Automated earthquake detection and high-resolution catalog construction in the Jailolo volcanic region, Indonesia

Abstrak

The Jailolo region, located on Halmahera Island, is characterized by intense seismic activity due to its proximity to two subduction zones and the Halmahera volcanic arc. To optimize the analysis of local earthquake activity in the region, we used an automated workflow that combines the deep-learning phase picker EQtransformer with the unsupervised machine learning association algorithm GaMMA to construct a high-resolution microseismic catalog. Phase arrivals were first identified using the EQTransformer and subsequently associated with each other using GaMMA, a Bayesian Gaussian Mixture Model with spatial clustering for earthquake event detection. This workflow produced 6,174 preliminary detections, of which 2,902 high-quality events were retained after relocation and filtering. The earthquakes predominantly occurred at depths shallower than 50 km, with an average local magnitude (ML) of 0.91. The obtained catalog revealed a large number of previously unreported small-magnitude earthquakes in the International Seismological Center (ISC) catalog. The completeness magnitude (Mc = 0.63) indicates a significant improvement in detection capability, enabling detailed analysis of microseismicity. The study highlights the effectiveness of automated machine-learning techniques in improving the resolution of seismicity patterns in a tectonically complex region such as Jailolo.

Identifier

• OpenAlex: https://openalex.org/W7154914699
• DOI: https://doi.org/10.15625/2615-9783/24522