Advancing Seismic Event Detection: Integrating Machine Learning with Waveform-Stacking Techniques

In the realm of seismic and microseismic event detection and localization, our research marks a significant step forward by integrating machine learning with advanced waveform-stacking techniques for detecting, locating and characterising seismic events. This integration is crucial for unravelling the complex spatio-temporal patterns of seismicity sequences. Our study addresses the challenges posed by noise-dominant microseismic events, which are typically overlooked by conventional detection methods. Building upon the foundational work on migration and stacking, we have developed an automated, data-driven method utilising a neural network trained in seismic phase arrival identification. This approach, underpinned by stacking and migration techniques, is enhanced by the incorporation of a spatial octree to precisely and efficiently localise seismic sources. These enhancement gives insights into complex seismic sequences, such as volcanic swarms and regional tectonic sequences. The software framework facilitates extensive feature extraction, such as local and moment magnitudes, enabling the study of seismic events across various scales and tectonic settings. This is exemplified in our validation studies using data from the Eifel region, Germany, and the Reykjanes Peninsula, Iceland. These regions, known for their diverse seismic activities including tectonic earthquakes and fluid-induced swarm activity, provide a rich dataset for testing our method's efficacy in different geological contexts. Our research contributes to the session's overarching goal of understanding the physical processes behind complex seismic sequences. By enhancing detection and localization capabilities, we aim to offer new perspectives and tools for the geophysical community to investigate the triggering mechanisms of these sequences with unprecedented resolution.