Classical mechanisms of volcanic eruptions mostly involve pressure buildup and magma ascent towards the surface¹. Such processes produce geophysical and geochemical signals that may be detected and interpreted as eruption precursors^1–3. On 22 May 2021, Mount Nyiragongo (Democratic Republic of the Congo), an open-vent volcano with a persistent lava lake perched within its summit crater, shook up this interpretation by producing an approximately six-hour-long flank eruption without apparent precursors, followed—rather than preceded—by lateral magma motion into the crust. Here we show that this reversed sequence was most likely initiated by a rupture of the edifice, producing deadly lava flows and triggering a voluminous 25-km-long dyke intrusion. The dyke propagated southwards at very shallow depth (less than 500 m) underneath the cities of Goma (Democratic Republic of the Congo) and Gisenyi (Rwanda), as well as Lake Kivu. This volcanic crisis raises new questions about the mechanisms controlling such eruptions and the possibility of facing substantially more hazardous events, such as effusions within densely urbanized areas, phreato-magmatism or a limnic eruption from the gas-rich Lake Kivu. It also more generally highlights the challenges faced with open-vent volcanoes for monitoring, early detection and risk management when a significant volume of magma is stored close to the surface.

Magma propagation is an unsteady process controlled by magma‐crust interaction. To provide information on its dynamics, we invert complementary ground deformation data spanning the 8 hours preceding the May 26, 2016 eruption at Piton de la Fournaise volcano (La Réunion, France). Data are inverted using 3‐D boundary element models combined with a Monte Carlo inversion method. The final geometry of the displacement source is determined based on four interferograms spanning the whole propagation phase while the dynamics of the propagation is inferred from temporal inversion of continuous GNSS data, using the final geometry as an a priori to constrain the source. The best modeled magma path consists in a 2700‐meter‐long sill located 800 meters above sea level and connected to the eruptive fissure by a sub‐vertical dike. The quick opening of the horizontal part of the intrusion could have been favored by limited flank sliding during the early stage of propagation. The intrusion then stalled for ∼5 hours, while pressure increased slightly, until final upward propagation and eruption. Volume budget suggests that the eruption was fed by a single batch of magma quickly disconnected from its source. The delay prior to the eruption may reflect a limited magma supply. Finally, two mechanisms, potentially acting together, might have favored the eruption : a driving role of magmatic gas and/or, as often observed at Piton de la Fournaise, an eastward flank slip.

Mixed‐mode fluid‐filled cracks represent a common means of fluid transport within the Earth's crust. They often show complex propagation paths which may be due to interaction with crustal heterogeneities or heterogeneous crustal stress. Previous experimental and numerical studies focus on the interplay between fluid overpressure and external stress but neglect the effect of other crack parameters. In this study, we address the role of crack length on the propagation paths in the presence of an external heterogeneous stress field. We make use of numerical simulations of magmatic dike and hydrofracture propagation, carried out using a two‐dimensional boundary element model, and analogue experiments of air‐filled crack propagation into a transparent gelatin block. We use a 3‐D finite element model to compute the stress field acting within the gelatin block and perform a quantitative comparison between 2‐D numerical simulations and experiments. We show that, given the same ratio between external stress and fluid pressure, longer fluid‐filled cracks are less sensitive to the background stress, and we quantify this effect on fluid‐filled crack paths. Combining the magnitude of the external stress, the fluid pressure, and the crack length, we define a new parameter, which characterizes two end member scenarios for the propagation path of a fluid‐filled fracture. Our results have important implications for volcanological studies which aim to address the problem of complex trajectories of magmatic dikes (i.e., to forecast scenarios of new vents opening at volcanoes) but also have implications for studies that address the growth and propagation of natural and induced hydrofractures.

The ever‐increasing amount of Synthetic Aperture Radar (SAR) data motivates the development of automatic processing chains to fully exploit the opportunities offered by these large databases. The Synthetic Aperture Radar Interferometry (InSAR) Mass processing Toolbox for Multidimensional time series is an optimized tool to automatically download SAR data, select the interferometric pairs, perform the interferometric mass processing, compute the geocoded deformation maps, invert and display the velocity maps and the 2D time series on a web page updated incrementally as soon as a new image is available. New challenges relate to data management and processing load. We address them through methodological improvements dedicated to optimizing the InSAR pair selection. The proposed algorithm narrows the classical selection based on the shortest temporal and spatial baselines thanks to a coherence proxy and balances the use of each image as Primary and Secondary images thanks to graph theory methods. We apply the processing to three volcanic areas characterized with different climate, vegetation, and deformation characteristics: the Virunga Volcanic Province (DR Congo), the Reunion Island (France), and the Domuyo and Laguna del Maule area (Chile‐Argentina border). Compared to pair selection based solely on baseline criteria, this new tool produces similar velocity maps while reducing the total number of computed differential InSAR interferograms by up to 75%, which drastically reduces the computation time. The optimization also allows to reduce the influence of DEM errors and atmospheric phase screen, which increase the signal‐to‐noise ratio of the inverted displacement time series.

Surface deformation accompanying dike intrusions is dominated by uplift and horizontal motion directly related to the intrusions. In some cases, it includes subsidence due to associated magma reservoir deflation. When reservoir deflation is large enough, it can form, or reactivate preexisting, caldera ring‐faults. Ring‐fault reactivation, however, is rarely observed during moderate‐sized eruptions. On February 21, 2015 at Ambrym volcano in Vanuatu, a basaltic dike intrusion produced more than 1 m of coeruptive uplift, as measured by InSAR, synthetic aperture radar correlation, and Multiple Aperture Interferometry. Here, we show that an average of ∼40 cm of slip occurred on a normal caldera ring‐fault during this moderate‐sized (VEI < 3) event, which intruded a volume of ∼24 × 10⁶ m³ and erupted ∼9.3 × 10⁶ m³ of lava (DRE). Using the 3D Mixed Boundary Element Method, we explore the stress change imposed by the opening dike and the depressurizing reservoir on a passive, frictionless fault. Normal fault slip is promoted when stress is transferred from a depressurizing reservoir beneath one of Ambrym's main craters. After estimating magma compressibility, we provide an upper bound on the critical fraction (f = 7%) of magma extracted from the reservoir to trigger fault slip. We infer that broad basaltic calderas may form in part by hundreds of subsidence episodes no greater than a few meters, as a result of magma extraction from the reservoir during moderate‐sized dike intrusions.