Langer, Horst

Infrasound monitoring plays an important role in the framework of the surveillance of Mt. Etna, Europe’s largest active volcano. Compared to seismic monitoring, which is particularly effective for buried sources, infrasound signals mirror the activity of shallow sources like Strombolian explosions or degassing. The interpretation of infrasound signals is difficult to the untrained eye, as we have to account for volcanic and non-volcanic sources. The problem of handling large and complex data sets can be tackled with machine learning, namely pattern recognition techniques. Here, we focus on so-called ‘Unsupervised Learning’, where we identify groups of patterns being similar to each other. The degree of similarity is based on a metric measuring the distance among the features of the patterns. This work aims at the identification of typical regimes of infrasound radiation and their relation to the state of volcanic activity at Mt. Etna. For this goal, we defined features describing any infrasound pattern. These features were obtained using wavelet transform. We applied ‘Self-Organizing Maps’ (SOM) to the features projecting them to a 2-D representation space—the ‘map’. An intriguing aspect of SOM resides in the fact that the position of the patterns on the map can be expressed by a colour code, in a manner that similar patterns are assigned a similar colour code. This simplified representation of multivariate patterns allows to follow the development of their characteristics with time efficiently. During a training phase we considered a reference data set, which encompassed a large variety of scenarios. We identified typical groups of patterns which correspond to a specific regime of activity, being representative of the state of the volcano or noise conditions. These groups form areas on the 2-D maps. In a second step, we considered a test data set, which was not used during the training phase. Applying the same pre-processing as for the training data, we blindly assigned the test patterns to the regimes found before, identifying the one whose colour code is most similar to the one calculated to the test pattern. We are thus able to assess the validity of the prediction. The classification scheme presented provides a reliable assessment of the state of activity and adds useful and supplementary details to the results of the real-time automatic system in operation at Istituto Nazionale di Geofisica e Vulcanologia—Osservarorio Etneo. This is of particular importance when no visible information of the volcanic activity is available either for unfavourable meteorological conditions or during night time.

The monitoring of the seismic background signal – commonly referred to as volcanic tremor - has become a key tool for volcanic surveillance, particularly when field surveys are unsafe and/or visual observations are hampered by bad weather conditions. Indeed, it could be demonstrated that changes in the state of activity of the volcano show up in the volcanic tremor signature, such as amplitude and frequency content. Hence, the analysis of the characteristics of volcanic tremor leads us to pass from a mere monoparametric vision of the data to a multivariate one, which can be tackled with modern concepts of multivariate statistics. For this aim we present a recently developed software package which combines various concepts of unsupervised classification, in particular cluster analysis and Kohonen maps. Unsupervised classification is based on a suitable definition of similarity between patterns rather than on a-priori knowledge of their class membership. It aims at the identification of heterogeneities within a multivariate data set, thus permitting to focalize critical periods where significant changes in signal characteristics are encountered. The application of the software is demonstrated on sample sets derived from Mt. Etna during eruptions in 2001, 2006 and 2007-8.

The study of the kinematics and stress field related to seismicity makes an important contribution to the understanding of tectonic processes. In this kind of analysis, a crucial issue is identifying seismically homogeneous areas, which implies data classification and cluster creation. We present an approach that combines unsupervised learning techniques in order to reveal patterns in the focal mechanisms data set. In particular, a combination of two popular clustering algorithms, that is, self-organizing maps and Fuzzy C-means, was applied to focal mechanisms of events located in the Central Mediterranean region, characterized by a complex geodynamic framework. The analysis allowed identifying eight groups of focal mechanisms and their spatial distribution in the crust, and revealing the tectonic style of key sectors of southern Italy and of the neighboring offshore areas. A compressive regime was found between the lower Tyrrhenian Sea and southeastern Sicily, whereas extension prevails along the Calabrian Arc and the southern Apennines. A NW-SE transcurrent faulting between the Aeolian Islands and the Ionian Sea forms a transfer zone between these two domains.

Some researchers view radon emissions as a precursor to earthquakes, especially those of high magnitude [e.g., Wang et al., 2014; Lombardi and Voltattorni, 2010], but the debate in the scientific community about the applicability of the gas to surveillance systems remains open. Yet radon “works” at Italy’s Mount Etna, one of the world’s most active volcanoes, although not specifically as a precursor to earthquakes. In a broader sense, this naturally radioactive gas from the decay of uranium in the soil, which has been analyzed at Etna in the past few years, acts as a tracer of eruptive activity and also, in some cases, of seismic–tectonic phenomena. To deepen the understanding of tectonic and eruptive phenomena at Etna, scientists analyzed radon escaping from the ground and compared those data with measurements gathered continuously by instrumental networks on the volcano. Here Etna is a boon to scientists—it’s traced by roads, making it easy to access for scientific observation. Dense monitoring networks, managed by the Istituto Nazionale di Geofisica e Vulcanologia, Catania–Osservatorio Etneo (INGV-OE), have been continuously observing the volcano for more than 40 years. This continuous dense monitoring made the volcano the perfect open-air laboratory for deciphering how eruptive activity may influence radon emissions.

With over fifty eruptive episodes (Strombolian activity, lava fountains, and lava flows) between 2006 and 2013, Mt Etna, Italy, underscored its role as the most active volcano in Europe. Seven paroxysmal lava fountains at the South East Crater occurred in 2007-2008 and 46 at the New South East Crater between 2011 and 2013. Month-lasting lava emissions affected the upper eastern flank of the volcano in 2006 and 2008-2009. On this background, effective monitoring and forecast of volcanic phenomena are a first order issue for their potential socio-economic impact in a densely populated region like the town of Catania and its surroundings. For example, explosive activity has often formed thick ash clouds with widespread tephra fall able to disrupt the air traffic, as well as to cause severe problems at infrastructures, such as highways and roads. For timely information on changes in the state of the volcano and possible onset of dangerous eruptive phenomena, the analysis of the continuous background seismic signal, the so-called volcanic tremor, turned out of paramount importance. Changes in the state of the volcano as well as in its eruptive style are usually concurrent with variations of the spectral characteristics (amplitude and frequency content) of tremor. The huge amount of digital data continuously acquired by INGV’s broadband seismic stations every day makes a manual analysis difficult, and techniques of automatic classification of the tremor signal are therefore applied. The application of unsupervised classification techniques to the tremor data revealed significant changes well before the onset of the eruptive episodes. This evidence led to the development of specific software packages related to real-time processing of the tremor data. The operational characteristics of these tools – fail-safe, robustness with respect to noise and data outages, as well as computational efficiency – allowed the identification of criteria for automatic alarm flagging. The system is hitherto one of the main automatic alerting tools to identify impending eruptive events at Etna. The currently operating software named KKAnalysis is applied to the data stream continuously recorded at two seismic stations. The data are merged with reference datasets of past eruptive episodes. In doing so, the results of pattern classification can be immediately compared to previous eruptive scenarios. Given the rich material collected in recent years, here we propose the application of the alert system to a wider range (up to a total of eleven) stations at different elevations (1200-3050 m) and distances (1-8 km) from the summit craters. Critical alert parameters were empirically defined to obtain an optimal tuning of the alert system for each station. To verify the robustness of this new, multistation alert system, a dataset encompassing about eight years of continuous seismic records (since 2006) was processed automatically using KKAnalysis and collateral software offline. Then, we analyzed the performance of the classifier in terms of timing and spatial distribution of the stations.

Eighteen paroxysmal episodes occurred on Mt Etna in 2011, and provided rich material for testing automatic procedures of data processing and alert systems in the context of volcano monitoring. The 2011 episodes represent a typical picture of activity of Mt Etna: in 2000 and 2001, before the 2001 flank eruption, more than one hundred lava fountains were encountered. Other major lava fountains occurred before the flank eruptions of 2002/03 and 2008. All these fountains, which are powerful but usually short lived phenomena, originated from the South-East Crater area and caused the formation of thick ash clouds, followed by the fallout of material with severe problems for the infrastructure of the metropolitan area of Catania. We focus on the seismic background radiation – volcanic tremor – which plays a key role in the surveillance of Mt Etna. Since 2006 a multi-station alert system has been established in the INGV operative centre of Catania exploiting STA/LTA ratios. Besides, it has been demonstrated that also the spectral characteristics of the signal changes correspondingly to the type of volcanic activity. The simultaneous application of Self Organizing Maps and Fuzzy Clustering offers an efficient way to visualize signal characteristics and its development with time, allowing to identify early stages of eruptive events and automatically flag a critical status before this becomes evident in conventional monitoring techniques. Changes of tremor characteristics are related to the position of the source of the signal. The location of the sources exploits the distribution of the amplitudes across the seismic network. The locations were extremely useful for warning throughout both a flank eruption in 2008 as well as the 2011 lava fountains, during which a clear migration of tremor sources towards the eruptive centres could be noticed in advance. The location of the sources completes the picture of an imminent volcanic unrest and corroborates early warnings flagged by the changes of signal characteristics. On-line data processing requires computational efficiency, robustness of the methods and reliability of data acquisition. The amplitude based multi-station approach offers a reasonable stability as it is not sensitive to the failure of single stations. The single station approach, based on our unsupervised classification techniques, is cost-effective with respect to logistic efforts, as only one or few key stations are necessary. Both systems have proven to be robust with respect to disturbances (undesired transients like earthquakes, noise, short gaps in the continuous data flow), and false alarms were not encountered so far. Another critical aspect is the reliability of data storage and access. A hardware cluster architecture has been proposed for failover protection, including a Storage Area Network system. We outline concepts of the software architectures which allow easy data access following predefined user policies. We envisage the integration of seismic data and those originating from other scientific fields (such as volcano imagery, geochemistry, deformation, gravity, magneto-telluric), in order to facilitate cross-checking of the findings encountered from the single data streams, in particular allowing their immediate verification with respect to ground truth.

The early-warning of a volcanic unrest requires continuous, reliable information from monitoring before volcanic activity starts. An optimal source of such information are seismic data, which overcome problems due to prohibitive conditions for field surveys or cloud cover that may hinder visibility. Given the large amount of digital data accumulating in short times, techniques of automatic pattern recognition are necessary in the context of effective extraction of information and data reduction. We designed a multi-station warning system based on pattern recognition techniques. In particular, a classification of patterns of volcanic tremor, the background seismic radiation, has been performed. Two unsupervised classifiers, Self-Organizing Maps (SOM) and fuzzy clustering were applied to automatically detect patterns which are typical footprints of an impending volcanic unrest. Plotting the SOM colors on DEM allows us their geographical visualization according to the stations of detection; this spatial location may give hints on areas potentially impacted by eruptive phenomena. The method implies continuous processing of recorded data streams; it was tested and tuned over year-long data streams on the base of eruptive phenomena occurred at Etna, Italy, in recent years. Here we present results of the application of the classifier, which forecasted in hindsight patterns associated with fast-rising magma (typical of lava fountains) as well as a relatively long lead time of the outburst (lava flows from eruptive fractures). The performance of the multi-station system was evaluated by using Receiver Operating Characteristics (ROC) curves; the result is indicative of a good detection accuracy that cannot be achieved from a mere random choice.

Over fifty eruptive episodes with Strombolian activity, lava fountains, and lava flows occurred at Mt Etna volcano between 2006 and 2013. Namely, there were seven paroxysmal lava fountains at the South-East Crater in 2007-2008 and 46 at the New South-East Crater between 2011 and 2013. Lava emissions lasting months affected the upper eastern flank of the volcano in 2006 and 2008-2009. Effective monitoring and forecast of such volcanic phenomena are particularly relevant for their potential socio-economic impact in densely populated regions like Catania and its surroundings. For example, explosive activity has often formed thick ash clouds with widespread tephra fall able to disrupt the air traffic, as well as to cause severe problems at infrastructures, such as highways and roads. Timely information about changes in the state of the volcano and possible onset of dangerous eruptive phenomena requires efficacious surveillance methods. The analysis of the continuous background seismic signal, the so-called volcanic tremor, turned out of paramount importance to follow the evolution of volcanic activity [e.g., Alparone et al., 2003; Falsaperla et al., 2005]. Changes in the state of the volcano as well as in its eruptive style are usually concurrent with variations of the spectral characteristics (amplitude and frequency) of tremor. The huge amount of digital data continuously acquired by INGV’s broadband seismic stations every day makes a manual analysis difficult. In order to tackle this problem, techniques of automatic classification of the tremor signal are applied. In a comparative study, the robustness of different methods for the identification of regimes in volcanic activity were examined [Langer et al., 2009]. In particular, Langer et al. [2011] applied unsupervised classification techniques to the tremor data recorded at one station during seven paroxysmal episodes in 2007-2008. Their results revealed significant changes in the pattern classification well before the onset of the eruptive episodes. This evidence led to the development of specific software packages, such as the program KKAnalysis [Messina and Langer, 2011], a software that combines an unsupervised classification method (Kohonen Maps) with fuzzy cluster analysis. The operational characteristics of these tools - fail-safe, robustness with respect to noise and data outages, as well as computational efficiency - allowed on-line processing at the operative centre of the INGV-Osservatorio Etneo in 2010 and the identification of criteria for automatic alarm flagging. The system is hitherto one of the main automatic alerting tools to identify impending eruptive events at Etna. The software carries out the on-line processing of the new data stream coming from two seismic stations, merged with reference datasets of past eruptive episodes. In doing so, results obtained for new data are immediately compared to previous eruptive scenarios. Given the rich material collected in recent years, we are able to apply the alert system to eleven stations at different elevations (1200-3050 m) and distances (1-8 km) from the summit craters. Critical alert parameters were empirically defined to obtain an optimal tuning of the alert system for each station. To verify the robustness of this new, multistation alert system, a dataset encompassing about eight years of continuous seismic records (since 2006) was processed automatically using KKAnalysis and collateral software off-line. Then, we analyzed the performance of the classifier in terms of timing and spatial distribution of the stations. We also investigated the performance of the new alert system based on KKAnalysis in case of activation of whatever eruptive centre. Intriguing results were obtained in 2010 throughout periods characterized by the renewal of volcanic activity at Bocca Nuova-Voragine and North-East Crater, and in the absence of paroxysmal phenomena at South-East Crater and New South-East Crater. Despite the low-energy phenomena reported by volcanologists (i.e., degassing, low-to moderate explosions), the triggered alarms demonstrate the robustness of the classifier and its potential: i) to identify even subtle changes within the volcanic system using tremor, and ii) to highlight the activation of a single eruptive centre, even though different from the one for which the classifier was initially tested. It is worth noting that in case of activation of weak sources, the successful performance of the classifier depends upon the general level of signals originating from other sources in that specific time span.

A multidisciplinary study on gas emission and volcanic tremor characteristics of Mt. Etna B. Behncke, S. Falsaperla, S. Giammanco, H. Langer, M. Neri, E. Pecora, G. Salerno Istituto Nazionale di Geofisica e Vulcanologia, Sezione di Catania, Osservatorio Etneo,P.zza Roma 2, 95125, Catania, Italy The 2008-2009 eruption of Mt. Etna was heralded by episodes of paroxysmal summit activity, with strong Strombolian activity and spectacular lava fountains and flows, starting from spring 2007. In this study, we present analyses of a three-month period (from February to April, 2007) which led to the first paroxysm. In doing so, we merge volcanic tremor data and gas measurements of SO2 and Radon. This multidisciplinary study allows characterizing a stage during which the volcano feeder was affected by fluid recharge, producing to repeated episodes of temporary increases in volcanic tremor amplitude, without any visible phenomenon at the surface. We investigate on these spurious changes in tremor characteristics and their relationship to gas emission. Ruling out other exogenous sources, we hypothesize that certain changes represented aborted eruptions, where the magma failed to reach the surface.

Mt Etna is a well monitored basaltic volcano for which high-quality, multidisciplinary data set are continuously available for around-the-clock surveillance. Particularly, volcano-seismic data sets cover decades long local recordings, temporally encompassing different styles of eruptive activity, from Strombolian eruptions to lava fountains and lava flows. Intense earthquakes swarms have often heralded effusive activity. However, from the seismic point of view, volcanic tremor has proved to be one of the most reliable indicators of impending eruptive activity. Indeed, changes in the volcano feeder show up in the signature of tremor, its spectral characteristics and source location. Some of us (Langer and Messina) have recently developed a new software for the classification of volcanic tremor data, combining Self Organizing Maps (also known as Kohonen Maps) along with Cluster and Fuzzy Analysis. This software allows us to analyse the background seismic radiation at permanent broadband stations located at various distance from the summit craters to identify transitions from pre-eruptive to eruptive activity. Throughout the analysis of the data flow, the software provides an unsupervised classification of the spectral characteristics (i.e., amplitude and frequency content) of the signal. The information embedded in the spectrum is interpreted to assign a specific state of the volcano. An application of this new software is proposed here on the eruptive events at Etna of 2007-2009, which consisted of 7 episodes of lava fountaining, periodic Strombolian activity at the summit craters, followed by lava emissions on the upper east flank of the volcano, with start on 13 May 2008 and end on 6 July 2009. In the study period the source of volcanic tremor was always shallow (less than 3 km) and within the volcano edifice. The upraise of magma to the surface was fast and associated with changes of volcanic tremor features, which covered time windows of variable duration from several hours to a few minutes. We discuss the possible reasons of such variability in the light of the characteristics of the overall seismicity preceding the eruptions in the study period, taking into account field observations and rheology of the ascending magma as well.