Research Information System
Turk Deprem Arastirma Dergisi, vol.7, no.1, pp.90-100, 2025 (Scopus)
In analyses aimed at determining the dynamic parameters of earthquakes, the accurate detection of the arrival times of primary seismic waves (P and S waves) is a fundamental prerequisite for solving seismological problems. Studies based on these arrival times contribute to various research areas, such as understanding the Earth’s crust and mantle structure. Recent developments in machine learning—or more broadly and commonly known as artificial intelligence technologies—have made it possible to automatically detect earthquake waves from waveform data. Due to its complex tectonic structure, Türkiye has high seismic activity, as it is located at the intersection point of the Eurasian, African, and Arabian plates. In this study, the performance of a deep learning algorithm that automatically detects the arrival times of P and S phases in earthquakes occurred in Türkiye from 2013 to 2019 belonging to KOERI (Kandilli Observatory and Earthquake Research Institute) network was evaluated. The results show that machine learning can make more accurate predictions compared to traditional statistical methods and is effective in reducing human-induced errors. The findings of the study reveal that deep learning-based seismic phase detection algorithms trained with large databases can provide increased accuracy and speed in seismological analyses when adapted to local needs. It is recommended that future studies conduct comparative examinations of models also trained with local data and develop algorithms that do not require human intervention in phase detection.