Seismicity in the Sea of Marmara Obtained Using Machine Learning Algorithms

The Sea of Marmara represents one of the most critical seismic gaps due to its high fault slip rate (~20 mm/yr), the long interval since the last major earthquake (~250 years), and its proximity to densely populated metropolitan areas. Understanding the complexity of faulting and seismicity in this region is therefore essential. In this study, we utilize a convolutional neural network-based detection and phase picking algorithm (Mousavi et al., 2020) combined with a phase associator employing a grid-search location method (Zhang et al., 2019), significantly increasing the number of detected events using the same dataset as the Kandilli Observatory and Earthquake Research Institute (KOERI) data center (BDTIM) stations. Each waveform is manually reviewed to accurately distinguish real earthquakes from false positives. Furthermore, by incorporating data from AFAD and the local Prince Islands Real-Time Earthquake Monitoring System (PIRES), we construct an accurate and detailed seismicity map of the Sea of Marmara. Our results demonstrate that seismicity patterns can be greatly refined by integrating data from multiple networks and applying state-of-the-art methods for earthquake detection, location, and association. (This study is funded by TÜBİTAK Project No. 121Y407.)