacements. We defined a grid space of 80 × 25 km along the rupture plane framing presumable high-slip patches. For the final results, data-model correlation is above %85. The inversion results are summarized in Figure 7 (top).
Co-seismic slip occurs on roughly a 65 km long and 25 km wide fault area. Maximum depth remains at 13 km in depth-sectional projection due to the inclined character of the normal fault. There are three high slip patches along the fault plane, where the co-seismic slips ranging between 0 and 26 cm. The western patch has a length of 25 km hosting co-seismic slips reaching up to 13 cm. Its center is well defined by the slip pattern at 8 km depth. It ruptures the depth range of 4-12 km. There is a smooth transition to the neighboring central patch, which has a length of 16 km. The central patch hosts the highest co-seismic slips reaching up to 26 cm. Its center is located much shallower at 3 km depth. It ruptures the depth range of 0-10 km. The eastern patch has a length of 18 km. It is clearly separated from the central patch. It hosts the lowest co-seismic slips remaining at 5 cm. Its center is located much at 6 km depth and it ruptures almost the entire depth range, from the surface to the basement of the seismogenic zone. Overall pattern indicates that the pre-earthquake coupling, which is storing the slip deficit, occurred on three patches. study, one of the earliest examples of contributions of the new CORS-TR to the earthquake geodesy in Turkey is presented, in particular at a large size earthquake.
Combining GPS observations with surface geological data, we conclude the fault geometry from the surface to depths of about 17 km has listric characters. The main shock and aftershocks of the 2017 Bodrum-Kos earthquake suggest that the fault mechanism can be classified into two groups based on focal mechanism solutions. (1) NE-trending oblique-to strike-slip faults with moderatedepth hypocenters (up to 14 km) at the western end of the GFZ, (2) approximately E-W trending normal faults with shallow earthquakes (up to 17 km) at the middle and western section of GFZ.
The instrumental period earthquakes suggest a westward propagation along the GFZ. Source mechanism solutions for most of these earthquakes indicate that normal fault mechanisms with a small strike-slip component have been observed on the E-W-oriented normal fault systems in the Gulf of Gökova, which confirms that the extension is in a north-south direction.