An Enhanced Deep-Learning Catalog of the Mw 8.8 Maule Aftershock Sequence

We re-examine the aftershock sequence of the Mw 8.8 Maule earthquake in south-central Chile using deep learning on 10 months of continuous seismic data from 156 temporary stations along the rupture zone (March 2010–March 2011). By integrating back-projection and matched filtering with PhaseNet (a deep-learning phase picker), we initially identify 99,137 earthquakes. We then relocate these events using NonLinLoc with source-specific station terms and waveform coherence. We select a subset of 8,894 earthquakes for template matching and obtain a final catalog of 374,058 earthquakes—nearly 12 times more than previous studies—achieving a magnitude of completeness of Mw 1.7, which is an order of magnitude better. The spatiotemporal evolution of the seismicity reveals intricate seismic structures, including a highly active shallow cluster in the Pichilemu-Vichuquén region (33.5°S–35°S) showing a complex L-shaped geometry and deeper slab-related seismicity near Concepción (37°S–38°S). Spatial and temporal variation of the b-value further highlight heterogeneous post-seismic deformation driven by multiple fault system activations. This study demonstrates how modern analytical techniques, particularly machine learning, extract valuable insights from older datasets, enabling the discovery of previously undetected small-amplitude seismicity and refining our understanding of earthquake dynamics and seismic hazards.