Abstract
A timely, high-resolution earthquake catalog is crucial for estimating seismic evolution and assessing hazards. This study aims to introduce a deep-learning-based real-time microearthquake monitoring system (RT-MEMS) for Taiwan, designed to provide rapid and reliable earthquake catalogs. The system integrates continuous data from high-quality seismic networks via SeedLink with deep learning models and automated processing workflows. This approach enables the generation of an earthquake catalog with higher resolution and efficiency than the standard catalog announced by the Central Weather Administration, Taiwan. The RT-MEMS is designed to capture both background seismicity and earthquake sequences. The system employs the SeisBlue deep learning model, trained with a local dataset, to process continuous waveform data and pick P- and S-wave arrivals. Earthquake events are then associated and located using a modified version of PhasePAPY. Three stable RT-MEMS have been established in Taiwan: one for monitoring background seismicity along a creeping fault segment and two for monitoring mainshock-aftershock sequences. The system can provide timely information on changes in seismic activity following major earthquakes and generate long-term catalogs. The refined catalogs from RT-MEMS contribute to a more detailed understanding of seismotectonic structures and serve as valuable datasets for subsequent research.