An earthquake sequence occurred in the Atacama region of Chile throughout September 2020. The sequence initiated by a mainshock of magnitude = 6.9, followed 17 hours later by a = 6.4 aftershock. The sequence lasted several weeks, during which more than a thousand events larger than = 1 occurred, including several larger earthquakes of magnitudes between 5.5 and 6.4. Using a dense network that includes broad-band, strong motion and GPS sites, we study in details the seismic sources of the mainshock and its largest aftershock, the afterslip they generate and their aftershock, shedding light of the spatial temporal evolution of seismic and aseismic slip during the sequence. Dynamic inversions show that the two largest earthquakes are located on the subduction interface and have a stress drop and rupture times which are characteristics of subduction earthquakes. The mainshock and the aftershocks, localised in a 3D velocity model, occur in a narrow region of interseismic cou-

An earthquake sequence occurred in the Atacama region of Chile throughout September 2020. The sequence initiated by a mainshock of magnitude Mw6.9, followed 17 hours later by a Mw6.4 aftershock. The sequence lasted several weeks, during which more than a thousand events larger than Ml 1 occurred, including several larger earthquakes of magnitudes between 5.5 and 6.4. Using a dense network that includes broad-band, strong motion and GPS sites, we study in details the seismic sources of the mainshock and its largest aftershock, the afterslip they generate and their aftershock, shedding light of the spatial temporal evolution of seismic and aseismic slip during the sequence. Dynamic inversions show that the two largest earthquakes are located on the subduction interface and have a stress drop and rupture times which are characteristics of subduction earthquakes. The mainshock and the aftershocks, localised in a 3D velocity model, occur in a narrow region of interseismic coupling (ranging 40%-80%), i.e. between two large highly coupled areas, North and South of the sequence, both ruptured by the great Mw~8.5 1922 megathrust earthquake. High rate GPS data (1 Hz) allow to determine instantaneous coseismic displacements and to infer coseismic slip models, not contaminated by early afterslip. We find that the total slip over 24 hours inferred from precise daily solutions is larger than the sum of the two instantaneous coseismic slip models. Differencing the two models indicates that rapid aseismic slip developed up-dip the mainshock rupture area and down-dip of the largest aftershock. During the 17 hours separating the two earthquakes, micro-seismicity migrated from the mainshock rupture area up-dip towards the epicenter of the Mw6.4 aftershocks and continued to propagate upwards at ~0.7 km/day. The bulk of the afterslip is located up-dip the mainshock and down-dip the largest aftershock, and is accompanied with the migration of seismicity, from the mainshock rupture to the aftershock area, suggesting that this aseismic slip triggered the Mw6.4 aftershock. Unusually large post-seismic slip, equivalent to Mw6.8 developed during three weeks to the North, in low coupling areas located both up-dip and downdip the narrow strip of higher coupling, and possibly connecting to the area of the deep Slow Sleep Event detected in the Copiapo area in 2014. The sequence highlights how seismic and aseismic slip interacted and witness short scale lateral variations of friction properties at the megathrust.

Abstract. What controls the location and segmentation of mega-earthquakes in subduction zones is a long-standing problem in Earth sciences. Prediction of earthquake rupture extent mostly relies on interplate coupling models based on Global Navigation Satellite Systems providing patterns of slip deficit between tectonic plates. We here investigate if and how the strongly and weakly coupled patches revealed by these models relate to the distribution of deformation along the plate interface, i.e., basal erosion and/or underplating. From a mechanical analysis of the topography applied along the Chilean subduction zone, we show that extensive plate interface deformation takes place along most of the margin. We show that basal erosion occurs preferentially at 15 km depth while underplating does at 35 ± 10 and 60 ± 5 km depth, in agreement with P-T conditions of recovered underplated material, expected pore pressures and the spatial distribution of marine terraces and uplift rates. South of the Juan Fernández Ridge, large sediment input favors shallow accretion and underplating of subducted sediments, while along northern Chile, extensive basal erosion provides material for the underplating. We then show that, along the accretionary margin, the two last major earthquakes were limited along their down-dip end by underplating while, along the erosive margin, they were surrounded by both basal erosion and underplating. Segments with heterogeneously distributed deformation largely coincide with lateral earthquake terminations. We therefore propose that long-lived plate interface deformation promotes stress build-up and leads to earthquake nucleation. Earthquakes then propagate along fault planes shielded from this long-lived permanent deformation, and are finally stopped by segments of heterogeneously distributed deformation. Slip deficit patterns and earthquake segmentation therefore reflect the along-dip and along-strike distribution of the plate interface deformation. Topography acts as a mirror of distributed plate interface deformation and should be more systematically studied to improve the prediction of earthquake ruptures.

During the weeks following a large earthquake, transient aseismic slip, referred as afterslip, develops at the fault and dominates the deformation observed at the Earth' surface. The development of continuous GNSS networks at subduction zones for almost three decades has allowed to capture the main pattern of the afterslip that followed some major megathrust events: afterslip initiates immediately after the earthquake, with a slip rate decreasing through time as 𝐴𝐴 1 𝑡𝑡 leading to a logarithmic growth of the cumulative slip (e.g., Perfettini et al., 2010). Afterslip takes place at areas of the fault surrounding the coseismic rupture (Perfettini et al., 2010), with little if any slip inside the coseismic rupture. The evolution of the cumulative number of aftershocks mimics the afterslip evolution both in time and space, and the moment released through aftershocks is only a small fraction of the afterslip equivalent moment (Hsu et al., 2006;Perfettini & Avouac, 2004).