Pyroclast Tracking Velocimetry illuminates bomb ejection and explosion dynamics at Stromboli (Italy) and Yasur (Vanuatu) volcanoes



A new image processing technique—Pyroclast Tracking Velocimetry—was used to analyze a set of 30 high-speed videos of Strombolian explosions from different vents at Stromboli (Italy) and Yasur (Vanuatu) volcanoes. The studied explosions invariably appear to result from the concatenation of up to a hundred individual pyroclast ejection pulses. All these pulses share a common evolution over time, including (1) a non-linear decrease of the pyroclast ejection velocity, (2) an increasing spread of ejection angle, and (3) an increasing size of the ejected pyroclasts. These features reflect the dynamic burst of short-lived gas pockets, in which the rupture area enlarges while pressure differential decreases. We estimated depth of pyroclast release to be approximately 1 and 8 m below the surface at Stromboli and Yasur, respectively. In addition, explosions featuring more frequent pulses also have higher average ejection velocities and larger total masses of pyroclasts. These explosions release a larger overall amount of energy stored in the pressurized gas by a combination of more frequent and stronger ejection pulses. In this context, the associated kinetic energy per explosion, ranging 103–109 J appears to be a good proxy for the explosion magnitude. Differences in the pulse-defining parameters among the different vents suggest that this general process is modulated by geometrical factors in the shallow conduit, as well as magma-specific rheology. Indeed, the more viscous melt of Yasur, compared to Stromboli, is associated with larger vents producing fewer pulses but larger pyroclasts.

1 Introduction

Strombolian activity is characterized by frequent (minutes), impulsive release of magmatic products (pyroclasts and gases) through mild explosions [Cashman and Sparks, 2013] and is typical of mafic volcanoes with a low- to medium-viscosity magma. These explosions have been attributed to the burst of large pressurized gas pockets [e.g., Sparks, 1978; Jaupart and Vergniolle, 1988; Burton et al., 2007; James et al., 2008]. The gas, exsolved from the magma at depth, forms bubbles that grow, coalesce, and rise into the conduit where they keep growing due to decompression. Inside the conduit, the gas may organize in slugs, i.e., elongated bubbles (up to 500 m long) with diameters approaching that of the conduit [Vergniolle and Jaupart, 1986; James et al., 2008; Del Bello et al., 2012; Gonnermann and Manga, 2013]. Alternatively, the gas may reach the surface as a series of small bubbles [Bani et al., 2013]. Close to the surface, these pressurized bubbles or slugs burst, causing the ejection of molten pyroclasts and a variable quantity of ash into the air [Chouet et al., 1974]. In order to characterize the Strombolian activity of volcanoes, various studies describe the explosions and their products, including, e.g., description of the process by thermal imagery [Patrick, 2007; Sahetapy-Engel and Harris, 2009; Harris et al., 2012], estimation of pressure release from the waveform of the associated acoustic signal [Ripepe et al., 2013; Medici et al., 2014], and estimation of the intensity of the explosion from the monitoring network video footage [Taddeucci et al., 2013].

Recent studies, with higher time resolution, revealed that Strombolian explosions are, in most cases, not the result of a single burst, but a more complex process. By studying the ejection of pyroclasts using high-speed cameras, Taddeucci et al. [2012b] pointed out that each explosion at Stromboli volcano (Italy) is composed of several short ejection pulses, lasting ~1 s or less, thus emphasizing the need for high temporal resolution observations to better constrain the mechanism of Strombolian explosions. These newly defined ejection pulses were used to define physical conditions and dynamic evolution of Strombolian explosions [Taddeucci et al., 2012a]. However, their systematic parameterization is still at an early stage, because of the inherent difficulties in data processing of high-speed imagery.

In this paper, the ejection of bombs during Strombolian explosions at Yasur (Vanuatu) and Stromboli (Italy) volcanoes is analyzed by applying pyroclast tracking velocimetry (PyTV) to high-speed optical videos. The PyTV is a suite of image analysis tools developed and calibrated [Gaudin et al., 2014] in order to extract from videos the most complete database of the velocity, angle, and size distributions of bomb-sized pyroclasts produced by Strombolian explosions. Compared to previous studies, more than a factor of ten more pyroclasts are tracked, enabling the detailed structure of Strombolian explosions to be revealed with unprecedented spatial and temporal resolution. The relationships between the ejection parameters provide new constraints to the burst mechanisms, highlighting differences and variability, as well as similarities in the vent characteristics and between the two study volcanoes.

2 Data Acquisition

2.1 Study Volcanoes

Two volcanoes presenting a persistent Strombolian activity were targeted in this study. Yasur is a low relief volcano (361 m a.s.l.) located on Tanna Island (Vanuatu), 150 km east of the New Hebrides trench [Louat et al., 1988]. The summit of the edifice hosts an irregularly shaped depression (resulting from two or three nested craters) consisting of three vent areas called A, B, and C from south to north (Figure 1, top) [Nabyl et al., 1997; Oppenheimer et al., 2006]. The number and location of individual vents within each area are highly variable. Explosions, subdivided into high-energy and low-energy classes, recur at a frequency of ten explosions per hour or more [Bani et al., 2013].

Figure 1.

Views (both facing North-West) of the different eruptive vent areas at Yasur and Stromboli at observation time. Insets show the location of the two volcanoes.

Stromboli volcano (Aeolian Islands) is located between the Southern Tyrrhenian Sea back arc and the Calabrian Arc forearc region [Peccerillo, 2001; Bonaccorso et al., 2003]. Within the 926 m a.s.l. high Stromboli island, eruptive activity takes place at a crater terrace located at about 700 m a.s.l. The terrace hosts three main vent areas [Harris and Ripepe, 2007], persistent since at least 1776 [Washington, 1917] and named according to their geographical position (South-West, hereafter SW, Central, and North-East, hereafter NE; see Figure 1, bottom). As at Yasur, the number and position of individual vents, and their morphologies, are highly variable [Harris and Ripepe, 2007] at the scale of days to weeks. Strombolian activity at these vents recurs at a frequency ranging from a few to a few tens of explosions per hour, with average explosion intensity being directly proportional to average frequency [Taddeucci et al., 2013].

2.2 Measurement Setup

High-speed videos were recorded at Stromboli by a NAC HotShot 512 high-speed camera (512 × 512 pixels definition; 500 Hz frame rate; 32 s recording time [Taddeucci et al., 2012b]) and at Yasur with an Optronis CR600x2 camera (1280 × 1024 pixels definition; 500 Hz frame rate; 14 s recording time). Both sensors are sensitive in the visible and near-infrared wavelengths, so that glowing bombs appear brighter than the background, and the gas mixture remains relatively transparent.

On both volcanoes, the observation point was located about 300 m from the vents (Figure 1). At this distance, our 300 mm focal length lens provides a resolution of about 2 cm/pixel, (see Table 1 for specific values). The exposure time and the diaphragm aperture were set to optimize the contrast between the pyroclasts and the background in each explosion. Of the 24 videos recorded at Stromboli and the 21 videos from Yasur, 5 and 11 were discarded, respectively (Table 1) because of the contrast between the pyroclasts and the background was insufficient, preventing the PyTV algorithm from running effectively, or because two vents were erupting at the same time. In some videos, a too-high pyroclast concentration results in saturation of the images.

Table 1. Characteristics of the High-speed Videos of the Strombolian Explosions Investigated in This Study
VentDateDistanceaPixel SizeNo. of Explosions Analyzed/RecordedNo. of Pulses Analyzed/ObservedNotes
  1. a

    Measured by laser telemeter.

  2. b

    In some videos, multiple vents were erupting at the same time. These videos are discarded from the analysis of individual explosions.

SW117/06/09320 m0.021 m4/574/220Long-lasting sequence of pulses with similar velocities
SW227/10/09290 m0.018 m5/655/299Short-lived sequence of high-frequency pulses with strong velocity fluctuations
NE119/06/09339 m0.022 m7/1085/297High-frequency pulses with very high velocity at the explosion onset
NE218/06/09312 m0.020 m3/315/19Short-lived groups of few, low-velocity pulses.
B1b10/07/11268 m0.0136/1544/93Few, well-separated, and large pulses preceded and followed by smaller ones
B12/07/11268 m0.013 m5/641/59Variable number of pulses reaching high velocities

At Yasur, we focused on the activity of vent area B, on 10 July 2011 and 12 July 2011, while the activity was at alert level 2 on a scale ranging from 0 to 4. Both vent morphology and explosion dynamics changed on a short time scale (a few hours) during the observation period. The vent diameter was approximately 10 m. Vents in area B were hidden by the crater rim and inferred to be located a few meters below the field of view. On 10 July 2011, area B hosted two vents, about 15 m apart (B1 and B2 from east to west). Observations mainly focused on the most active vent, B1, even if activity from vents A and B2 was also visible in some videos. On 12 July 2011, area B featured a single vent (hereafter denoted vent B). Filmed explosions are mostly of Type 1 of Patrick [2007], i.e., with no significant ash component, but a few are gradational to Type 2a, i.e., still dominated by the ejection of bombs but with an appreciable amount of ash. In addition to the filming, we also sampled pyroclasts from the ongoing activity to assess their chemical composition.

At Stromboli, we focused on the activity of two vents of the NE vent area, filmed on 18 June 2009 and 19 June 2009, and two in the SW vent area, filmed on 17 June 2009 and 27 October 2009 [also studied by Taddeucci et al., 2012a, 2012b] from the Pizzo Sopra la Fossa viewpoint. During these periods the activity was at a low to average level []. Unlike Yasur, in the relatively short filming intervals (less than 4 h) the vents, either in direct view of the camera or inferred to be a few meters below it, displayed no significant change in the morphology and activity. Vent diameter in all cases is inferred from the videos to range from 2 to 3 m. All filmed vents ejected incandescent bombs with no significant ash component (Type 1 explosions of Patrick [2007]), with the only exception being the NE vent area on 18 June 2009, which produced some ash at the beginning of the explosion, partly masking coarser bombs in the videos (Type 2a of Patrick [2007]) Both at Stromboli and at Yasur, we observed bombs deforming plastically in-flight, in good agreement with the previous studies on their viscosity [Gurioli et al., 2014].

3 Data Processing

Bright pyroclasts viewed against a darker background are tracked on the videos by using the pyroclast tracking velocimetry (PyTV) algorithm [Gaudin et al., 2014]. After a preprocessing step that removes the background from the images, a gray level threshold is applied on each frame in order to outline the pyroclasts. The pyroclasts are tracked by using image features, i.e., pixels associated with a high gray level gradient that remains constant with time. The displacement of these features in two successive frames is estimated by solving the optical flow equation [Moroni and Cenedese, 2005; Shindler et al., 2012]. Each pyroclast is then associated with its closest feature, making it possible to compute the evolution of its position in different frames and the reconstruction of its complete track.

The final product of the application of PyTV and subsequent post-processing is a Lagrangian description of the Strombolian explosions, where the evolution of the position, velocity, and angle over time of each pyroclast is estimated. In order to study the time evolution of the explosion, a subset of this database is extracted by associating each pyroclast with a single velocity and angle value, which are computed at the time that it rises through a horizontal detection line (Eulerian description). The height of this detection line, hereafter denoted hd, was set to 4 m above the vent as the best compromise between being close enough to the vent, so that the velocity and angle of the pyroclasts are not significantly modified from their initial values (~2% for a particle traveling at 40 m/s), and high enough to avoid the effects of ash and gas clouds and image saturation. The final database with the time evolution of the velocity, angle, size of the ejected bombs, and the mass ejection rate is available in the supporting information, while an example is given in Figure 2.

Figure 2.

Output of the automatic Pyroclast Tracking Velocimetry (PyTV) analysis of a high-speed video of a Strombolian explosion (see supporting information for similar information on all the studied explosions). For a given detection height above the vent, here set at 4 m, the analysis provides the evolution over time of, from top to bottom: (1) vertical velocity of individual bombs; (2) bomb ejection angle with respect to vertical; (3) bomb size (equivalent diameter), and (4) mass eruption rate (in gray) and cumulative mass of bombs ejected (black line). Note the closely spaced points that define the different ejection pulses (the range of assessed pulses is given in the header).

Bomb size is computed as equivalent diameter from the square root of the projected surface averaged on all the frames. In the raw data (Figure 2 and supporting information), we report all the detected pyroclasts. However, a large proportion of pyroclasts smaller than 10 cm (~5 px) is not efficiently tracked by the PyTV algorithm. Hence, we exclude them from the size distribution and mass analysis. Pyroclast mass is then estimated assuming a spherical shape and a density of 1.2 × 103 kg/m3 [Ripepe et al., 1993; Lautze and Houghton, 2007; Polacci et al., 2008]. The total mass is corrected to take into account the particles hidden by the ash cloud. Sensitivity analysis [Gaudin et al., 2014] shows that at least 50% of the bombs larger than 10 cm are effectively tracked by PyTV, reducing uncertainties on the total mass of the bombs ejected during an explosion to less than a factor of two.

Bomb velocity at the detection line is computed from the displacement in five successive frames and has a precision of 1 m/s. Note that the retrieved values are not the absolute velocities but only their projection on a plane perpendicular to the line of sight. As a consequence, the absolute velocity of a pyroclast with an angle of 30° toward the camera will be underestimated by 15%. The velocities are split into pseudo-vertical and pseudo-horizontal components, and apparent angles with respect to the vertical are computed.

For further details on the rationale and application of the post-processing method described above, as well as for an analysis of the uncertainties in the retrieved parameters, please refer to Gaudin et al. [2014].

4 Results

4.1 Vent-Averaged Ejection Parameters of Bomb-sized Pyroclasts

During the few hours of observations on each measurement day, none of the Stromboli or Yasur vents changed substantially in terms of style and intensity of the activity. Consequently, initial characterization of the activity at each vent during this specific time interval is achieved by compiling the distribution of velocity, ejection angle, and size of the ejected bombs from all explosions occurring there (Figure 3).

Figure 3.

Distribution of size, ejection angle, and velocity of bombs (>10 cm in size) for all the explosions from the six studied vents. See Table 1 for the main characteristics of explosions analyzed for each vent.

Except for vent SW2, ejection velocity distributions are asymmetric with 20–40 m/s modal values (Figure 3). A small proportion of bombs reach velocities 5 to 10 times higher than the modal values (up to 200 m/s in some explosions at Stromboli). The observed proportion of relatively “fast” bombs is higher at Yasur than at Stromboli. Vent SW2 of Stromboli stands out with relatively few slow bombs and a modal value of ~100 m/s. Note that these values are smaller than those found by Taddeucci et al. [2012a, 2012b], because, in our study, only the bombs coarser than 10 cm are efficiently tracked. The ejection angle is symmetrically distributed around a mean ejection angle, characteristic of each vent, with a spread (standard deviation of the ejection angle) varying from 20° to 90° (Figure 3). In all cases, the mean ejection angle of bombs, characterizing each single vent, deviates less than 30° from the vertical. Finally, at all vents the smallest bombs always represent the most abundant class in our detection range (Figure 3). The total mass of bombs in the 9.5–10.5 cm size range is 2 to 10 times greater than the total mass of the 47.5–52.5 cm size range, although the difference is less pronounced at Yasur than at Stromboli. In some cases (Str-SW2 and Str-NE1), a secondary peak is present for the coarsest bombs, which might be an artifact due to the very small number of the coarsest clasts.

4.2 Ejection Pulses

The time evolution of the different studied parameters (Figure 2 and supporting information) clearly shows that the ejection of bombs in Strombolian explosions is not steady over the duration of an explosion, but organized in pulses. Pulses are recognizable by a cluster of fast bombs arriving in a short time interval and displaying a non-linear decay of the vent ejection velocity over time (Figure 4). In the studied explosions, up to a hundred pulses are recognizable within a single explosion, each pulse comprising 21 to 1901 detected bombs. The duration of individual pulses ranges from 0.05 to 2 s, and their repetition rate can reach 10 pulses per second in SW2. The higher rates are often associated with strong overlapping pulses, making it impossible to determine to which pulse the slowest bombs are associated.

Figure 4.

Changes in ejection parameters over time during a single ejection pulse (explosion Str-NE_2). Top: pyroclasts vertical velocity. Fitting the velocity decay with equations (1) and (2) provides an estimate of h + hd. Center: ejection angle measured from vertical (positive toward SW, in this case). Bottom: pyroclast size. Note the decreasing ejection velocity and increasing size and spread angle of pyroclasts over time. Histograms give the number distribution of the pyroclasts for the corresponding parameter.

We visually assessed a range for the number of pulses within each explosion (Figure 2), but only pulses that could be unequivocally isolated were quantitatively analyzed. The proportion of analyzed versus assessed pulses is 90% at Yasur, but only 35% at Stromboli, where the pulse rate is generally higher (Table 1). The non-linear decay of the ejection velocity (vc) during each pulse has been fit against time (t) to obtain a characteristic depth associated with each pulse:

display math(1)

where hd the height of the detection line above the vent (fixed at 4 m, see section 4.1). The two unknowns, t0 and h, represent, respectively, the time of the burst, and a characteristic depth that will be discussed in section 6.1.1 (Figure 4). The mean velocity of the individual particles from the depth h + hd and the observation line vc is derived from the observed velocity v as:

display math(2)

where g the gravity (g = 9.81 m/s2). We did not correct h for the effects of drag since the latter depends on (1) the unknown velocity of the gas inside the conduit and in the atmosphere and on (2) the size of the bombs, resulting in a different fit for each size bin. Despite large uncertainties, (20% in well-defined pulses of Stromboli explosions, and up to 75% in the worst-defined pulses), the characteristic depth h is always shallower than 30 m. Apart from vent SW2, h is on average smaller at Stromboli (h ~ 0–2 m) than at Yasur (h ~ 5–11 m) (Figure 5).

Figure 5.

Correlation between the maximum velocity and the other characteristic parameters of the ejection pulses from (top) Stromboli and (bottom) Yasur volcanoes. Each point represents a single ejection pulse, color coded for the source vent. Green dashed lines are the linear correlations and related coefficients between different parameters, for both types of pulses of Stromboli (high velocities and low velocities), and for Yasur. Coefficients marked with a star denote non-significant correlations (signification level below 90%).

During a single pulse, the ejection rate increases at the beginning, peaks after about 25% of the total duration, when the velocity is about 25% of the maximum velocity, and then decreases (Figure 4). Finally, the ejection angle of the bombs also displays a characteristic evolution during an individual pulse: the spread angle, defined as the standard deviation of the ejection angle of all bombs at any time, increases over time by 20° on average, i.e., bomb trajectories become less collimated as the pulse progresses. At the same time, the mean ejection angle of bombs may also vary significantly during a pulse, the shift sometimes exceeding 40° in some Stromboli pulses (Figure 4).

Since they present common characteristics, ejection pulses can be described by a small set of parameters, designed to be independent and self-consistent despite the different observational conditions: (i) the maximum ejection velocity, measured at the beginning of the pulse; (ii) the characteristic depth h, computed after equations (1) and (2); (iii) the spread angle of the trajectories (measured for bombs in the 40–60 m/s velocity range); (iv) the shift in ejection angle; (v) the tenth percentile of the bomb size distribution; and (vi) the total ejected mass.

The relationships among the above parameters appear to be characteristic of the different vents (Figure 5). At Stromboli, two types of pulses are noticeable: slow pulses (maximum velocity below 40 m/s: SW1, NE2, and most of NE1 pulses), showing a strong positive correlation between the velocity and the total mass, the size of the pyroclasts and the spread of ejection angles; and fast pulses (maximum velocity 40–160 m/s: SW2 and the first pulses of NE1 explosions), showing different relationships, i.e., only h correlated positively with the velocity, while pyroclasts sizes and total mass display an anti-correlation with the velocity. At Yasur, there is no clear distinction between pulse types, and correlations are less-defined, and maximum velocity correlates positively with the spread angle and negatively with the size (Figure 5).

4.3 Explosion Parameters

Differences in the activity at the vents of the two volcanoes are marked not only at the scale of the ejection pulse but also at the scale of the whole explosion. At Stromboli, each Strombolian explosion is easily identified with a large number (5 to 100) of closely spaced pulses. In addition, the explosions of SW2 and NE1 show distinctive trends of pulse successions in which the average pulse velocity shifts gradually over time (see supporting information). By contrast, at Yasur distinguishing the onset and ending of individual explosions is harder, due to the presence of a background of small pulses continuously occurring at intervals of seconds to minutes at the vents. The main explosions, corresponding to the high energy events of Bani et al. [2013], are marked by orders-of-magnitude stronger (faster and with more pyroclasts) pulses and may begin with the arrival of one or multiple large pulses, but also with a gradual increase in the size and number of pulses. Also, we note that larger pulses occur either in isolation or closely grouped in time (see supporting information).

Here, we define three basic parameters to describe each Strombolian explosion in terms of ejection pulses: (i) the mean pulse rate, i.e., the number of assessed pulses divided by the explosion duration; (ii) the mean ejection velocity, defined as the mean velocity of all bombs in all pulses; and (iii) the mean mass per pulse, i.e., the total mass of bombs divided by the number of pulses in the explosion. The three parameters correlate positively at both volcanoes (Figure 6). In addition, for a given mean pulse velocity, the pulse rate is generally higher at Stromboli (up to 7 pulses per second on average) than at Yasur (never exceeds 3 pulses per second).

Figure 6.

Correlation between the pulse frequency, defined as the number of ejection pulses divided by the observed duration of the explosion, and the total mass and velocity of bombs (>10 cm in size) (each point represents a single explosion). Note that, at the scale of the explosion, the mean velocity and the total mass of bombs correlate positively with the pulse rate.

As a consequence, the total kinetic energy Ec of the bombs can serve as a proxy for the total energy release during an explosion. Ec is computed by summing up the kinetic energy of each bomb, as follows:

display math(3)

where mi and vi are the mass and the velocity of individual bombs, respectively. For the explosions recorded in our study, the total kinetic energy ranges over more than 4 orders of magnitude (Figure 7), but each vent is characterized by a specific range of kinetic energy. This range of variation is much greater than uncertainties (due to PyTV and video acquisition limitations, including videos saturated by the simultaneous ejection of too many pyroclasts and videos that did not capture the whole duration of one explosion) estimated to be less than a factor of two.

Figure 7.

Estimated kinetic energy associated with bomb ejection from the different explosion vents (B and B1 at Yasur, and SW1, SW2, NE1, and NE2 at Stromboli). At vents NE1, SW1, and SW2, the kinetic energy is a lower bound because the very large amount of ejecta led to image saturation, and/or the duration of recordings was too short.

5 Discussion

5.1 Ejection Pulses and Strombolian Explosions

Our study of the eruptive activity of two volcanoes renowned for Strombolian explosions (including the “type locality”) provides strong evidence that such explosions are not the result of a single, simple burst event. In contrast, we demonstrate that Strombolian explosions occur by successive pressure release pulses that last from 0.05 to 2 s, as measured 4 m above the vent. The frequency can reach 10 pulses per second, which is 10 times higher than reported by previous manual analysis of high-speed videos [Taddeucci et al., 2012b]. The ubiquitous occurrence of ejection pulses in all the analyzed Strombolian explosions from the two volcanoes suggests that they can be regarded as the fundamental unit of the Strombolian style of volcanic activity.

The ejection velocity decay that defines the pulses was previously modeled according to shock tube theory [Taddeucci et al., 2012a]. The same fit cannot be used here, since it requires knowledge of the maximum velocity of small pyroclasts, which are not efficiently tracked by the PyTV codes. However, equation (1) resembles that of Alatorre-Ibargüengoitia et al. [2010], and most of the retrieved values of h are close to the estimation of Taddeucci et al. [2012a] (see below). Equation (1) is suitable for bombs that are ejected at or above the depth h below the vent rim and thus undergo a short acceleration phase (Figure 8).

Figure 8.

Theoretical evolution of the ejection of pyroclasts during a single pulse. The pyroclast velocity, measured by the camera at height hd above the vent, is used to trace the bombs trajectory both inside and outside the conduit, assuming a constant velocity (zero drag and gravity). At time t0 all trajectories converge toward a characteristic depth (h), which then represents the deepest possible source for bombs (dotted gray lines). The observed velocity decay matches a greater acceleration of shallower pyroclasts with respect to deeper ones. (A) and (B) denote the acceleration and the ballistic transport phases, respectively.

The theoretical duration of the pulses can be estimated by computing the time needed for the slowest bombs to reach the detection line under gravity deceleration (and neglecting drag):

display math(4)

For h + hd ranging from 4 to 24 m, pulses should last between 0.9 and 2.2 s. The upper boundary (2.2 s) is in good agreement with the observations. On the other hand, pulses shorter than the predicted lower boundary (0.9 s) are observed when pulse frequencies are very high (e.g., SW2). Indeed, if pulse frequency is higher than pulse duration, the slow-tail bombs from a given pulse will be caught up by the following pulse while still inside the conduit. In this case, the slowest bombs that can escape the conduit will have a velocity:

display math(5)

where Δt is the delay between two successive pulses. Vent SW2 (h = 9 m; Δt = 0.1–0.2 s) has a velocity distribution of bombs markedly different from the others (Figure 3), and the computed minimum velocity ranges from 45 to 90 m/s, in good agreement with the observations (see supporting information). For the other Stromboli and Yasur vents, the minimum velocity computed by equation (5) is not sufficient to reach the detection height. Outside the conduit, some of the bombs caught by the following pulse may maintain their own velocity, causing the pulses to overlap.

The observed values of h, ranging from 0 to 30 m with a mode around 7 m, correspond to the lower range in Taddeucci et al. [2012a] who found depths up to 190 m, which corresponds well with the estimated size of the slugs (50–250 m at Yasur [Kremers et al., 2013] and 20–300 m at Stromboli [Chouet et al., 1999; Harris and Riepepe, 2007]). The use of PyTV, with respect to manual tracking, allowed detection of smaller, individual pulses.

The ubiquity of the pulses in Strombolian explosions and their similar characteristics suggest that all pulses are generated by a common mechanism. In most cases, the intensity of the pulses remains sustained with time, implying that pulses must be produced by individual gas pockets (Figure 2 and supporting information) whose volume seems to be limited by a threshold depending on the erupting vent. Each of these gas pockets appears to be released at very shallow depth (1–8 m in average, and never more than 30 m).

During the release of gas from a large bubble or slug, pulses could be generated by the repeated collapse of (1) portions of conduit walls, (2) pyroclasts falling back in the crater, and/or (3) the liquid film lining conduit walls [Taddeucci et al., 2012b]. The viscoelastic response of the liquid film or the conduit walls to a pulse release may also play a significant role, especially for Stromboli explosions where the high pulse frequency, up to 10 per second, is hard to explain only by gravitational collapse. Pulses could also be generated inside the conduit after fragmentation, by pressure fluctuations in the gas-particles mixture [Dartevelle and Valentine, 2007]. Alternatively, the gas pockets associated with pulsed could be generated at depth and reach the surface as a train of bubbles.

5.2 Pressure and Energy Release and the Effect of Magma Properties and Vent Geometry

Within each pulse, the velocity decay is invariably accompanied by a peak in the ejection rate (Figure 4) and an increase in the size and spread angle of bombs ejected over time (Figure 2). The combination of these trends can be explained by the decreasing pressure differential and the dynamics of the pressure release event driving each pulse. At the beginning of the pulse-forming event, the maximum pressure differential drives the maximum acceleration and fragmentation of the magma. The pressure gradient may be enhanced by initial focusing of the magma rupture in a small area, resulting in low dispersion of the bombs around the mean ejection direction. Then, the increasing size of the rupture area increases the bomb mass ejection rate while the pressure differential decreases, reducing the ejection velocity. Meanwhile, the average ejection angle may vary due to the change in the location in the rupture point of the bubble as occasionally observed in the videos of shallow explosions. At the end of the pulse, the pressure differential decrease hinders efficient magma fragmentation and bomb acceleration. In addition, the ejected bombs become less well-sorted in size, which induces higher differential velocities, more frequent collisions, and thus more divergent trajectories, i.e., increasing spread angle (Figure 2).

The total kinetic energy released during single explosions range over at least 4 orders of magnitude (0.02 to >110 MJ; Figure 7). Larger total energies produce higher mean velocities and greater ejected masses (Figure 5). Possibly, higher gas pressure not only accelerates bombs faster but removes more magma from the conduit due to drag and enhanced fragmentation. These two parameters are also positively correlated with the release frequency of ejection pulses (Figure 6), suggesting that this pulse frequency is also modulated, to first order, by the overall pressure driving the explosion.

The total kinetic energy can be compared to the total mechanical energy available in the pressurized gas pocket (neglecting the thermal component, which may still be dominant), which can be computed thermodynamically as:

display math(6)

where pb and Vb are the pressure and volume of the gas pocket, and patm the atmospheric pressure. According to previous studies, for slugs ranging from 20 to 300 m at Stromboli [Chouet et al., 1999; Harris and Riepepe, 2007], that is to say volumes from 100 to 1500 m3 and pressure ranging from 0.2 to 0.7 MPa [Del Bello et al., 2012; Taddeucci et al., 2012a], the total available energy in a single explosion would range between 5.9 and 725 MJ, suggesting that only about 0.1–10% of the available energy is used to propel bomb-sized pyroclasts, while the rest is used to accelerate smaller pyroclasts and gas and/or is dissipated in the form of elastic response of the hosting rock, acoustic waves, etc.

In addition to the energy released during an explosion, the vent and volcano conditions have a strong control on the averaged bomb parameters (Figure 3) and their relationship (Figure 6). For example, two different regimes are distinguishable at Stromboli. The ejection pulses of SW1 vent and the first pulses of NE2 vent differ from those at other vents by higher ejection velocities, and a different relationship between velocity, depth, mass, and spread (Figure 5). These peculiarities point toward an energy threshold delimiting two types of pulse regime, each of them primarily determined by the maximum velocity of the ejected particles.

Finally, specific differences can be observed between the two study volcanoes. For a given pulse rate, Yasur explosions have higher velocities than Stromboli (Figure 6). In addition, the pyroclasts are ejected from a deeper location (8 m vs 1 m in average), and the bombs tend to be larger (Figure 5). This may reflect higher melt viscosity at Yasur, i.e., 104.1 Pa s vs. 103.1 Pa s at Stromboli (see Table 2), which would favor higher pressure buildup, and/or wider vents (~10 m of diameter at Yasur vs. 2.5 m at Stromboli).

Table 2. Chemical Composition, Liquidus Temperature, and Correspondent Viscosity of Glasses in the Studied Volcanoes
  1. a

    Data from Métrich et al. [2010].

  2. b

    Electron microprobe analysis (analytical procedures as per Gozzi et al. [2014]) of Yasur pyroclasts from the study period (mean of 35 analyses).

  3. c

    Standard deviation.

  4. d

    Sum of H2O and CO2.

  5. e

    Calculated at liquidus temperature with the MELTS algorithm [Ghiorso and Sack, 1995].

T liquidus (°C)11451110 
η (Pa s)e103.1104.1 

6 Conclusions

This study shows that, during a Strombolian explosion, pyroclasts are ejected in a substantially similar pattern that is independent of a specific vent or volcano, thus suggesting a common physical mechanism. The release of pressurized gas is invariably achieved gradually through pyroclast ejection pulses, each lasting a few tenths of seconds. Pyroclasts are propelled from shallow (0–30 m) depths and then driven up the upper-conduit/vent up into the atmosphere. Defining characteristics of the ejection pulses include a non-linear decrease of ejecta velocity with time, a peak in the ejection rate with coarser bombs ejected at the end of the pulse, and an increase in the ejection spread angle with time. These observations fit well the process of bubble burst, where the rupture area increases in size and shifts while the pressure differential decreases.

The characteristic parameters of the pulses and, particularly, the combination of pulses in a single Strombolian explosion are primarily controlled by the total energy available. The available mechanical energy of the rising gas, a function of its volume and pressure, determines not only the mean velocity and the ejected mass of pyroclasts, but also the frequency at which ejection pulses are released during an explosion. The kinetic energy released through bomb ejection ranges over more than 4 orders of magnitude in our observation windows and varies from vent to vent, suggesting that the gas pocket characteristics are primarily determined by the physical properties of the feeder magma and conduit geometries.

Generally speaking, vent conditions influence all the relevant explosion characteristics (duration, mean velocity, and total mass of ejecta). The organization of the explosions in discrete pulses supports the hypothesis that the maximum energy that can be released at one time through a conduit is controlled by the conduit geometry and physical properties of magma. Our results indicate that several ejection regimes may occur, depending on the total available energy, even during a single explosion.

From a hazard management perspective, we emphasize that the various parameters of Strombolian explosions cannot be studied independently. For instance, our results suggest that stronger explosions are expected to eject a dangerous combination of more and faster pyroclasts over a wider angle, as evinced by the relationship we found between ejection velocity, mass, and spread angle.


Data supporting this paper is available as supporting information (fs01.pdf), and at INGV Roma—Department of Seismology and Tectonophysics, HP-HT lab. Data sets: Yasur_0711, Stromboli_0609, and Stromboli_1009. We thank C. Acerra, M. Albano, M. Mari, S. Rao, and E. Garaebiti (Vanuatu Geohazards Observatory) for invaluable field and technical support; funding was provided by the INGV-DPC “V2” and “Paroxysm”, FIRB-MIUR “Research and Development of New Technologies for Protection and Defense of Territory from Natural Risks”, and FP7-PEOPLE-IEF-2008-235328 “NEMOH” ITN projects. Larry Mastin and an anonymous scientist provided useful reviews that significantly improved the manuscript.