Anomaly Detection for the Identification of Volcanic Unrest in Satellite Imagery

Geologists seek to understand the relationship between volcanic unrest and eruptions by identifying subtle Volcanic Thermal Features (VTFs) in high-resolution satellite imagery. This analysis requires the careful curation of large databases of relevant volcanic thermal information. However, volcanic unrest is characterized by highly subtle thermal anomalies. Manual identification on a global scale is highly labor- and time-intensive. We propose Hotspotter: an end-to-end system to automatically detect subtle volcanic thermal anomalies in satellite images and derive relevant thermal statistics. Previous solutions for automated VTF detection have limited data size and geographic diversity. To accommodate an unprecedentedly large and diverse volcanic dataset, we propose an automated pipeline combining unsupervised anomaly detection with supervised classification to filter anomalous regions. Hotspotter gives 90% anomaly detection accuracy and robust generalization to new volcanoes. Our automated approach can accelerate scientists' search for VTFs to help identify relevant thermal precursors and enable more precise forecasts of global volcanic eruptions.

A deep learning approach to detecting volcano deformation from satellite imagery using synthetic datasets

The possibility of forecasting volcanic eruptions remains a major challenge for the volcanological scientific community. To date, various techniques based on volcano-tectonic seismicity, endogenous gas emission and satellite imagery have been widely applied in an effort to understand and anticipate short-term volcanic behaviour leading to eruptions. The rescaled range analysis (R/S) applied to time series of volcano-tectonic earthquakes is a quantitative method for determining the short-term and long-term memory of seismic activity during volcanic unrest. By using the Hurst exponent, it is possible to identify the precise transition from anti-persistence to persistence in volcano-tectonic earthquake time-series (VT) associated with volcanic dike ascent. We calculated the Hurst exponent of volcano-tectonic earthquakes during the 2021 Tajogaite eruption (La Palma, Canary Islands), the temporal evolution of the GEOS diagram and its correlation with the sustained dynamics of the volcanic eruption. Our study suggests that the volcanic unrest system transitions from anti-persistence to persistence approximately two days before the eruption, indicating a non-return point and the imminent onset of the eruption. Furthermore, we identified five magma deep injections during the eruption. The final stage and potential cessation of the eruption can also be inferred from the asymptotic trend of the Hurst exponent.Supplementary InformationThe online version contains supplementary material available at 10.1038/s41598-025-28566-6.

<p>By 2021, Vulcano, Aeolian Islands (Italy), experienced a dramatic increase in different monitoring parameters, including microseismicity, ground deformation, fumarole temperatures, and volatile emissions of steam, carbon, and sulfur dioxide. The volcanic unrest was noticeable in September 2021, causing the Civil Protection to raise the alert level from green to yellow on October 1st. Here we present a number of ground- and satellite-based thermal methodologies used to detect and characterize the change of state of the La Fossa hydrothermal system between January 2021 and January 2022. We analyzed: (i) the temperature and (ii) CO2 flux data acquired at 15 cm‐depth on a N-S profile N-S and grid in the geothermally heated area during three field surveys in June, September 2021 and January 2022; (iii) a time series acquired with a radiometer including temperatures and number of vents inside the fumarole field from 1994 to 2022; (v) thermal images acquired by a hand-held thermal camera during four field surveys in March, June and September 2021, plus January 2022; (v) nighttime multi-spectral satellite images acquired by ASTER, ECOSTRESS and VIIRS sensors from January 2021 to January 2022. Satellite images show a clear increase in the radiant heat flux/land surface temperature as well as in the number of thermally anomalous pixels, this thermal anomaly has been observed from mid-September. However, by combining ground and satellite techniques the starting point of this change can be tracked thermally from at least June 2021. Our experience suggests that the methods, essentially based on the thermal monitoring, <span>could be used to herald upcoming crises. This method </span><span>has been applied on a close conduit volcano and highlighted changes of trend in the solfataric release. Further tests, aiming to reduce (filter or define) the external effects on the land surface temperature, and to define the correlations with the long term monitoring data (either ground-based or by remote sensing) in this area, would assess a standardized methodology to monitoring the subtle, but diffuse fluid release. The assessed methodology could then be applied to other active hydrothermal systems, to herald thermal changes on the surface, related to the increasing energy released from a deep source.</span></p>

Santorini Island is of volcanic origin and has historically faced repeated volcanic and seismic activity. In early 2025, increased volcanism and intensified earthquake activity, similar to 2011-2012, caused residents’ concern. This study aims to characterize ground deformation on Santorini Island during its volcanic unrest in 2025 using InSAR observations. For this purpose, 74 synthetic aperture radar (SAR) images of Sentinel-1A satellites in descending and ascending orbits were acquired from early January 2024 to late March 2025. Line-Of-Sight (LOS) velocity values of the descending and ascending orbits were decomposed to determine the east-west and vertical displacement velocities. According to the results obtained, uplifts up to +60 mm/year velocity values were detected in the central parts of the island called Caldera, and subsidence up to –30 mm/year velocity values were detected in the outer regions. In addition, eastward horizontal movements reaching velocities of +60 mm/year and westward horizontal movements reaching velocities of –50 mm/year were also detected throughout the island. In the second stage of the study, a total of 4 points were selected on the islands of Thira, Thirasia, Nea Kameni, and Palea Kameni, considering the Kameni and Kolumbo fault zones. For these points on the island of Santorini, the displacements occurring over 15 months were analysed by time series analysis, and the temporal behaviour of the deformation (increasing/decreasing trend) was monitored. The analysed data indicate that the ongoing horizontal and vertical movements on the island could be caused by volcanic rather than seismic effects, which is consistent with previous studies. This situation shows that volcanic risk assessments in the region should be monitored for the upcoming processes.

Molten sulfur is found in various subaerial volcanoes. However, limited records of the pools and flows of molten sulfur have been reported: therefore, questions remain regarding the physicochemical processes behind this phenomenon. A suite of new sulfur flows, some of which active, was identified at the Lastarria volcano (northern Chile) and studied using satellite imagery, in situ probing, and temperature and video recording. This finding provides a unique opportunity to better understand the emplacement mechanisms and mineral and chemical compositions of molten sulfur, in addition to gaining insight into its origin. Molten sulfur presented temperatures of 124–158°C, with the most prolonged sulfur flow reaching 12 m from the source. Photogrammetric tools permitted the identification of levees and channel structures, with an estimated average flow speed of 0.069 m/s. Field measurements yielded a total volume of 1.45 ± 0.29 m3 of sulfur (equivalent to ∼2.07 tons) mobilized during the January 2019 event for at least 408 min. Solidified sulfur was composed of native sulfur with minor galena and arsenic- and iodine-bearing minerals. Trace element analysis indicated substantial enrichment of Bi, Sb, Sn, Cd, as well as a very high concentration of As (>40.000 ppm). The January 2019 molten sulfur manifestations in Lastarria appear to be more enriched in As compared to the worldwide known volcanoes with molten sulfur records, such as the Shiretoko-Iozan and Poás volcanoes. Furthermore, their rheological properties suggest that the “time of activity” in events such as this could be underestimated as flows in Lastarria have moved significantly slower than previously thought. The origin of molten sulfur is ascribed to the favorable S-rich chemistry of fumarolic gases and changes in host rock permeability (fracture opening). Molten sulfur in Lastarria correlates with a peak in activity characterized by high emissions of SO2 and other acid species, such as HF and HCl, in addition to ground deformation. Consequently, molten sulfur was framed within a period of volcanic unrest in Lastarria, triggered by changes in the magmatic-hydrothermal system. The appearance of molten sulfur is related to physicochemical perturbations inside the volcanic system and is perhaps a precursor of eruptive activity, as observed in the Poás and Turrialba volcanoes.

Identifying the observable signals that warn against volcanic unrest and impending eruptions is one of the greatest challenges in the management of natural disasters. In this regard, satellite data has become a strong focus of global interest, offering abundant datasets from multi-missions and valuable tools to study Earth and improve physical models.The SAFARI project aims at developing a comprehensive space-based strategy for next-generation quantitative volcano hazard monitoring integrating the most recent satellite imagery capabilities and the relative products with the newest technologies mainly in the field of Machine Learning (ML) and Soft Computing. The main objectives of SAFARI include: (i) following the manifestations of unrests and impending eruptions, as well as (ii) forecasting the areas potentially threatened by volcanic products through eruptive scenarios. For this purpose, SAFARI intends to characterize the state of volcanic activity (quiet, unrest and eruptive phases) by taking advantage of a variety of satellite data, including active and passive sensors ranging from optical to microwave frequencies, and to extract quantitative satellite-derived input parameters to physical models for rapid and accurate scenario forecasting during eruptions. Well-established products from space-based volcano monitoring such as: (i) volcanic radiative power, (ii) surface displacement and (iii) volcanic gas emission (e.g., SO2, BrO) time series are processed jointly and supported by less frequently used but still informative time series such as (iv) ground skin temperature of the volcanic edifices, (v) change detection time series, (vi) time-varying volcanic ash indices, (vii) ash top height time series, (viii) gravity field variation and also (ix) time varying indices giving information about deformation phases of the volcanic edifice (i.e., inflation/deflation) as well as (x) crucial parameters related to the volcanic source (e.g., depth, volume variation) by using data assimilation to deformation models. SAFARI merges and assembles the latest developments from different INGV teams, in a way to analyze Earth observation (EO) data with a retrospective and multi-disciplinary approach, employing traditional statistical or numerical analysis, latest generation Graphic Processing Units (GPUs) architectures and newer and more sophisticated ML algorithms to classify time series, detect anomalies, and predict or estimate significant parameter values. The methodologies in SAFARI are developed and verified at four active volcanoes worldwide: Etna and Vulcano (Italy), continuously monitored by dense ground based networks managed by INGV, which will provide a first controlled experiment, and Nyiragongo (D.R. Congo) and Sangay (Ecuador), characterized by high volcanic hazard but with modest permanent monitoring networks, where satellite remote sensing is a key monitoring tool.The results of the SAFARI project and its underlying data source and methodologies, as well as the potential of the whole integrated processing chain, aim at becoming an effective tool for volcanic hazard analysis and impact quantification never used to date in volcanology, improving safety and reducing risk associated to eruptive events worldwide.

<p>Many (~250) volcanoes worldwide are occupied by glaciers. This can be problematic for volcano monitoring, since glacier ice potentially masks evidence of volcanic activity. However, some of the most devastating and costly volcanic eruptions of the last 100 years involved volcano-glacier interactions (e.g. Nevado del Ruiz 1985, Eyjafjallajökull 2010). Therefore, improving methods for monitoring glacier-covered volcanoes is of clear societal benefit. Optical satellite remote sensing datasets and techniques are perhaps most promising, since they frequently have a relatively high temporal and spatial resolution and are often freely available. These sources often show the effects of volcanic activity on glaciers, including ice cauldron formation, ice fracturing, and glacier terminus changes. In this study, we use satellite sources to investigate possible links between volcanic activity and changes in glacier velocity. Despite some studies reporting periods of glacier acceleration triggered by volcanic unrest, the potential of using the former to monitor the latter has yet to be investigated. Our approach is to observe how glacier surface velocity responded to past volcanic events in Alaska and Chile by applying feature-tracking, mostly using optical satellite imagery. The overall aim is to systematically track changes in the glacier velocity, with hope of improving volcano monitoring and eruption prediction. </p>

Multiparameter volcanic unrest has been recorded since April 2019 on the Montagne Pelée volcano located on Martinique. There have been only very few periods of seismic unrest since the last magmatic eruption of 1929−1932. This is therefore a rare opportunity to examine its origin. In April 2019 the number of shallow volcano-tectonic (VT) earthquakes increased drastically above the reference monthly rate of 19 VT/month and then exceeded it consistently for several months. Deep (> 10 km) VT events occurred at the onset of the unrest and harmonic tremor was first recorded in November 2020. Continuous Global Navigation Satellite System data reveal that a minor horizontal deformation began around mid-2021. The modeling of these data favors an inflation source located at about 1 km below and slightly SW of the summit, in the area of the hydrothermal system and where most of the shallow VT events are located. Zones of degraded and dead vegetation on the upper flanks of Montagne Pelée were detected with satellite imagery starting in November 2019 and shown to be associated with elevated passive CO2 soil degassing. This protracted unrest most likely reflects the ascent of a limited volume of deep magmatic fluids that reinvigorated the shallow hydrothermal circulation.

Identifying early indicators of volcanic eruptions is a fundamental part of natural hazard management but is notoriously difficult. Here we consider whether monitoring changes in glacier velocity can help. We use satellite images to investigate changes in the surface velocity of Cone Glacier (Alaska) between November 2017 and January 2022, a period encompassing two eruptions of Mount Veniaminof on which the glacier sits. Our data show high glacier velocities months prior to these eruptions and low velocities immediately before, during and after the 2018 eruption, likely caused by volcanically triggered ice melt and associated changes in subglacial water pressures. Evidence for elevated velocities months prior to eruptions is particularly important and indicates that glacier speed-up might be an early indicator of volcanic unrest. Thus, glaciers could serve as tools for volcano monitoring and eruption forecasting since more than 2500 glaciers globally are located within 5 km of an active volcano.