A dense focal mechanisms catalogue for the Atacama segment in Chile (24◦S - 31◦S) using deep learning polarity picking

The Atacama segment in Northern Chile, a persistent seismic gap since 1922, represents a complex and highly active subduction zone. The region’s seismicity spans a broad magnitude range (M -0.8 to 6.2) and encompasses diverse sources including intraslab, interface, upper-mantle, and outer-rise events. While a dense earthquake catalog exists, systematic information on focal mechanisms has remained sparse, limiting detailed understanding of stress distribution and seismotectonic processes.

Here, we construct a comprehensive focal mechanism catalog for the Atacama seismic gap using P-wave polarity inversions implemented via the SKHASH algorithm based on a grid-search of nodal planes. First-motion polarities were automatically picked using a CNN model trained on 3 millions human-picked examples from diverse tectonic settings to improve cross-regional transferability. Rigorous quality selection was applied, accounting for signal-to-noise ratio, azimuthal coverage and minimum allowed number of polarities, ensuring robust mechanism determination. Under current network configuration and resolution constraints, around ~30% of the catalog (initially counting ~166 000 events) can be resolved. 

The resulting catalog provides a detailed statistical overview of mechanisms across different seismic classes and magnitudes. Particular attention is given to markers of slab stress state - along-dip compressions and tensions, its distribution across the double plane seismicity zone, and to mechanisms of upper-plate seismicity, which may reflect fluid transfer and crustal heterogeneities.