Seismic events relocation and velocity models for the Atacama seismic gap at Central-Northern Chile (24.5-29°S)

<p>Information about seismic velocities and the seismicity distribution in subduction zones can reveal the effect physical properties have over large ruptures. The ANILLO (Y6) Seismic Network spans between 24.5&#176;S - 29&#176;S in Chile. Composed of 104 seismic stations, it was operative between 2020 and 2022. Deep Learning algorithms (EQTransformer) enable the identification of more than 30,000 seismic events during the eighteen months of operation time, in an area with an important lack of moderate-to-large events during the last century since the nucleation of the 1922 M8.5 Atacama earthquake. Using a subset of events with a good spatial distribution, low residuals, and many P- and S-arrivals, we obtained a well-distributed subcatalog of 1000 earthquakes, corresponding to 26,570 P- and 22,109 S-wave arrival times. Using VELEST for the selected events, inverting for station corrections, we computed a novel 1-D velocity model representative of this region by minimizing the subset residuals. This enabled us to relocate the complete catalog, reducing both the residuals and the location errors of all events. Starting from this 1-D velocity model, staggered tomographic inversions were performed using SIMUL2000, simultaneously inverting for seismic velocity models and hypocentral parameters according to the iterative damped least squares method. The complexity in the tomographic model was gradually increased, from 2-D to ultimately a 3-D fine model with low node separation. Preliminary results show that earthquake arrival times picked using Deep Learning algorithms can provide an exceptional database for tomographic studies, with residuals similar to human-picked databases. We obtain a good resolution underneath the seismic network in the forearc, with a reduced resolution and smearing in off-shore areas where location uncertainties, especially in the vertical direction, are greater. We suggest that the incorporation of OBS data would significantly improve the off-shore model. We observe an east-dipping, high Vp velocity structure subducting in the expected slab position, in which the relocated seismic events, with a low vertical location error, follow the slab geometry in distinct seismicity bands. Additionally, observed shallow seismicity could be related to crustal faults, mining activity, or even dehydration processes near the slab's upper surface. The event locations and velocity structure are interpreted in terms of the geological and geophysical characteristics of the subduction zone at these latitudes that may influence the behavior of large megathrust earthquakes.</p>