40Large volcanic eruptions on Earth commonly occur with collapse of the roof of a crustal magma 41 reservoir, forming a caldera. Only a few such collapses occur per century and lack of detailed 42 observations has obscured insight on mechanical interplay between collapse and eruption. We use Calderas are 1 -100 km diameter depressions found in volcanic regions of Earth and other planets. basaltic andesite) intrusive activity and eruptions (2,(9)(10)(11)(12). 59The consensus from field and modelling studies is that caldera collapse progresses from initial 60 surface downsag to fault-controlled subsidence (1, 8, 13, 14). The pre-collapse topography is obtained by subtracting the subsidence observed at the surface. As we recorded the caldera subsidence mainly on the ice (Fig. 1, Fig. S1), we made corrections and (Fig. 3A). We therefore conclude that suggestions of a large increase in ice flow out of the caldera 147 during these events (25) cannot be fitted with our data. 148Bedrock subsidence exceeding 1 m occurred within an area of 110 km 2 that extended beyond the 149 pre-existing caldera (Fig. 1, Fig. S1). After termination of collapse the total subsidence at the pre-150 existing caldera rims amounted to 3 to 11 meters ( Fig. 1D and 1E). Using subglacial radio-echo GPS station in the center of the caldera (Fig. 1A), including the rate of vertical rate of ice surface Cumulative number of M>4 caldera earthquakes, with magnitude evolution colored in red, blue and 176 grey representing clusters on the southern rim, the northern rim and smaller clusters, respectively 177 (see Fig. S5). E) Cumulative seismic moment for M>4 caldera earthquakes. from analysis of subaerial gas measurements (Fig. 4). This depth concurs with our regional on FTIR and Multi-GAS measurements (24). 194Seismicity and subsurface structure 195 We used seismic data and Distinct Element Method (DEM) numerical modelling (24), to 196 characterize the deeper collapse structure as the reactivation of a steeply-inclined ring fault (Fig. 5). 197We mostly observed seismicity at depths of 0-9 km beneath the northern and southern caldera rims 198( Fig. 5B), with earthquakes being more numerous on the northern rim. This spatial pattern of 199 seismicity is consistent with fracturing above a deflating magma reservoir that was elliptical in (Fig. 5C, D). Our best fitting models had preexisting faults dipping out at 80-85¡ from the caldera 207 center on the north side and at 85-90¡ toward the caldera center on the south side. The modeled pre- 208existing faults lay at 1-2 km below the surface on the north side and 3-4 km on the south side. 209Modeling of a more complex fault geometry or the inclusion of greater material heterogeneity may 210 further improve the data fit, but presently lacks robust geophysical constraints. components of the observed earthquakes at B ‡rdarbunga. We, however, narrowed down on 222 plausible solutions by using the micro-earthquakes (Fig. 5A). The moment tensor solutions are well 223 constrained, but the inferred d...

than the seismic waves (Hill et al., 2002), thus eruption triggering by static deformation is most likely to be observed at volcanoes located in proximity to an earthquake rupture plane. We study earthquakes at subduction zones with rupture lengths of several hundred kilometers, because they affect a large number of potentially active volcanoes, i.e., the 1952 Kamchatka M 9.0, the 1960 Chile M 9.5, the 1964 Alaska M 9.2, and the 2004 Sumatra-Andaman M 9.3 earthquakes (Fig. 1).

Changing stresses in multi-stage caldera volcanoes were simulated in scaled analogue experiments aiming to reconstruct the mechanism(s) associated with caldera formation and the corresponding zones of structural weakness. We evaluate characteristic structures resulting from doming (chamber inflation), evacuation collapse (chamber deflation) and cyclic resurgence (inflation and deflation), and we analyse the consequential fault patterns and their statistical relationship to morphology and geometry. Doming results in radial fractures and subordinate concentric reverse faults which propagate divergently from the chamber upwards with increasing dilation. The structural dome so produced is characterised bysteepening in the periphery, whereas the broadening apex subsides. Pure evacuation causes the chamber roof to collapse along adjacent bell-shaped reverse faults. The distribution of concentric faults is influenced by the initial edifice morphology: steep and irregular initial flanks result in a tilted or chaotic caldera floor. The third set of experiments focused on the structural interaction of cyclic inflation and subsequent moderate deflation. Following doming, caldera subsidence produces concentric faults that characteristically crosscut radial cracks of the dome. The flanks of the edifice relax, resulting in discontinuous circumferential faults that outline a structural network of radial and concentric faults; the latter form locally uplifted and tiltedwedges (halfgrabens) that grade into horst-and-graben structures. This superimposed fault pattern also extends inside the caldera. We suggest that major pressure deviations in magma chamber(s) are reflected in the fault arrangement dissecting the volcanoflanks and may be used as a first-

Flank instability and sector collapses, which pose major threats, are common on volcanic islands. On 22 Dec 2018, a sector collapse event occurred at Anak Krakatau volcano in the Sunda Strait, triggering a deadly tsunami. Here we use multiparametric ground-based and space-borne data to show that prior to its collapse, the volcano exhibited an elevated state of activity, including precursory thermal anomalies, an increase in the island’s surface area, and a gradual seaward motion of its southwestern flank on a dipping décollement. Two minutes after a small earthquake, seismic signals characterize the collapse of the volcano’s flank at 13:55 UTC. This sector collapse decapitated the cone-shaped edifice and triggered a tsunami that caused 430 fatalities. We discuss the nature of the precursor processes underpinning the collapse that culminated in a complex hazard cascade with important implications for the early detection of potential flank instability at other volcanoes.

Large tectonic earthquakes lead to significant deformations in the months and years thereafter. These so-called post-seismic deformations include contributions mainly from afterslip and viscoelastic relaxation, quantification of their relative influence is of importance for understanding the evolution of post-seismic crustal stress, strain and aftershocks. Here, we investigate the post-seismic deformation processes following the 2011 M w 9.0 Tohoku earthquake using surface displacement data as observed by the onshore global positioning system network in the first ∼1.5 yr following the main shock. We explore two different inversion modelling strategies: (i) we simulate pure afterslip and (ii) we simulate the combined effect of afterslip and viscoelastic relaxation. By assuming that the afterslip is solely responsible for the observed post-seismic deformation, we find most afterslip activities to be located close to the downdip area of the coseismic rupture at 20-80 km depth with a maximum cumulative slip of ∼3.8 m and a seismic moment of 2.3 × 10 22 Nm, equivalent in moment to an M w 8.84 earthquake. By assuming a combination of afterslip and viscoelastic components, the best data fit is found for an afterslip portion that is spatially consistent with the pure afterslip model, but reveals a decreased seismic moment of 2.1 × 10 22 Nm, or M w 8.82. In addition, the combined model suggests an effective thickness of the elastic crust of ∼50 km overlying an asthenosphere with a Maxwell viscosity of 2 × 10 19 Pa s. Temporal analysis of our model inversions suggests that the rate of afterslip rapidly decreases with time, consistent with the state-and rate-strengthening frictional law. The spatial pattern of afterslip coincides with the locations of aftershocks, and also with the area of coseismically increased Coulomb failure stress (CFS). Only a small part of the coseismically increased CFS was released by the afterslip in 564 d after the event. The effect of the viscoelastic relaxation within this initial stage only plays a secondary role, but it shows an increasing tendency, that is, the contribution of viscoelastic relaxation increases with time. Further geodetic observations are needed for a robust quantification of the role of the viscoelastic relaxation in the post-seismic deformation.