Dynamics of Deep Earthquakes

Deep earthquakes (>70km) are a puzzling yet insufficiently understood process occurring inside the Earth. Despite their potential for significant magnitudes (up to 9) and associated hazards, their physics remains elusive. The theoretical inhibition of earthquakes with increasing temperature and pressure at depth contrasts with over a century of observed occurrences, prompting the persistent question: What causes deep earthquakes? We argue that to tackle this question we need a methodological shift to obtain new observations of deep Earth processes and dynamics, which must be integrated into realistic physical models of slabs. The SODA project introduces deep learning methods to analyse seismological data aiming at improving the detection, characterization and classification of deep events. The outcome will be a new generation of observations offering an unprecedented view of the spatiotemporal evolution (e.g., spatial migrations) of deep events, directly informing us of the physical processes occurring in subduction zones (e.g., fluid movements, aseismic deformation). We will then determine the complex thermal structure, mineralogy, and geometry of subduction zones. Finally, we will use machine learning to unveil the relationships in between our new seismological observations and the modelled physical properties of the slabs, to quantitatively unravel the boundary conditions controlling the occurrence and the dynamics of deep events. The integrated approach of the SODA project will provide an unparalleled quantitative view of the deep Earth dynamics, shedding new light on the physical processes occurring at the mantle level and providing a quantitative assessment of the mechanisms controlling the occurrence and style of deep earthquakes. Our project will thus represent a major leap forward in the understanding of deep earthquakes, but will also have a major impact on the understanding of Earth dynamics and plate tectonics as whole.