Presentation Information
[2101]Deep Learning Approach to Detection and Spatial Characterization of Microearthquakes in a Geothermal Field
○Muhamad Firdaus Al Hakim1, Katsuaki Koike1, Indra Nugroho2 (1. Kyoto University, 2. PT Supreme Energy)
Chairperson:
Keywords:
Microearthquake,Deep Learning,EQTransformer,Geothermal Monitoring
Large volumes of continuous seismic data present a significant challenge for manual earthquake event detection. This study applies a deep learning approach to analyze MEQ occurrences from continuous seismic data as a monitoring tool at a geothermal field. We tested eight pre-trained weight configurations of EQTransformer and PhaseNet via the SeisBench framework to identify the best-performing combination for this dataset. The seismic data were recorded continuously by 15 stations over one month in January 2021 as benchmark test. The results demonstrated that EQTransformer with the 'volpick' weights achieved the highest recall value and overall score for the one-month dataset. GaMMA phase association yielded 577 detected events that surpass the 279 MEQ events in the existing conventional catalog. Preliminary hypocenter determination was performed using the Geiger approach for both the deep learning and manual catalogs to obtain initial hypocenter locations. Both seismic clusters showed a consistent ENE-WSW-trending pattern in the southern part of the study area which coincides with the injection zone. However, the centroids of both clusters were separated by approximately 1 km likely due to differences in origin time estimation and arrival time picks between the two approaches. These results provide significant insight into the potential of the deep learning approach for injection-induced MEQ analysis and offers an efficient alternative to manual methods for seismic catalog construction and long-term monitoring.