Combining Borehole Seismic Data and Deep-Learning Methods to Monitor Seismic Sequences in the Istanbul-Marmara Region, NW Türkiye

Abstract

The North Anatolian Fault Zone beneath the Sea of Marmara in NW Türkiye displays a 50% probability of an $M\ge 7$ earthquake within the next decades. The proximity of the fault to the Istanbul megalopolis highlights the need for high-resolution microseismicity monitoring to identify seismicity transients including pre-mainshock clustering. We analyze six 4.0 $\le$ M_L $\le$ 5.8 earthquake sequences in the eastern and central Marmara region from 2016 to 2023 and focus on the occurrence of seismicity immediately prior to the mainshocks. We combine borehole and surface seismic waveform data with deep-learning phase picking, automated phase association, and non-linear relocation to compile a comprehensive seismicity catalog. Compared with conventional processing, our workflow allows us to detect about eight times more events in time windows around the mainshocks in the area instrumented with borehole sensors. We then apply waveform cross-correlation, refined location techniques, and template matching to quantify waveform similarity and characterize seismic activity preceding moderate events. We observe a strong increase in waveform similarity and decrease in inter-event distance prior to a 2016 M_L 4.5 event in the Çınarcık Basin, consistent with earlier results from manual processing, and identify additional highly similar events preceding the mainshock. The 2019 M_L 5.8 Silivri sequence shows increased waveform similarity but more heterogeneous spatial clustering. We demonstrate that borehole seismic arrays are critical for resolving these subtle signals due to higher signal-to-noise ratios. Our workflow provides a robust basis towards near-real-time microseismic monitoring in the Istanbul-Marmara region and systematically identifying seismicity transients and localization processes.