Structure of the North Anatolian Fault Zone imaged via teleseismic scattering tomography

1 School of Earth and Environment, The University of Leeds, Leeds, LS2 9JT, 2 Department of Geological Sciences, University of Manitoba, Winnipeg, Manitoba, Canada, 3 Department of Geology and Geophysics, School of Geosciences, University of Aberdeen, King’s College, Aberdeen, AB24 3UE, United Kingdom, 4 Kandilli Observatory and Earthquake Research Institute, Department of Geophysics, Boğaçizi University, 34684 Cengelköy, Istanbul, Turkey, 5 Department of Geophysical Engineering, Sakarya University, Esentepe Campus, 54187, Sakarya, Turkey 6 COMET, School of Earth and Environment, The University of Leeds, Leeds, LS2 9JT


S. Rost et al.
into the upper mantle, we exploit the scattered seismic wavefield following the P-wave arrivals 22 of teleseismic events (Frederiksen & Revenaugh, 2004), allowing us to resolve the fine scale 23 structure of the lithosphere in the study region using data from the 18-month DANA deploy-24 ment (DANA, 2012) deployed across the NAFZ in the region of the 1999 ruptures (Fig. 1a). The 25 P-wave coda contains energy from P-to-P and P-to-S scattering at small-scale heterogeneities 26 along the ray-paths. Structure can be recovered from the scattered seismic energy through mi-27 gration approaches ranging from common-conversion-point or common-scattering-point stacking 28 (e.g. Dueker & Sheehan, 1997) to full depth migration (e.g. Ryberg & Weber, 2000). Here we are 29 using a tomographic waveform approach based on linear inverse theory of the scattered wavefield 30 (Ji & Nataf, 1998;Frederiksen & Revenaugh, 2004). 31 We find that the two strands of the NAFZ evident in the shallow structure coincide with main 32 interfaces and interface terminations throughout the crust and into the upper mantle indicating that 33 the fault zone structure may extend to depths of at least ∼75 km in this region. We find evidence 34 for small-scale variation of structure in the vicinity of the strands that might indicate the detection 35 of heterogeneity related to past deformation along the present day fault.  been attributed to partial melts or pore fluid flow from the upper mantle beneath the NAFZ.

107
We apply the teleseismic scattering tomography approach by Frederiksen & Revenaugh (2004) 108 to the DANA dataset to resolve the small-scale structure beneath the array. The scattered seis-109 mic wavefield is more sensitive to short-wavelength variations in material properties than is the 110 path-integrated sensitivity of transmitted phases such as used in e.g. seismic traveltime tomogra-111 phy. The P-to-p and P-to-s scattered energy in the coda of teleseismic P-waves travelling along  given by: The wavefield can then be divided into a primary (background) and scattered component (u = 144 u 0 + δu) with the unperturbed wavefield satisfying the unperturbed wave equation Assuming that the scattered wavefield is much weaker than the unperturbed wavefield this gives 146 the first-order Born approximation by discarding higher-order terms: with Q i being a term of the unperturbed wavefield and the perturbed model parameters which is 148 given as equation 13.22 in Aki & Richards (2002).
with u 0 being a solution for the unperturbed medium. solving the forward problem of the waveform inversion. 159 We assume the incident P-wave to be planar (Fig. 2) with a known slowness vector, a condi-160 tion well met for teleseismic records. The scattered wavefield is derived from the seismic obser- with I being an MxM identity matrix and λ a weighting factor, representing uniform damping. We    Fig. 4). This velocity model is derived from seismic exper-  Fig. 4a) with the first arrival muted. Scattered phases can be seen coherently across the 270 traces. The synthetic traces (Fig. 4c) show similar structure although clearly are not able to cap-271 ture the full complexity of the data due to the simplicity of the model (Fig. 4b). Noise is added to    amplitude variation Gaussian noise compared to the direct wave amplitude to produce this noisy 274 synthetic dataset (Fig. 4d).

275
The synthetic model is parameterised with 5 km cell spacing horizontally and 2 km vertically.   of 6 km starting at 78 km (Fig. 5 a and c). No density variation was added to the model. The from localized perturbation, the recovered model will be a band-pass filtered version of the input. 284 We apply a Wiener optimum filter to minimize the effects of the inversion process, mainly to 285 reduce sidelobes to aid interpretation. The optimization filter, as described for example by Gubbins

286
(2004), is obtained by minimising the residual between the desired output g t (Fig. 5a,c) and the 287 signal obtained by convolution of the filter f 0 t with the actual output x t (Fig. 5b,d) 288 The effect of the inversion and the filter terms acting on a single trace of the synthetic model 289 are shown in Supplemental Figure S1. Although the input model in this test does not contain any 290 density (ρ) heterogeneity, Fig. 5e shows that the inverted model for the density structure is affected 291 by cross-talk between the different parameters (more examples given in Supplemental Figure S2).

317
The results of the tomographic scattering inversion of the DANA dataset are shown in Fig. 6 and 318 Fig. 7. Slices in Fig. 6 and 7 were extracted from the three-dimensional inversion volume along 319 North-South (Fig. 6) and East-West (Fig. 7) profiles at locations shown in Fig. 1. The locations 320 of the profiles were chosen to be in similar locations to those shown in Fig. 6 of Kahraman et al.

334
Profiles for V P (Fig. 6 a) and V S (Fig. 6 b) have been extracted along a longitude of 30.20 • E.

335
Areas with limited resolution as determined from the recovery tests (Fig. 5) have been masked 336 in this profile in transparent grey. The V P profile (Fig. 6a)   The eastern North-South profile at 30.51 • E (Figs. 6c and 6d) shows more structure, especially in 364 the crust, than the western profile despite the close proximity of the two profiles. 365 We observe a strong, fast V P anomaly at a depth of ∼34 km terminating halfway through 366 the Armutlu block and re-emerging at a depth of ∼42 km just north of the northern strand in 367 the Istanbul zone. In V S we observe a more continuous structure with a high velocity anomaly at 368 ∼36 km depth in the south stepping to ∼42 km at ∼40.7 • N coinciding with the northern strand.

369
The V P anomaly is weak in the Armutlu block on this profile and seems to terminate at 40.6 • E, while the V S anomaly is more continuous, but also weakens in this region. The amplitude variation 371 of these anomalies cannot be explained by the limitations of the sampling (see Fig. 5).

372
Especially striking in this profile is the complex V S structure in the Sakarya Zone down to 373 depths of about 30 km manifesting as series of fast and slow anomalies between ∼10 km and 374 32 km (see supplemental Figure S7). The V P structure is similar but weaker than V S . The structure 375 terminates abruptly at the southern strand with little crustal structure in the Armutlu block. The  Similar to the western profile we identify a slow anomaly in both V P and V S at depths of 379 ∼56 km and ∼52 km, respectively. The V S anomaly seems to show more complexity. We identify 380 a slow anomaly at ∼76 km depth in the Sakarya zone in V P which cannot be identified in V S . This 381 anomaly seems to terminate at the southern branch. Fast anomalies are detected at ∼92 km in V P 382 and V S across the profile with shallower fast anomalies for V P and V S at ∼76 km depth beneath 383 the Istanbul zone and the Armutlu block. In V P there is evidence of this interface splitting into a 384 deeper interface deepening to ∼102 km across the southern strand.

397
In contrast to the Sakarya Zone, the Armutlu Block (Figs. 7c and 7d for V P and V S , respectively) 398 shows more structure down to depths of 40 km. A fast anomaly at ∼40 km terminates around 399 30.4 • E and appears as shallow as 30 to 32 km further east in V P . V S also shows the termination 400 but a less pronounced step. The step around 30.6 • E seems to coincide with the profile moving  The strong Moho signal is underlain by a slow anomaly around 52 km depth again terminating 416 around 30.6 • E for V P . V P shows a fast anomaly at ∼74 km depth, which like the Moho signal in 417 this block, terminates at about 30.6. • E; the corresponding structure in V S is weaker and discontin-418 uous. A strong fast anomaly at ∼92 km depth can be seen in V P and V S , but again the V S structure 419 is complex.  (Fig. 1). The depth slices through 427 the model shown in Fig. 6 and 7 show strong changes between the two north-south trending profiles 428 despite their close proximity. Interpreted NS sections are shown in Fig. 8  show western and eastern stacks in Fig. 9 for both V P and V S .

431
Comparing the individual slices and the stacked velocity-depth profiles shows that many features 432 are coherent along stretches of the profile, but can change on short scale-lengths.  In the east the Moho seems much weaker and discontinuous across all three tectonic blocks.  Fig. S14, S15), which is more pronounced in V P but also 456 detectable in V S , indicating strong contrasts in crustal structure between these two blocks. to the neighbouring tectonic units, which could be related to the absence of metamorphism and the 485 lack of major deformation (Okay, 1989).   Fig.   517 6c,d). In general the AA shows almost no heterogeneity in the crust. Fig 6 a,b) shows evidence for 518 heterogeneity at crustal depths coinciding with the surface expression of the NNAFZ.

519
The Moho step detected in the eastern profiles (e.g. Fig 6c, d)) seems to coincide with the  We show that scattering tomography in conjunction with dense recordings of the seismic wave-