Recent improvements in the frequency, type, and availability of satellite images mean it is now feasible to routinely study volcanoes in remote and inaccessible regions, including those with no ground‐based monitoring. In particular, Interferometric Synthetic Aperture Radar data can detect surface deformation, which has a strong statistical link to eruption. However, the data set produced by the recently launched Sentinel‐1 satellite is too large to be manually analyzed on a global basis. In this study, we systematically process >30,000 short‐term interferograms at over 900 volcanoes and apply machine learning algorithms to automatically detect volcanic ground deformation. We use a convolutional neutral network to classify interferometric fringes in wrapped interferograms with no atmospheric corrections. We employ a transfer learning strategy and test a range of pretrained networks, finding that AlexNet is best suited to this task. The positive results are checked by an expert and fed back for model updating. Following training with a combination of both positive and negative examples, this method reduced the number of interferograms to ∼100 which required further inspection, of which at least 39 are considered true positives. We demonstrate that machine learning can efficiently detect large, rapid deformation signals in wrapped interferograms, but further development is required to detect slow or small deformation patterns which do not generate multiple fringes in short duration interferograms. This study is the first to use machine learning approaches for detecting volcanic deformation in large data sets and demonstrates the potential of such techniques for developing alert systems based on satellite imagery.

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version will undergo additional copyediting, typesetting and review before it is published in its final form, but we are providing this version to give early visibility of the article. Please note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Space-borne Synthetic Aperture Radar (SAR) Interferometry (InSAR) is now a key geophysical tool for surface deformation studies. The European Commission’s Sentinel-1 Constellation began acquiring data systematically in late 2014. The data, which are free and open access, have global coverage at moderate resolution with a 6 or 12-day revisit, enabling researchers to investigate large-scale surface deformation systematically through time. However, full exploitation of the potential of Sentinel-1 requires specific processing approaches as well as the efficient use of modern computing and data storage facilities. Here we present Looking Into Continents from Space with Synthetic Aperture Radar (LiCSAR), an operational system built for large-scale interferometric processing of Sentinel-1 data. LiCSAR is designed to automatically produce geocoded wrapped and unwrapped interferograms and coherence estimates, for large regions, at 0.001° resolution (WGS-84 coordinate system). The products are continuously updated at a frequency depending on prioritised regions (monthly, weekly or live update strategy). The products are open and freely accessible and downloadable through an online portal. We describe the algorithms, processing, and storage solutions implemented in LiCSAR, and show several case studies that use LiCSAR products to measure tectonic and volcanic deformation. We aim to accelerate the uptake of InSAR data by researchers as well as non-expert users by mass producing interferograms and derived products.

Satellites enable widespread, regional or global surveillance of volcanoes and can provide the first indication of volcanic unrest or eruption. Here we consider Interferometric Synthetic Aperture Radar (InSAR), which can be employed to detect surface deformation with a strong statistical link to eruption. Recent developments in technology as well as improved computational power have resulted in unprecedented quantities of monitoring data, which can no longer be inspected manually. The ability of machine learning to automatically identify signals of interest in these large InSAR datasets has already been demonstrated, but data-driven techniques, such as convolutional neutral networks (CNN) require balanced training datasets of positive and negative signals to effectively differentiate between real deformation and noise. As only a small proportion of volcanoes are deforming and atmospheric noise is ubiquitous, the use of machine learning for detecting volcanic unrest is more challenging than many other applications. In this paper, we address this problem using synthetic interferograms to train the AlexNet CNN. The synthetic interferograms are composed of 3 parts: 1) deformation patterns based on a Monte Carlo selection of parameters for analytic forward models, 2) stratified atmospheric effects derived from weather models and 3) turbulent atmospheric effects based on statistical simulations of correlated noise. The AlexNet architecture trained with synthetic data outperforms that trained using real interferograms alone, based on classification accuracy and positive predictive value (PPV). However, the models used to generate the synthetic signals are a simplification of the natural processes, so we retrain the CNN with a combined dataset consisting of synthetic models and selected real examples, achieving a final PPV of 82%. Although applying atmospheric corrections to the entire dataset is computationally expensive, it is relatively simple to apply them to the small subset of positive results. This further improves the detection performance without a significant increase in computational burden (PPV of 100%). Thus, we demonstrate that training with synthetic examples can improve the ability of CNNs to detect volcano deformation in satellite images, and propose an efficient workflow for the development of automated systems.

International audienceWe investigate how surface load variations around volcanoes act on shallow magma chambers. Numerical calculations are carried out in axisymmetric geometry for an elliptical chamber embedded in an elastic medium. Magma compressibility is taken into account. For variable chamber shape, size and depth, we quantify how unloading events induce magmatic pressure change as well as variation of the threshold pressure required for dyke initiation at the chamber wall. We evaluate the triggering effect of these surface events on onset of eruptions and find it depends strongly on the surface load location and the magma chamber shape. We apply this model to two active Icelandic subglacial volcanoes: Grímsvötn and Katla. The 2004 eruption of Grímsvötn was immediately preceded by a jökulhlaup, a glacial outburst flood of 0.5km3. We show that this event may have triggered the eruption only if the system was very close to failure conditions. Katla volcano is covered by the Mýrdalsjökull ice cap. An annual cycle, with up to 6m change in snow thickness, occurs from winter to summer. As the seasonal snow load is reduced, a pressure decrease of the same order of magnitude as the load is induced within the magma storage zone. Our model predicts that, in the case of a spherical or horizontally elongated magma chamber, eruptions are more likely when the snow cover is smallest, which appears consistent with the fact that all the last nine major historical eruptions at Katla occurred during the summer period. The model predicts an increase in Coulomb stress around the caldera, up to 7km from its centre, during unloading periods, enough to trigger earthquakes. Stress due to snow load variations, with focusing of it in weak zones near the caldera boundary, is considered a contributing factor to seasonal seismicity observed beneath Mýrdalsjökull

Forecasting explosive eruptions relies on using monitoring data to interpret the patterns and timescales of magma transport and mixing. In September 2017, a distal seismic swarm triggered the evacuation of around 140,000 people from Agung volcano, Bali. From satellite imagery and 3D numerical models, we show that seismicity was associated with a deep, sub-vertical magma intrusion between Agung and its neighbour Batur. This, combined with observations of the 1963 eruption which caused more than thousand fatalities, suggests a vertically and laterally interconnected system experiencing recurring magma mixing. The geometry of the 2017 dyke is consistent with transport from a deep mafic source to a shallow andesitic reservoir controlled by stresses induced by the topographic load, but not the regional tectonics. The ongoing interactions between Agung and Batur have important implications for interpretation of distal seismicity, the links between closely spaced arc volcanoes, and the potential for cascading hazards.

The strong tidal triggering of mid-ocean ridge earthquakes has remained unexplained because the earthquakes occur preferentially during low tide, when normal faulting earthquakes should be inhibited. Using Axial Volcano on the Juan de Fuca ridge as an example, we show that the axial magma chamber inflates/deflates in response to tidal stresses, producing Coulomb stresses on the faults that are opposite in sign to those produced by the tides. When the magma chamber’s bulk modulus is sufficiently low, the phase of tidal triggering is inverted. We find that the stress dependence of seismicity rate conforms to triggering theory over the entire tidal stress range. There is no triggering stress threshold and stress shadowing is just a continuous function of stress decrease. We find the viscous friction parameter A to be an order of magnitude smaller than laboratory measurements. The high tidal sensitivity at Axial Volcano results from the shallow earthquake depths.

Pressure influences both magma production and the failure of magma chambers. Changes in pressure interact with the local tectonic settings and can affect magmatic activity. Present-day reduction in ice load on subglacial volcanoes due to global warming is modifying pressure conditions in magmatic systems. The large pulse in volcanic production at the end of the last glaciation in Iceland suggests a link between unloading and volcanism, and models of that process can help to evaluate future scenarios. A viscoelastic model of glacio-isostatic adjustment that considers melt generation demonstrates how surface unloading may lead to a pulse in magmatic activity. Iceland's ice caps have been thinning since 1890 and glacial rebound at rates exceeding 20 mm yr −1 is ongoing. Modelling predicts a significant amount of 'additional' magma generation under Iceland due to ice retreat. The unloading also influences stress conditions in shallow magma chambers, modifying their failure conditions in a manner that depends critically on ice retreat, the shape and depth of magma chambers as well as the compressibility of the magma. An annual cycle of land elevation in Iceland, due to seasonal variation of ice mass, indicates an annual modulation of failure conditions in subglacial magma chambers.