Automated First-arrival Phase Picking on the 2022 Cianjur Aftershock Earthquakes Using Machine Learning

Abstrak

Abstract Rapid event identification and phase picking of arrival time are crucial in seismology. Therefore, an advanced tool is necessary to detect the phase of seismic waves capable of processing extensive data quickly, accurately, and efficiently. This study focuses on the aftershock of the Mw 5.6 Cianjur 2022 earthquake recorded by 19 3-C seismometers from November 22 to December 22, 2022, and evaluates the reliability of the first-arrival phase picking obtained from machine learning. We compare the results of the machine learning-based approach with the previous manual catalog. We employed PhaseNet, a deep learning-based seismic phase detection method, to identify the seismic phase of aftershocks, which was then integrated with the GaMMA earthquake associator. PhaseNet detected more than 13.000 P and S phases, associated with GaMMA for 647 events. We also used the Wadati diagram to analyze the reliability of the identified phases. Consequently, 440 events were located using NonLinloc. The average difference in time between Nonlinloc and catalog event estimations is 0.875 seconds. The average differences in latitude, longitude, and depth of the earthquake are 1.155 km, 1.283 km, and 2.315 km, respectively. This study shows that PhaseNet performs well and can detect more events than the manual catalog, but it still requires quality control from the user, as can be resolved simply using the Wadati diagram. Machine learning-based catalogs offer significant advantages in efficiency, seismic association, and location detection.

Identifier

• OpenAlex: https://openalex.org/W7131260187
• DOI: https://doi.org/10.1088/1755-1315/1590/1/012008