Long-period Events Research Articles

Event classification, seismicity, and eruption forecasting at Great Sitkin Volcano, Alaska: 1999–2023

The frequency content of volcanogenic seismicity is often used to classify events and their spatial and temporal progression is then used to map subsurface volcanic processes. The progression of volcano-seismic events and associated source processes also plays a critical role in eruption forecasting. Here we develop and evaluate a computerized methodology for characterizing volcano-seismic event types using Frequency Index and Average Peak Frequency. We apply and test this technique at Great Sitkin Volcano, Alaska, classifying over 9000 hypocenters between 1999 and 2023. This 24-year time span covers periods of seismic quiescence, earthquake activity on nearby tectonic (bookshelf) faults, precursory unrest from 2016 to 2021, and the explosive onset in May 2021 of the ongoing effusive eruption. We use the spatial and temporal evolution of classified event types to map the active volcanic and tectonic processes, develop a conceptual model of the subsurface magmatic system, and perform a retrospective analysis of eruption forecasts at Great Sitkin Volcano between 2016 and the present. The classification and progression of hypocenters suggests the subsurface Great Sitkin Volcano magmatic system consists of a mid- to lower- crustal source zone between 10 and 40 km depth and an upper crustal magma storage area between −1 and 10 km depth (hypocenter depth is referenced to sea level and negative depths reflect height above sea level). The earliest precursors occurred in July 2016 and consisted of deep long-period and volcano-tectonic earthquakes at mid-crustal depths suggesting the subsequent unrest and eruption were triggered by a deeper intrusion of magma. This mid-crustal seismic activity was immediately followed by the onset upper-crustal long-period events and volcano-tectonic earthquakes VTs suggesting a strong linkage between the shallow and deeper portions of the magmatic system. The upper crustal area was likely capped by the 1974 lava dome until the magmatic explosion on May 26, 2021.

Seismoacoustic Wavefield at Popocatépetl Volcano, Mexico, Captured by a Temporary Broadband Network from 2021 to 2022

Abstract Popocatépetl is a highly active stratovolcano in central Mexico with recurrent activity of Vulcanian-type explosions and frequent degassing. The proximity of Popocatépetl volcano to Mexico City, one of the most populated cities in the world, demands continuous monitoring to achieve an adequate volcano risk assessment. We present an overview of the first high-dynamic-range and high-broadband (0.01–200 Hz; 400 Hz sampling rate) seismoacoustic network (PoPiNet), which we operated around Popocatépetl volcano from August 2021 to May 2022. Here, we show preliminary results of the explosions recorded in September 2021. We deployed five seismoacoustic stations within 4–25 km horizontal distance (range) from the vent. We identify infrasonic waveforms associated with tremor and explosions, with pressures ranging from 16 to 134 Pa and dominant frequencies between 0.2 and 5.0 Hz. The frequency content of the recorded signals at the closest stations to the volcano spans the sub-bass (20–60 Hz) and bass (60–250 Hz) ranges. The associated seismic signals of moderate explosions exhibit air-to-ground coupled waves with maximum coherence values at frequencies up to 5 and 25 Hz for the farthest and closest stations to the volcano, respectively. Conversely, we observe infrasound signal amplitudes from relatively small explosions reaching maximum pressures of 10 Pa that do not couple into the ground, even at the closest stations. These infrasound signals are associated with type-I long-period events as reported in previous investigations. The waveform consistency suggests repetitive and nondestructive sources beneath the volcano.

Automated Detection of Volcanic Seismicity Using Network Covariance and Image Processing

Abstract Seismicity at restless volcanoes commonly features a variety of signal types reflecting both volcanotectonic and fluid-driven source processes. However, traditional catalogs of seismicity are often incomplete, especially concerning events with emergent onsets such as those driven by the dynamics of magmatic and hydrothermal fluids. The detection of all discrete events and continuous seismic tremors, regardless of the underlying source processes, would therefore improve the ability of monitoring agencies to forecast eruptions and mitigate their associated hazards. We present a workflow for generalized detection of seismic events based on the network covariance matrix (Seydoux et al., 2016). Our contributions enable the method to simultaneously detect continuous and short-duration (<∼10 s) events, provide information about the frequency content of the signals, and to refine the initial detection times by an order of magnitude (from window lengths of 75 to 7.5 s). We test the workflow on a 15-month record of seismicity with 23 stations at Mammoth Mountain, California (July 2012–October 2013) and detect 62% of long-period events and 94% of volcanotectonic events in the existing Northern California Earthquake Data Center catalog. In addition, ∼3000 events are not included in the catalog, and thousands of tremor signals are found. The method is suitable for near-real-time analysis of continuous waveforms and can provide a valuable supplement to existing algorithms to improve the completeness of catalogs used for monitoring volcanoes.

The 2015 earthquake swarm in the Fentale volcanic complex (FVC): A geohazard risk for Ethiopia's commercial route to the Djibouti port.

We investigate rare seismicity in the neighborhood of the Fentale volcano, main Ethiopian rift that occurred in spring 2015 using data recorded by the Ethiopian Seismic Station Network. Over 1350 earthquakes are located and a seismic swarm is observed to the northeast of Fentale volcano, with the largest earthquake having a magnitude of 4.3 ML. Several VT (volcano-tectonic) and LP (long-period) events are observed in the swarm, indicating that it is induced by magmatic intrusion, consistent with dike-induced deformation in the same area inferred from satellite geodesy. The seismic activity commenced at about 40 km depth and migrated southwestward as it shallowed, towards Fentale volcano, suggesting that the dike feeding system might be influenced by the magma rich Afar Depression further to the northeast rather than the adjacent Fentale volcano. An earlier cluster of seismicity in the same area was recorded by a temporary deployment of seismic stations that ran from November 2001 to January 2003; extensive clusters of less well recorded seismic activity in the region occurred in June 1989, January–February 1981, October–November 2003, and 2004. Such frequent seismicity at the Fentale Volcanic Complex coupled with the adjacent fast-growing Lake Basaka poses an eminent and significant risk to Ethiopia's access to Djibouti port, its most important commercial route. This should be a wakeup call to concerned stakeholders to take precautionary measures by developing risk management and rapid response strategies.

Bridging Supervised and Unsupervised Learning to Build Volcano Seismicity Classifiers at Kilauea Volcano, Hawaii

Abstract Real-time classification of volcano seismicity could become a useful component in volcanic monitoring. Supervised learning provides a powerful means to achieve this but often requires a large amount of manually labeled data. Here, we build supervised learning models to discriminate volcano tectonic events (VTs), long-period events (LPs), and hybrid events in Kilauea by training with pseudolabels from unsupervised clustering. We test three different supervised models, and all of them achieve >93% accuracy. We apply the model ensemble to the six-day seismicity during the eruption in 2018 and show that they were mainly VTs (62%), in comparison with the dominance of LPs prior to the eruption (68%). The success of our method is aided by the accuracy of the majority of pseudolabels and the consistency of the three models’ performance. Using Shapley additive explanations, we show that the frequency contents at 1–4 Hz are the most important to differentiate volcano seismicity types. This work, together with our previous clustering analysis, provides an example of bridging unsupervised and supervised learning to construct potential real-time seismic classifiers from scratch.

Temporal evolution of the magmatic system at Nevado del Ruiz Volcano (Colombia) inferred from long-period seismic events in 2003–2020

Long-period (LP) seismic events have been broadly assumed to be generated by oscillations of crack fluid-filled resonators. We investigated the waveforms and spectral characteristics of LP seismic events recorded by five stations at Nevado del Ruiz Volcano, Colombia. The observed event types of LP events exhibited decaying harmonic oscillations with either monochromatic or polychromatic spectral features (NLP events) or events characterized by non-decaying harmonic oscillations (BLP events). The NLP events occurred over four periods: September 2003 (period I), January 2004–October 2004 (period II), October 2010–March 2012 (period III), and October 2019–April 2020 (period IV). Their source locations were determined to be within the first kilometer of the conduit placed below the Arenas crater. Their locations involved a distinct resonator for each of the four periods but a single resonator for all the events during a single period. Assuming a crack geometry at their sources and based on their locations, we considered a dusty gas (mixture of ash and water vapor) filling a crack resonator. We estimated the crack model parameters to investigate the temporal processes generating variations in frequency and quality factor Q. Then, the estimated parameters indicate that NLP events were triggered by magma fragmentation. We found that events from period III preceded phreatomagmatic eruptions, increased ash and gas emissions and SO2 flux, and increased overall seismic activity. Three years after this period III, a dome was emplaced at the base of the Arenas crater. Besides, events from period IV also preceded by ~3 years a significant increase in the overall seismic activity, which led to a change in alert activity at the volcano. We argue that BLP events were driven by resonances provoked by the unstable degassing through the uppermost magma plug's open cracks (tuffisite veins).In contrast, NLP events occurred by overpressure released through fractures along the conduit's walls (shear fractures). The shear fractures at the conduit wall were in a region of high strain rate where a decrease in confining pressure triggered fragmentation, resulting in dusty gas and, subsequently, a resonance of the crack. NLP and BLP events were in a narrow relationship with the ascent of vesiculated magma to the surface. Our results show that the LP events help diagnose the state of magmatic fluids and understand the eruption processes of Nevado del Ruiz Volcano. LP events are not always followed by explosive eruptions, but they do indicate significant overpressure developing in the shallow conduit.

The 2013–2020 seismic activity at Sabancaya Volcano (Peru): Long lasting unrest and eruption

Sabancaya volcano is the youngest and second most active volcano in Peru. It is part of the Ampato-Sabancaya volcanic complex which sits to the south of the ancient Hualca Hualca volcano and several frequently active faults, thus resulting in complex volcano-tectonic interactions. After 15 years of repose, in 2013, a series of 4 earthquakes with magnitude >4.5 occurred within 24 h, marking the beginning of a new episode of unrest. Several additional swarms of earthquakes occurred in the following years until magmatic eruptive activity started on 6 November 2016. This activity is ongoing as of this writing, with an average of 50 explosions per day. In this study, we present results of multiparametric monitoring of Sabancaya's activity observed during 2013–2020. Seismic data are used to create a one-dimensional seismic velocity model, to catalog, locate, and characterize earthquakes, to detect repeating earthquake families, and to monitor seismic velocity variations by ambient noise cross-correlation. These analyses are complemented by visual and remote sensing observations and ground deformation measurements. All monitored parameters showed significant changes on 6 November 2016, the day of eruption onset, thus dividing the eruptive activity into pre-eruptive and eruptive stages.The unrest is characterized by high levels of seismic activity with hundreds of events detected per day. Volcano-tectonic (VT) earthquakes were dominant during the pre-eruptive period while long-period (LP) events and explosions have been most numerous since the eruption onset. Earthquake locations highlight long-lasting seismogenic zones along multiple previously active regional faults, as well as along newly identified faults. This VT seismicity is mainly distributed in a sector from the northwest to the east of the volcanic complex at distances of up to 30 km from the crater. We focus our analysis on two eruptive episodes: the eruption onset and subsequent crater migration from south to north, and the increase of lava dome extrusion rate in 2019. Both episodes are accompanied by seismic velocity decreases of up to 0.2% and are preceded by a few weeks by bursts of distal VT activity, including numerous repeating earthquakes. These repeated events were located on several remote tectonic faults (5–25 km from the vent). We suggest that these phenomena could be due to the injection of a batch of magma in the deep reservoir and/or conduit, which would generate 1) a pressure wave propagating in the hydrothermal system, triggering the bursts of seismic activity and 2) slow rising of magma by melting old material filling the conduit that eventually produced the eruptive and dome growth acceleration events.

Seismicity during the recent activity (2009–2020) of Turrialba volcano, Costa Rica

Turrialba is a stratovolcano located at the easternmost part of the Costa Rican volcanic front. After remaining quiescent for more than a century, in 1996 it started to show signs of unrest, until a first phreatomagmatic explosion occurred on January, 2010. Since then, the activity evolved from phreatic to magmatic, in a series of distinct eruptive phases. In this paper, we investigate the seismic records that span the whole eruptive process (2010-present), in order to identify precursory signals and characterize the volcanic evolution. A long-term analysis was carried out based on the continuous records, as well as seismic catalogs (volcano-tectonic seismicity, harmonic tremor, etc.). In addition, the gradual character of the evolution of this eruption allowed for the analysis of independent precursory stages. Thus, we inspected in detail the most important of those periods, particularly, prior to the first 2010 phreatomagmatic eruption, and prior to the 2016 transition to an open vent system. Temporary tremor amplitude decreases were found to precede most of the eruptive phases. In total, 9 pre-eruptive tremor abatement periods were identified spanning several days (5–44), which often concurred with a decrease in the SO2 flux. The analysis of the volcano-tectonic seismicity highlights the migration of magma from a deep (6–10 km) reservoir beneath the neighboring Irazú volcano towards Turrialba volcano, especially between the years 2015 and 2016. This activity peaked on December 2016 when a Mw 5.5 earthquake took place between both volcanoes. Harmonic tremor episodes thrived in the later phase when the system finally opened (2017–2018). In the short-term, compounded tonal seismic signals were identified as precursor events, such as long-period events followed by harmonic tremor or by a multichromatic coda similar to tornillo-type events. The co-occurrence of tremor amplitude decreases and tonal seismic signals is interpreted to be caused by a sealing of the hydrothermal system, which blocked the circulation of fluids and permitted the resonances in the inner cavities. This process leaded to pressure accumulation and the consequent eruptions. Thus, trough a series of cycles of sealing and rupture the system of conduits gradually opened. The seismic characterization of this eruption constitutes insightful knowledge useful for monitoring and risk assessment purposes.

Performance of a Rotational Sensor to Decipher Volcano Seismic Signals on Etna, Italy

AbstractVolcano‐seismic signals such as long‐period events and tremor are important indicators for volcanic activity and unrest. However, their wavefield is complex and characterization and location using traditional seismological instrumentation is often difficult. In 2019 we recorded the full seismic wavefield using a newly developed 3C rotational sensor co‐located with a 3C traditional seismometer on Etna, Italy. We compare the performance of the rotational sensor, the seismometer and the Istituto Nazionale di Geofisica e Vulcanologia‐Osservatorio Etneo (INGV‐OE) seismic network with respect to the analysis of complex volcano‐seismic signals. We create event catalogs for volcano‐tectonic (VT) and long‐period (LP) events combining a STA/LTA algorithm and cross‐correlations. The event detection based on the rotational sensor is as reliable as the seismometer‐based detection. The LP events are dominated by SH‐type waves. Derived SH phase velocities range from 500 to 1,000 m/s for LP events and 300–400 m/s for volcanic tremor. SH‐waves compose the tremor during weak volcanic activity and SH‐ and SV‐waves during sustained strombolian activity. We derive back azimuths using (a) horizontal rotational components and (b) vertical rotation rate and transverse acceleration. The estimated back azimuths are consistent with the INGV‐OE event location for (a) VT events with an epicentral distance larger than 3 km and some closer events, (b) LP events and tremor in the main crater area. Measuring the full wavefield we can reliably analyze the back azimuths, phase velocities and wavefield composition for VT, LP events and tremor in regions that are difficult to access such as volcanoes.

Volcanic tremor and long period events at Mt. Etna: Same mechanism at different rates or not?

Volcanic tremor and long period (LP) events are typical seismic signals recorded on active volcanoes and are characterized by different durations, longer than minutes and a few seconds - tens of seconds for the former and latter, respectively. As they share the same frequency content, they are often grouped together in the literature and referred to by the unique name of LP seismicity. The common spectral features, together with observations in some volcanoes of individual LP events merging to form volcanic tremor, led to hypotheses that LP events and volcanic tremor share the same source mechanism. However, it is still open to debate whether volcanic tremor can be considered a simple coalescence of LP events or not. In this work, to help answer such a question, we analysed volcanic tremor and LP events recorded at Mt. Etna during the period February 2019–June 2020, characterized by minor eruptive activity, varying from weak ash emission to explosive and effusive eruptions at all the summit craters. Results from spectral, amplitude and location analyses, as well as the different scaling laws explaining the distributions of the duration/number of events versus size, led us to infer that LP events and volcanic tremor at Mt. Etna are not due to a common source mechanism.

Conduit resonance models for long-period seismicity at Galeras volcano (Colombia), during 2004–2010

This study analyzes the long-period (LP) seismicity of Galeras volcano from the period 2004–2010, based on resonance models of a magma column. In this model an initial disturbance is propagated through the magma column's walls in the form of waves until it reaches a receiver located on the surface. The input parameters for the crust and magma were obtained from previous studies, and the solution of the systems of equations was found using a finite element method. Models for two groups of LP events were found: group G1 associated with the emplacement of the lava domes in 2006 and 2008, and group G2 related to the seismicity preceding the emplacement of these domes. Group G1 is modeled by the resonance of a magma column about 2800 m long with its top near the surface and group G2 is related to a column close to 2000 m in length. The main frequency of initial perturbation differs between the groups. Additionally, the results of this study cast doubt on the effectiveness of location methods of fluid-related seismicity based on amplitude attenuation.

LP or VT signals? How intrinsic attenuation influences volcano seismic signatures constrained by Whakaari volcano parameters

Long Period (LP) seismicity is considered a key indicator of fluid migration in the volcanic edifice, with a source mechanism typically ascribed to fluid resonance in the volcano conduit and fractures. Our viscoelastic numerical wave propagation models at laboratory and field scales are based on quantitative estimates of wave speeds and intrinsic attenuation from Whakaari rocks saturated with a range of fluid types. Such models display path effects in homogeneous and layered media that influence the frequency content and duration of the seismic waveforms. We find that volcano-tectonic (VT) earthquakes can resemble short-duration LP events when propagating through fully or partially gas-saturated ash tuff deposits. Meanwhile, in a shallow volcanic subsurface comprising low-velocity, attenuative ash tuff and high-velocity, low-attenuation lava layers, VT signals may be distorted to resemble long-duration LP events. Such geological models are built based on interpretations of the shallow hydrothermal system at Whakaari volcano, and discussed in the context of other volcanoes around the world.

Hydrogeochemical temporal variations related to changes of seismic activity at Tenerife, Canary Islands

This work presents the results of a hydrogeochemical study conducted in two galleries on the island of Tenerife: Fuente del Valle and San Fernando, during a 2-year period with weekly sampling (July 2016 to July 2018). Groundwater temperature, pH, and electrical conductivity exhibited stable values throughout the study period, while significant temporal variations were observed in the total alkalinity, SO42−/Cl− ratio (in meq·L−1), and isotopic composition (δ18O and δD). δ13CTDIC values showed an endogenous signature for both galleries, while δ18O and δD values indicated a meteoric origin with interaction with endogenous gases, mainly CO2. Changes in isotopic ratios coincided with variations observed in alkalinity and SO42−/Cl− ratio. On October 2, 2016, a seismic swarm of long-period events was recorded on Tenerife followed by a general increase of the seismic activity in and around the island. A correlation was observed between some hydrogeochemical parameters in the groundwaters of the galleries related to observed changes of the seismic activity.

Aseismic strain episodes at Campi Flegrei Caldera, Italy

Abstract. Since 2004 a research project has been developed to monitor subsurface deformation of Italian volcanoes using borehole strainmeters and long-baseline tiltmeters. Six Sacks-Evertson dilatometers were installed around Campi Flegrei caldera and Vesuvius during 2004–2005 (Scarpa et al., 2007), and in 2008 these instruments were supplemented by two arrays of 28–280 m long water-tube tiltmeters in underground tunnels. Relevant strainmeter and tiltmeter data have been collected and analysed from the instruments installed near Campi Flegrei caldera during the recent unrest episodes. In the period 2004–2005 strain, tilt and GPS data from Campi Flegrei indicate the onset of surface deformation that accompanied a low rate of vertical displacement that continued to 2006, corresponding to an increase of CO2 emission. This strain episode preceded caldera microseismic activity by a few months, as was observed also during a significant inflation episode in 1982. Other transient strain episodes occurred in October 2006, which were accompanied by a swarm of VT (Volcano-Tectonic) and LP (Long Period) events, in 2009, at the time of renewed gas emission activity at Solfatara, and again in March 2010, several minutes before a seismic swarm. The time scale of these transient strain events ranges from some hours to several days, putting tight constraints on the origin of ground uplifts at Campi Flegrei. Their location is compatible with a source inferred from long term deformation signals, located about 4 km beneath Pozzuoli. A proposed mechanism for these aseismic strain episodes is that they are associated with magma growth in reservoirs with occasional pressure relief associated with the leakage of gas.

The 2002–2005 Changbaishan Volcanic Unrest Triggered by the 2002 M 7.2 Wangqing Deep Focus Earthquake

One of the most active intraplate volcanoes in East Asia, Changbaishan volcano experienced unrest from July 2002 to July 2005. On 2002/06/28, the M 7.2 Wangqing deep-focus earthquake occurred ∼290 km northeast of Changbaishan volcano. While some studies have suggested a possible triggering relationship, the physical mechanism of such distant interaction is still not well understood. Using a template matching technique, which cross-correlates waveform of known events with continuous data, we perform systematic detection of microseismic events recorded by station CBS near Changbaishan volcano from July 1999 to July 2007. The detected earthquakes can be further categorized into three different types: volcano-tectonic (VT) events, long-period (LP) events and harmonic-spectra (HS) events. We detect 3763 VT events between July 2002 and July 2007. The intense VT earthquake swarm during the period from July 2002 to July 2005, along with recurring LPs and HSs and other geodetic/geochemical evidence, suggest magma movement during unrest. Compared with the hand-picked catalogue, the catalogue obtained by template matching technique reveals a delayed-triggering relationship between Wangqing deep-focus earthquake and unrest. The small magnitudes of the VT events and the limited numbers of LP and HS events suggest that the Wangqing mainshock likely triggered bubble excitation in the mid-crust magma system, resulting in overpressure and a small magma injection into the shallow magma chamber at a depth of ∼5 km, leading to the 3-years unrest.

クラック固有振動の数値計算に基づくガレラス山（コロンビア）での低周波地震の励起過程推定

クラック固有振動の数値計算に基づくガレラス山（コロンビア）での低周波地震の励起過程推定

ガレラス火山（コロンビア）における低周波地震の発生過程：減衰調和振動型（N型）と非調和振動型（B型）の違いについて

ガレラス火山（コロンビア）における低周波地震の発生過程：減衰調和振動型（N型）と非調和振動型（B型）の違いについて

Volcanic seismicity beneath Chuginadak Island, Alaska (Cleveland and Tana volcanoes): Implications for magma dynamics and eruption forecasting

Cleveland and Tana are remote volcanoes located in the central Aleutian volcanic arc on the eastern end of the Islands of Four Mountains (IFM). The persistently active Mount Cleveland volcano, on the western side of Chuginadak Island, is surrounded by several closely spaced Quaternary volcanic centers including Carlisle, Herbert, Kagamil, Tana, and Uliaga, and numerous small satellite vents on Chiginadak between Cleveland and Tana. The Alaska Volcano Observatory (AVO) installed two permanent broadband seismometers on Chuginadak Island in 2014, and we operated a temporary broadband network focused on the western side of the island in 2015–2016. Collectively, these stations provided the first seismic observations of this frequently active volcano and the surrounding Holocene-aged volcanic vents. During the study period (July 2014–January 2019), eruptive activity at Cleveland was characterized by small explosions separated by periods of lava effusion that formed small domes in the volcano's summit crater. We characterize seismicity beneath Chuginadak Island through automated analysis of event waveform frequency content, development of a one-dimensional P-wave velocity model, calculation of earthquake hypocenters, magnitudes, focal mechanisms, and identification of earthquake families. This analysis reveals the full range of seismic event types expected in a highly active volcanic environment and includes Volcano-Tectonic (VT) earthquakes, Long-Period (LP) events, and explosion signals. LP events appear to cluster at shallow depth beneath the active crater of Mount Cleveland and almost all of the explosions occur without identifiable short-term (hours to days) seismic precursors. VT earthquakes beneath Mount Cleveland occur at depths of 2 to 8 km below sea level (BSL) and range in magnitude from −0.2 to 1.8. VT focal mechanisms have horizontal P-axes that align with the regional axis of maximum stress. These observations, and a relatively slow one-dimensional seismic velocity model, are consistent with a shallow body of magma that is fed through a deeper conduit system. The time-history of VT earthquakes and shallow LP events suggest their occurrence may track the transfer of magma and fluids from the mid-crust to the shallow portions of the conduit system and may provide a means to anticipate future explosions and periods of dome growth. VT hypocenters also extend ~7 km northeast of Cleveland's summit at depths of 5 to 10 km BSL, under a group of Holocene-aged vents between Mount Cleveland and Tana. These earthquakes have vertically-oriented P-axes and a greater percentage occur in families. These observations, combined with observations of vent orientation and morphology and gas flux, suggest the area between Cleveland and Tana represents a zone of complicated volcano-tectonic interaction, similar to calderas elsewhere in the Aleutian arc. The presence of a larger volcanic system in the eastern IFM could influence magmatism and account for the multiple closely spaced volcanic centers in this region.

Chapter 7.3 Mount Melbourne and Mount Rittmann

Abstract Mount Melbourne and Mount Rittmann are quiescent, although potentially explosive, alkaline volcanoes located 100 km apart in Northern Victoria Land quite close to three stations (Mario Zucchelli Station, Gondwana and Jang Bogo). The earliest investigations on Mount Melbourne started at the end of the 1960s; Mount Rittmann was discovered during the 1988–89 Italian campaign and knowledge of it is more limited due to the extensive ice cover. The first geophysical observations at Mount Melbourne were set up in 1988 by the Italian National Antarctic Research Programme (PNRA), which has recently funded new volcanological, geochemical and geophysical investigations on both volcanoes. Mount Melbourne and Mount Rittmann are active, and are characterized by fumaroles that are fed by volcanic fluid; their seismicity shows typical volcano signals, such as long-period events and tremor. Slow deformative phases have been recognized in the Mount Melbourne summit area. Future implementation of monitoring systems would help to improve our knowledge and enable near-real-time data to be acquired in order to track the evolution of these volcanoes. This would prove extremely useful in volcanic risk mitigation, considering that both Mount Melbourne and Mount Rittmann are potentially capable of producing major explosive activity with a possible risk to large and distant communities.

A machine-learning approach for automatic classification of volcanic seismicity at La Soufrière Volcano, Guadeloupe

The classification of seismo-volcanic signals is performed manually at La Soufrière Volcano, which is time consuming and can be biased by subjectivity of the operator. We propose here a machine-learning-based model for classification of these signals, to handle large datasets and provide objective and reproducible results. To describe the properties of the signals, we used 104 statistical, entropy, and shape descriptor features computed from the time waveform, the spectrum, and the cepstrum. First, we trained a random forest classifier with a dataset provided by the Observatoire Volcanologique et Sismologique de Guadeloupe that consisted of 845 labeled events that were recorded from 2013 to 2018: 542 volcano-tectonic (VT); 217 Nested; and 86 long period (LP). We obtained an overalll accuracy of 72%. We determined that the VT class includes a variety of signals that cover the VT, Nested and LP classes. After visual inspection of the waveforms and spectral characteristics of the data set, we introduced two new classes: Hybrid and Tornillo. A new random forest classifier was trained with this new information, and we obtained a much better overall accuracy of 82%. The model is very good for recognition of all event classes, except Hybrid events (67% accuracy, 70% precision). Hybrid events are often considered to be a mix of VT and LP events. This can be explained by the nature of this class and the physical processes that include both fracturing and resonating components with different modal frequencies. By analyzing the feature weights and by training a model with the most important features, we show that a subset of the 14 best features is sufficient to obtain a performance that is close to that of the model with the whole feature set. However, these best features are different from the 13 best features obtained for another volcano in Peru, with only one feature common to both sets of best features. Therefore, the model is not universal and it must be trained for each volcano, or it is too specific to the one station used here.