Fault zones all around the world are formed by tectonic plate motions and are natural boundaries between blocks. They are generally locked to the bottom of the seismogenic layer and cannot slip freely compared to the velocities within the blocks because of the friction between rocks. Therefore, movement in these regions is generally minimal and causes earthquakes when the motion of the blocks overrides the friction force. After discharge (earthquake), faults begin to accumulate strain and this cycle continues until the next earthquake (Reid, 1910; Yavaşoğlu, 2011).

The NAF (North Anatolian Fault) is a tectonic plate boundary between the Anatolian and Eurasian plates. It slowly moves at ∼20 mm yr-1 to the west by the overthrusting Arabian Plate from the south and compresses the plate motion with the help of the massive Eurasian Plate in the north. Those tectonic forces constitute the North Anatolian Fault, which lies between the Karliova triple junction from the east and the Aegean Sea to the west for almost 1200 km. The width of the fault trace ranges between 100 m and 10 km. The Anatolian Plate moves 20–25 mm yr-1 to the west relative to the Eurasian Plate. There are velocity variations along the fault; that is, the west region moves faster than the eastern part and is a right-lateral strike-slip fault (Fig. 1) (Ketin, 1969, 1976; McClusky et al., 2000; Cakir et al., 2005; Şengör et al., 2005; Reilinger et al., 2006; Yavaşoğlu et al., 2011; Bohnhoff et al., 2016).

Earthquake mechanisms might have different characteristics in some regions. Faults may move freely without an earthquake, and this motion is reported at some unique places like the Hayward fault (Schmidt et al., 2005), the Superstition Hills fault (Wei et al., 2011) and the İsmetpaşa segment (Cakir et al., 2012), which can be observed from the surface (Ambraseys, 1970; Yavasoglu et al., 2015). This phenomenon is called “aseismic creep” and may occur in two different ways. If the creep takes place to the bottom of the seismogenic layer and the surface velocities are equal or close to the long-term tectonic velocities, there will not be enough strain accumulation for a large-scale earthquake (Şaroğlu and Barka, 1995; Cakir et al., 2005). On the other hand, if that free motion is not observed to the bottom of the seismogenic layer or observed surface velocities are smaller than the tectonic velocities, strain will accumulate to a final earthquake (Fig. 2) (Karabacak et al., 2011; Ozener et al., 2013; Yavasoglu et al., 2015). Also, aseismic creep in a region may occur continuously or fade out after some period (Kutoglu et al., 2010).

The NAF is reported to have segments that have shown aseismic creep since 1970 at İsmetpaşa, with a more recent discovery at Destek (Ambraseys, 1970; Karabacak et al., 2011). Aseismic creep at İsmetpaşa is reported to occur along ∼70–80 km, from Bayramören (east) to the Gerede (west) (Fig. 3). It was discovered at the wall of the İsmetpaşa train station in 1970, and several minor and large-scale studies have monitored the area since then (Table 1). That segment has hosted three destructive earthquakes (1943 Tosya Mw=7.2, 1944 Gerede Mw=7.2, 1951 Kursunlu Mw=6.9) that may have triggered or affected the creep (Şaroğlu ve Barka, 1995; Cakir et al., 2005; Karabacak et al., 2011; Kaneko et al., 2013) (Fig. 4).

In addition, creep at the Destek segment was reported in 2003 on a field trip around the region. Unlike the İsmetpaşa segment, there are few studies at this segment, and also the length of this segment is unclear. The 1943 Tosya earthquake, which is reportedly the biggest earthquake in the segment, affected this area (Karabacak et al., 2011) (Table 2).

All the studies around those segments indicate the continuity of creep, but the results are inconsistent and cannot clearly identify whether that event has an increasing trend or not. Most research (Ambraseys, 1970; Aytun, 1982; Eren, 1984; Altay and Sav, 1991; Deniz et al., 1993; Kutoglu et al., 2008, 2009, 2013; Karabacak et al., 2011; Ozener et al., 2013; Bilham et al., 2016) is generally on a microscale and focused on İsmetpaşa or a network near this village with geodetic methods, while others focus on a macroscale with InSAR (Deguchi, 2011; Fialko et al., 2011; Köksal, 2011; Kaneko et al., 2013; Cetin et al., 2014; Kutoglu et al., 2013), which needs a ground truth (Figs. 5, 6).

Those results cannot reveal the creep trend clearly. In addition, a ground network is required to exhibit the fault characteristics clearly along the segments. For this reason, we established a ground network to form profiles around those segments and made three observations annually from 2014 to 2016.

Designing a monitoring network around tectonic structures is always related to the geological characteristics and fault geometry, which includes the locking depth and earthquake-related motions (coseismic movements) through the fault. Previous studies indicate that the velocities for the stations distant from the fault plane can be used to derive long-term plate velocities, while nearby station velocities are suitable to detect the locking depth of a fault (Taskin et al., 2003; Halıcıoğlu et al., 2009). In addition, the velocities of the observation stations gradually decrease when their locations approach the fault plane. Another factor is the number of stations, and this is related to the fault length and width, but the station locations perpendicular to the fault plane must not exceed (±1/√3) of the locking depth. Also, some research has specified this limit as double the depth (Taskin et al., 2003; Kutoglu and Akcin, 2006; Kutoğlu et al., 2009; Halıcıoğlu et al., 2009; Poyraz et al., 2011; Bohnhoff et al., 2016). For this purpose, the following equation is used in general to obtain the proper distances of the observation stations from the fault plane: 1Vx=VTπarctan(xD), where V is the fault-parallel velocity, VT is the long-term tectonic plate velocity, x is the distance to the fault plane and D is the locking depth of the fault (Halıcıoğlu et al., 2009).

The location of the stations may vary according to the geological surface elements, but they are generally established on both sides of the fault to form a profile on each block to obtain surface velocities (Yavasoglu et al., 2015).

The geologic structure at the tectonic block boundaries and fault plane geometry also affect the tectonic behavior. To better understand this mechanism, an established network around the fault zone is observed with different techniques periodically or continuously. The variation of the observations is a clue to detect those amplitudes, and GPS is the most common technique for that kind of study. This technique is very effective and efficient to collect data from ground stations established around faults (Poyraz et al., 2011; Aladoğan et al., 2017).

Profiles that intersect the fault plane vertically are used to estimate the locking depth. However, in regions like İsmetpaşa and Destek, there is an additional locking depth deduced from the previous studies, which indicates that the creeping layer of the seismogenic zone does not reach the bottom but is around 5–7 km of depth in those areas (Kaneko et al., 2013; Ozener et al., 2013; Cetin et al., 2014; Bilham et al., 2016; Rousset et al., 2016). For this reason, the aseismic layer's attenuation depth is another crucial element to understand the creeping mechanism (Fig. 2). Also, considering the 5–7 km depth value with Eq. (1), station locations are chosen as 3 and 10 km on both sides of the fault to form profiles, while the NAF general locking depth is around 15 km (McClusky et al., 2000; Poyraz et al., 2011; Bohnhoff et al., 2016).

Before the three observation periods, a network was planned to form four profiles at İsmetpaşa and one profile at the Destek segment, including surrounding continuous GPS stations (real-time kinematic continuously operating reference stations – RTK CORSs) (Fig. 7). The aim of this study was to monitor this network periodically to calculate the velocity field and combine the results with CORS station velocities to estimate the creep ratio within the İsmetpaşa and Destek segments (Yavasoglu et al., 2015).

While establishing the network, 3 and 10 km on both sides of the fault were first considered, but some minor changes took place according to the geological structure of the area. In addition, another profile between the second and third profiles was formed with the suitable location of two unplanned stations. Finally, there are five profiles within ∼70 km along İsmetpaşa and one profile along Destek.

Observations were completed around July and August for 3 years using relative geolocations based on carrier-phase observations with the GPS technique (Table 3). Force centering equipment and GPS masts were used when necessary. The first campaign was on the 235–238 and 241 GPS days in 2014, the second was on 215–221 GPS days in 2015, and the last one was between 210 and 220 GPS days in 2016.

After the first campaign, KZDY station was damaged and removed from the rest of the project. Raw data collected for a minimum of 8 h at each station for the rest of the project were evaluated with GAMIT/GLOBK software (Herring et al., 2015a, 2015b) first, then the results were used as input to the block modeling software TDEFNODE (McCaffrey, 2002, 2009). A total of 63 stations (22 campaign, 30 surrounding RTK CORS, 11 IGS) are used in this network to monitor the İsmetpaşa and Destek segments and the remaining region between them (Table 4).

GPS data for cGPS and the IGS station data were processed on campaign observation dates. In addition, observations for those stations during January (for 7 d) were included at the GAMIT/GLOBK step to increase the stabilization of the designed networks.

The networks were linked to the ITRF 2008 global coordinate system by using surrounding IGS sites (Table 5) (Yavaşoglu et al., 2011; Herring et al., 2015a, b). After the transformation with GLOBK, the root mean square (RMS) of the stations was only 0.7 mm yr-1.

Results show that the velocity of the stations located on the Anatolian Plate range from 15 to 20 mm yr-1 (Fig. 8), which is similar to previous studies (McClusky et al., 2000; Reilinger et al., 2006; Yavaşoglu et al., 2011).

The GLOBK results for all of the station velocities are used as input for block modeling to predict the aseismic creep ratio within the fault plane in the predefined segments (Table 6, Fig. 9).

The aseismic creep ratio is estimated by interpolation through the profiles using surface velocities except the third profile (Table 7).

The GAMIT process indicates abnormal deformation for the ORMN and KDZU campaign stations, so their data are removed from the block modeling step. Additionally, the creep estimation for that profile is unfeasible. Actually, this is not a drawback for block modeling because the remaining station velocities are all used to model the region uneventfully.

With the calculated surface velocities, the Destek segment also has a creep trend through the campaign period. The estimated creep rate in this study according to GLOBK results is 10.6±3.1 mm yr-1 in this region, and it indicates aseismic creep similar to recent studies (Fraser et al., 2009; Karabacak et al., 2011).

Station velocities are suitable to predict surface and block motions locally. On the other hand, observations inside the blocks provide adequate long-term block velocities and rotations with high precision. Blocks generally demonstrate a regular movement, but their motion differs at their boundaries from this overall velocity. They cannot move freely around the faults because of the friction of rocks, which is generally interpreted as a slowdown that may eventually result in a total lack of movement (Fig. 10). That difference in velocity is called “slip deficit” and causes earthquakes after the friction threshold is surpassed (Kutoglu and Akcin, 2006; McCaffrey, 2014; Yavasoglu et al., 2015).

Slip deficit represents that expected velocities of the blocks pass through some deformations regarding the geological structure when approaching the fault zone and frequently decreases. This is based on the geometry of the fault plane, which can only be predicted based on surface velocities. In that context, the TDEFNODE software was used in this study to predict the fault plane locking interaction regarding the depths, which calculates variations of the block motions, strain accumulation within the blocks, and rotations through interseismic or coseismic periods (Okada, 1985; McCaffrey, 2009; Yavaşoğlu, 2011).

Basic input for the software includes GPS velocities, blocks with Euler poles, fault geometry and locking depth. Interacting blocks are represented as elastic blocks and are assumed to have elastic deformation because of their rotation around Euler poles. All of the defined system is assumed to float inside a half-space in which one of the blocks is fixed and has zero strain or movement. Fault geometry is defined by the user with nodes, and their locking ratios (phi) can be defined manually or as a function of depth (Fig. 11). Then, the software predicts the underground velocities based on the routines of Okada (1985) and estimates the surface velocities according to the defined values. Fault geometry estimation is the key feature to minimize the difference between observed and predicted surface velocities with the help of the χ2 test result, which represents the accuracy of the entire model (McCaffrey, 2002; Aktuğ and Çelik, 2008; Yavasoglu et al., 2011).

TDEFNODE is not only used for interacting blocks for interseismic strain accumulation, but also for faults that are partially or fully free-slipping like aseismic creep. The software model is suitable to define the locking ratios of all nodes independently for (0–1). (0) represents the fault at that node freely slipping, and (1) is for a fully locked node. That allows the user to define the fault plane with layers by using depth contours and to predict the fault plane if those layers are partially or fully locked (http://www.web.pdx.edu/~mccaf/defnode.html, last access: 20 December 2018).

Aseismic creep is an earthquake-free motion along the Earth's surface, but in some cases it is difficult to detect whether this motion is a free-slipping event or an interseismic movement. Thus, the observation network around the fault plane should be planned carefully regarding the ±3–10 km station locations mentioned before (Fig. 12).

During the TDEFNODE process, one of the tectonic blocks should be chosen as fixed to estimate the fault parameters. Therefore, the Euler pole is defined as (0, 0, 0) for the Eurasian Plate and (30.7, 32.6, 1.2) for the Anatolian Plate. The values represent latitude, longitude and angular velocity, respectively (McClusky et al., 2000).

Figure 12 demonstrates the suitable distances to detect aseismic creep. If an aseismic creep is suspected on a fault plane, then the optimum locations for the observation stations should be around 3 and 10 km on both sides of the fault and can be resolved from the interseismic movements. Therefore, the observation stations mentioned before are established around the fault as profiles to detect these discrepancies and to detect the main locking depth of the fault as well as attenuation depths for the creep event. Their locations are suitable to evaluate both the creeping ratios and locking depths of the faults.

Station velocities all around the region indicate the motion of the Anatolian Plate relative to the Eurasian Plate. Movements range between 15 and 24 mm yr-1 inside the southern plate where the northern motion reaches down to ∼1 mm yr-1. That result is consistent with previous studies (∼24±2 mm yr-1) (McClusky et al., 2000; Reilinger et al., 2006; Yavasoglu et al., 2011). In addition, model locking depths and results are similar to a more recent study with InSAR, which indicates that the locking depth of the fault at the İsmetpaşa segment is around 13–17 km and long-term tectonic movement is about 24–30 mm yr-1 (Hussain et al., 2018).

Special features of the inspected segments are revealed by the network established near the fault plane. Regarding the surface velocities of the observation points, profiles on both the İsmetpaşa and Destek segments indicate movements. These range between 10.1–14.9 and 10.6 mm yr-1 for the İsmetpaşa and Destek segments, respectively.

Additionally, the modeled fault plane evaluation for observed and calculated station movements demonstrates similar results as the locking depths of both creeping and seismogenic layers (Fig. 13). Station velocities on the south of the NAF are faster than the north end as expected (Fig. 14). Regarding the long-term geodetic block motions, modeled weighted locking ratios indicate 13.0±3.3 mm yr-1 of aseismic creep over the İsmetpaşa segment. That movement does not include the whole fault plane, and thus the creeping layer seems to slip freely to 4.5 km depths from the surface and decays between 4.5 and 6.75 km. The seismic data and previous studies (Cakir et al., 2005; Yavaşoğlu et al., 2011; Hussain et al., 2018) indicate that the locking depth over the fault is ∼15 km. This result demonstrates that the fully locked portion of the fault plane is between 6.75 and 15 km, which is supported by the χ2 test result (1.00).

The Destek segment also has similar results for the observed and modeled velocities (Fig. 15). The surface velocities for the profile (006) at this region indicate velocity differences (Fig. 16). In addition, the modeled fault plane indicates that the creeping segment is limited to 4.3 km of depth from the surface and decays linearly between 4.3 and 6.0 km. The remaining layer of the fault seems to be fully locked down to the seismogenic layer. The free-slipping portion has 9.6 mm yr-1 of motion, which is similar to the estimated surface velocities (10.6±3.1 mm yr-1). The χ2 test result (1.01) and the seismic data confirm the accuracy of the model.

Moreover, paleomagnetic data indicate a predominantly clockwise rotation of the blocks bordered by the faults between the İsmetpaşa and Destek segments. Examining the results with this study promotes that behavior with the GPS field of the region, especially on the Anatolian side of the NAF (Figs. 13, 15) (İşseven and Tüysüz, 2006).

We find no clear evidence for attenuation at both segments. On the contrary, there is a slight increase at İsmetpaşa and almost 50 % of an increase at Destek relative to previous studies. The frequency of this phenomenon at both segments is unclear, but the results of Hussain et al. (2018) support the argument that the creep event will continue until the next large-scale earthquake.

The NAF is reported to have had a creeping phenomena at İsmetpaşa since 1970 that has been observed with different techniques for a long time period, with a recent discovery at Destek. All previous studies concentrated on whole segments or at least some regions along those segments. With this study, a GPS network covering the whole Anatolian region along the NAF is established for the first time, and the results for the velocity area were used as input for block modeling. Also, the first GPS network covering the Destek segment was established during this study.

The network design and location of the observation points were distinguished according to the main locking depth of the NAF and the attenuation depth for the aseismic creep event. Model results show similar outcomes for both the İsmetpaşa and Destek segments; the locking depth for those segments is ∼15 km, and attenuation for the creeping layer depths varies between ∼4 and 6 km.

Through all the models, results for this study indicate that the creeping behavior still continues at both the İsmetpaşa and Destek segments, with a ratio of 13.0±3.3 and 10.6±3.1 mm yr-1, respectively. Block modeling and seismic data indicate that the creeping segment does not reach the bottom of the seismogenic layer (∼15 km) and is limited to some depths, which may not prevent a medium- to large-scale earthquake in the long term. In addition, we found no evidence for the attenuation of aseismic creep. Also, the frequency of this movement at İsmetpaşa is unclear and it is not possible to predict the aseismic creep ratio precisely for the long term, but results might indicate a small increase in the trend relative to previous studies in the region.

Additionally, the creeping ratio seems to have increased almost 50 % at the Destek segment since previous studies, which might indicate a relief at that segment. However, according to the model, aseismic creep is limited to some depths (∼6.0 km) and the creep ratio is smaller than the long-term block movements. The increasing trend is not sufficient to release all the strain in that segment. This might indicate strain accumulation on both ends of the segment.

The network established by this study should be monitored periodically for a precise assessment of the frequency of aseismic creep, which may include possible clues for a clear fault plane definition and earthquakes. In addition, results indicate that this creep event will be monitored until the next earthquake, which might reveal valuable information for fault zone layout models.