Detection of P and S Wave Phases by Machine Learning using Northwestern Türkiye Local Seismic Network Data

In regions with intense seismic activity like earthquakes, rapid detection and resolution of earthquake parameters and understanding seismic activity and mechanisms are important in terms of reducing possible risks. Since this process is left to the knowledge and experience of users to a great extent in the solution stage, human errors in detection of seismic wave phase arrival times may negatively affect the reliability of model studies. In this study, machine learning, which has been successfully applied to data in various seismological fields, was applied to earthquakes occurring in the Biga Peninsula, encompassing the most complicated tectonic elements of the north-western Aegean region, has high window seismicity. Results were evaluated using the waveform database for events recorded by local (COMU—Çanakkale Onsekiz Mart University) and national (KOERI—Kandilli Observatory and Earthquake Research Institute, AFAD—Ministry of Interior Disaster and Emergency Management Presidency) seismic networks observing activity linked to tectonism in the region under consideration with the originally trained model of the PhaseNet machine learning algorithm. Data contains 918 earthquakes recorded at 118 stations from May 2020 to the end of 2021. Compared to classic methods, the machine learning model used in the study provided more accurate results for detecting P and S wave phases. Also, epicentre calculations based on machine learning algorithm appear to be in better spatial agreement with the distribution of active faults than calculations based on handpicks. Although the original model of PhaseNet has not been trained with local data from Türkiye, study shows it is possible to get meaningful results by making adjustments on the algorithm or applying signal processing techniques on the data. Study suggests that enhancing machine learning algorithm with local training data can improve phase detection accuracy and epicenter prediction in seismic studies. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024.