[S-CG45] Science of slow-to-fast earthquakes

[SCG45-30] Detecting and characterizing slow earthquakes with deep learning on seismic data

★Invited Papers

*Jannes Munchmeyer¹, William B Frank², Sophie Giffard-Roisin¹, David Marsan¹, Anne Socquet¹ (1.Univ. Grenoble Alpes, Univ. Savoie Mont Blanc, CNRS, IRD, Univ. Gustave Eiffel, ISTerre, Grenoble, France, 2.Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of Technology, Cambridge, MA, USA)

Keywords:Deep learning, low-frequency earthquakes, seismic swarms, subduction

Slow earthquakes play an integral role in the relaxation of tectonic stresses on the subduction interface, showing complex interactions with fast deformation. Yet understanding slow earthquakes is difficult, because they are often hidden in seismic or geodetic noise. Here, we show different ways in which deep learning on seismic data can help detecting and characterizing slow earthquakes. Deep learning enables a data driven analysis of seismic data and, over the last years, has lead to substantial improvements in earthquake seismology. We first show how deep learning allows for a step-change in seismic catalogs, and how this revealed the initiation mechanisms of the 2023 shallow slow slip event around the Copiapó ridge, Chile. Second, we explain how deep learning can directly detect low-frequency earthquakes. Notably, we found that deep learning models transfer across world regions, e.g., models trained on data from Nankai can successfully recover low-frequency earthquakes in Japan. This points at universal waveform characteristics of low-frequency events across world regions. Lastly, we show how deep learning can also be used to show the absence of slow earthquakes. To this end, we conduct a systematic search for low-frequency earthquakes and tectonic tremors in the Atacama segement of the Chilean subduction zone, that turns up empty. We thereby provide an overview of different opportunities and challenges for the application of deep learning in the study of slow earthquakes.