We use regional seismo-acoustic data from AVO networks (Fig. 1). Infrasound array OKIF was a 4-element array located \(\sim \)60 km south of Bogoslof on Umnak Island (Fig. 1). The array consisted of Chaparral Physics Model 25Vx sensors sampled at 100 Hz by a Quanterra Q330 digitizer. The array aperture was \(\sim \)100 m and each element had wind noise reduction. The sensor-digitizer flat passband is \(\sim \)0.1–50 Hz. Array data are processed to determine coherent acoustic signals using a least squares algorithm (e.g., Szuberla and Olson (2004)). Infrasound data are considered to have originated from the volcano if the data segment had an acoustic trace velocity (0.25–0.45 km/s), back-azimuth within ± 15∘ of the actual, and a median cross-correlation maximum (MdCCM) above 0.5. MdCCM is determined by finding the median of the maximum cross-correlation between unique array element pairs (Lee et al. 2013) in 30 s of segments with 50% overlap. Acoustic travel time is removed assuming a constant propagation velocity of 0.34 km/s. See Haney et al. (2018) and Schwaiger et al. (2019) for more details on acoustic propagation between Bogoslof and the OKIF array. Data are filtered between 0.1–10 Hz for array processing. This relatively broad frequency band is chosen to capture the wide range of frequencies produced by Bogoslof and to be consistent with the frequency index calculations (see below). We determine infrasound event durations by summing the coherent data segments originating from Bogoslof. Delay-and-sum beamforming for a source originating from Bogoslof is applied to increase the signal-noise ratio. Other infrasound arrays in the region recorded Bogoslof eruptions, but are generally at much greater distances with increased path effects. OKIF provides the most complete acoustic dataset and recorded 41 out of the total 70 explosive events identified by AVO. AVO’s event identification involved detailed multiparameter analysis of the eruption sequence (Coombs et al. 2019)
We analyze seismic data from station MAPS located \(\sim \)73 km east of Bogoslof on Makushin volcano, Unalaska Island (Fig. 1). This station consists of a 3-component Guralp CMG-6TD seismometer with a built-in digitizer. Although other seismic stations are closer to Bogoslof, MAPS is broadband, digital, and ran nearly continuously during the eruption sequence. The sensor response is flat between 30 s and 100 Hz, and the response has been removed. We only analyze the vertical channel. Seismic travel time is accounted for assuming a 5-km/s velocity. All times listed are in UTC and plume heights are above sea level.
The frequency index (FI) (Buurman and West 2006) is used to examine changes in infrasound frequency content. FI is calculated using the following:
$$ \text{FI}={\log}_{10} \frac{A_{\text{upper}}} {A_{\text{lower}}} $$
(1)
where Aupper are the mean spectral estimates from an upper frequency band and Alower are the mean spectral estimates from a lower frequency band. Here, we define Alower between 0.1 and 1.0 Hz and Aupper between 1 and 10 Hz. Similar to infrasound array processing, spectra are computed in 30 s segments with 50% overlap. FI is calculated only for the infrasound data when the aforementioned array processing thresholds are met. Elevated low-frequency noise and the microbarom could influence the array detections and FI calculation, although this should be relatively minor due to our strict array processing thresholds.
The acoustic to seismic ratio is computed to examine changes in energy partitioning. Johnson and Aster (2005) defined the useful metric volcano acoustic–seismic ratio (VASR) as the ratio between the acoustic and seismic energy emitted from a volcano. The seismic energy calculation includes terms on density, attenuation, and site response that are likely complicated and poorly constrained for this eruption. Similarly, the acoustic energy calculation also relies on propagation and isotropic source radiation assumptions, which are also poorly constrained here. Because of this, we choose to simply analyze the ratio of the seismic and acoustic amplitude envelopes and refer to this modified metric as VASR*. We filter the acoustic waveforms between 0.1 and 10 Hz and the seismic data from 1.0 to 10 Hz with acausal butterworth filters. The higher seismic corner frequency is chosen to minimize contamination from the microseism. These filter bands capture the majority of the broadband seismo-acoustic energy from Bogoslof. The analytic signal of the Hilbert transform is then used to obtain the amplitude envelopes. Since we are interested in only gross source changes over long durations, envelopes are then averaged over 15-s intervals and smoothed using a 3-min duration convolutional filter. Similar to the infrasound FI, we examine VASR* only during data segments where infrasound array thresholds are met.
We acknowledge that seasonal propagation variations may affect the infrasound signals (Iezzi et al. 2018; Schwaiger et al. 2019). Due to the long duration of the 2016–2017 Bogoslof sequence, acoustic propagation likely varied between different periods of the eruption and certain events may have had more favorable propagation than others. Schwaiger et al. (2019) examine this in more detail. Because of these variations, we suggest VASR* is most useful over time periods of a single eruptive event or several days, but is difficult to compare over the entire sequence. Therefore, we only examine VASR* over the course of individual events. Infrasound frequency content should be less susceptible to changing propagation conditions than the amplitude.
Satellite observations used here are taken from available cloud-free images before and after selected eruptions. We primarily use Worldview-2 and -3 imagery, as well as Landsat 8, COSMO-SkyMed, and Sentinel-2. The DEM was created using the NASA Ames Stereo Pipeline (ASP) software package (Shean et al. 2016) with the January 31, 2017 Worldview-2 imagery. In this manuscript, we restrict our satellite-based analysis to qualitative observations, and focus on the changes in island morphology and how they relate to the seismo-acoustic data. More detailed satellite-based observations and interpretation are available in separate manuscripts, such as Waythomas et al. (2019) and Schneider et al. (2019). Details on satellite-derived SO2 and lightning observations are available in Lopez et al. (2019) and Van Eaton et al. (2019), respectively.
The majority of explosive events that occurred during the 2016–2017 eruption consisted of short-duration, explosion-like seismic, and acoustic signals with low FI. Figure 2 displays the FI for the 35 events detected at the OKIF infrasound array during the 2016–2017 Bogoslof eruption sequence. FI values for most events are below − 2, and the mean of all FI observations is − 2.3. Seven events have FI values that exceed − 1: December 12, 2016, January 31, 2017, March 8, 2017, June 7, 2017, June 10, 2017, July 10, 2017, and August 7, 2017 (Table 1). Satellite analysis suggests the vent was subaerial or “dry” during some point of all of these high FI eruptions, except for July 10, 2017 (Fig. 2). Below, we analyze most of these high FI eruptions and compare them to a more typical Bogoslof eruption from February 20, 2017.
Table 1 Summary of observations for events with FI greater than − 1, as well as the Feb. 20, 2017 eruption more typical of events during the eruption sequence
Infrasound event durations vary greatly and generally correlate with peak FI values (Fig. 2). Event durations range from 15 s for a couple of events up to 9690 s (161.5 min) on March 8, 2017. Eight events have durations longer than 1800 s (30 min): December 12, 2016, January 9, 2017, January 31, 2017, March 8, 2017, May 17, 2017, June 10, 2017, June 13, 2017, and August 7, 2017 (Table 1). These long duration events also have generally high FI values. Comparison of the peak FI and log10 of the duration for each event gives a linear correlation coefficient of 0.62.
Comparison of the range of FI values for each event with the event duration shows a strong correlation. Figure 3 plots the range of FI values (maximum FI–minimum FI) vs. the event duration for each eruption. The durations are plotted on a logarithmic scale. The correlation coefficient between the FI range and log10 of the duration is 0.91. This shows that longer events have a wider range of frequencies.
Satellite observations of Bogoslof during the 2016–2017 eruptive period generally showed a submerged vent, located within a small bay open to the sea or enclosed by a tephra ring. However, on several occasions, imagery also showed a vent on land (December 12), a lava dome that breached the water surface (early June), or water draining back into a dry crater following an eruption (January 13 and March 8). The red arrows in Fig. 2 represent independent satellite-based evidence of a dry or subaerial eruptive vent and are discussed in detail later. In general, these time periods correspond with events with high FI (>− 1) and long durations (> 1000 s/17 min). Below we analyze the seismo-acoustic character and FI and VASR* values for selected events and then discuss inferences on vent characteristics. Available satellite images before and after the eruption are used to infer the state of the vent. Table 1 summarizes the observations from the eruptive events this manuscript focuses on.
Figure 4a shows the seismo-acoustic waveforms, spectrograms, FI, and VASR* for the December 12, 2016 eruption (event 2). This event has relatively high FI values, peaking at − 0.5, but the character of the eruption is different from the other high FI events. This eruption consists of numerous short-duration explosions spaced closely together, which is typical of many of the Bogoslof eruptions with lower FI (e.g., February 20, 2017). However, the December 12 explosions have much higher frequency content. Satellite images of Bogoslof on December 14 suggest a small vent near the center of the island (Fig. 4b). The vent may thus have been above ground or submerged in only very shallow water. A December 1 image from before this event showed no vent or open water in the center of the island. Infrasound amplitudes peak at \(\sim \)1 Pa. VASR* is mostly low for this event but has a couple bursts just above 0.0001 Pa/μ m/s. Seismic background noise is relatively high. No significant plume, SO2, or lightning were detected from this eruption (Table 1). There was a weaker eruption earlier on December 12 apparent only in the infrasound data. It is notable that these two events represent the first infrasound detected from Bogoslof and likely the opening stages of the eruption sequence.
The Bogoslof eruption on January 31, 2017 (event 29) has marked changes in FI and VASR*. Figure 5 shows the seismo-acoustic waveforms, spectrograms, FI, and VASR* for 9 h surrounding this eruption, plotted in a similar manner to Fig. 4a. The infrasound begins around 05:21 with a number of relatively low-amplitude, short-duration signals (explosions). The frequency content of these events is low, with the FI during this period mostly below − 3. Around 08:30, the infrasound character changes to sustained, more broadband signals with higher amplitudes. FI during this period rises to approximately − 1. After a brief pause, eruption signals begin again around 09:45 and last for \(\sim \)60 min, with infrasound amplitudes peaking around 10:20 and FI increasing to around − 0.5. After a pause in the eruption, two infrasonic pulses occur over the next hour. The seismicity for this eruption primarily consists of transient earthquakes followed by intermittent tremor between \(\sim \)07:30 and 13:30; otherwise, no significant amplitude or frequency changes are apparent and the microseism and other ocean-related noise dominates frequencies below 1 Hz. The VASR* generally follows the infrasound amplitudes as the seismic amplitudes stay roughly constant. High VASR* values typically correspond to time periods with high FI.
Satellite images of the island prior to January 31 suggest the vent region was underwater and located in a small (\(\sim \)250-m wide) bay that likely represented the eruption crater (Fig. 5b). A WorldView-2 satellite image from after the eruption shows the morphology of the island markedly changed during the January 31 eruption (Fig. 5b), as the crater was no longer open to the sea and was below sea level. Water was clearly visible draining back into the vent. A DEM created from this imagery (Fig. 5b) suggests the vent was at least 20 m below sea level and that the eruption expelled all of the existing water and built a tephra collar that temporarily restricted seawater interaction at the vent. The topographic barrier separating the crater region from the sea is composed of pyroclastic deposits (Waythomas et al. 2019). The January 31 eruption also represented a change in observed volcanic emissions. Satellite data indicates the first eruption cloud was visible by 05:30 UTC and reached a height of 5.9 km. This eruption had a clear ash signal in the satellite data and produced trace ashfall on Unalaska Island and Dutch Harbor \(\sim \)100 km east of the volcano (Schneider et al. 2019). Ash was only detected previously from event 23 on January 18. Above-average SO2 was also detected (3.6 kt compared to an average of 2 kt) (Lopez et al. 2019).
Figure 6a displays the seismo-acoustic waveforms, spectrograms, FI, and VASR* for 2.5 h surrounding a typical eruption during the 2016–2017 eruption sequence, February 20, 2017 (event 36). No infrasound is apparent until a rapid onset around 02:11. Numerous short-duration, high-amplitude, impulsive bursts occur until the end of the eruption at \(\sim \)02:50. Infrasound frequencies are similar throughout the eruption, and are focused below 1 Hz. FI values range between − 2 and − 4. The explosions have high infrasound amplitudes, up to 19 Pa. Numerous earthquakes occur prior to the eruption onset at 02:11. The seismic amplitude then decreases and the earthquakes transition into seismic eruption tremor coincident with the infrasound detection. VASR* is high for the first eruption pulse, peaking at around 0.0012 Pa/μ m/s, then decreases over time. VASR* then varies irregularly during the eruption from 0.00068–0.0078 Pa/μ m/s, and is higher than the January 31 eruption, mostly because of the very high acoustic amplitudes. The eruption also produced an ash-poor plume to 6.1 km, 0.7 kt of SO2, and 21 lightning strokes (Lopez et al. 2019; Schneider et al. 2019; Van Eaton et al. 2019). WorldView-2 satellite imagery from this period (Fig. 6b) confirms the vent was submerged in a small lagoon before and after the eruption (Waythomas et al. 2019).
The March 8, 2017 eruption (event 37) contains periods of high infrasound FI and VASR*, similar to January 31; yet, these parameters have a more complex evolution. Figure 7a shows the infrasound and seismic signals for this event begin rapidly around 07:40 with a relatively high FI around − 1. Infrasound remains elevated until \(\sim \)10:20 UTC, with FI decreasing to − 2 to − 4 during the middle of eruption and then increasing again to near 0 around 09:45 UTC and then falling once again. Seismicity is quiet prior to the eruption, then increases in amplitude gradually during the middle of the eruption when the infrasound FI is low. VASR* is slightly elevated at the beginning of the eruption, then very low during the middle, then increases substantially toward the end of the eruption. Higher frequency harmonics are apparent in the infrasound spectrogram during periods of high FI.
Satellite images before and after the March 8 eruption similarly suggest a complex eruption progression (Fig. 7b). A Worldview image from March 3 shows a broad, roughly circular lake in the center of the edifice. The next clear image on March 11 shows that during the March 8 eruption the crater lake grew substantially, tephra was deposited along the western shore, and a small new vent opened on the northwest shore of the island. This event also had a relatively substantial, ash-rich plume between 10.6- and 10.8 km height, with trace to minor ashfall occurring on Unalaska (including Dutch Harbor) and a mariner just off-shore the northwest shore of Unalaska island. Extensive lightning and the highest SO2 (21.5 kt) of the eruption were also detected (Table 1). This was likely the most intense eruption of the entire eruption sequence.
Eruptions between June 6 and10 (events 44–48) are unique in the 2016–2017 eruption sequence in that an above-water lava dome was visible in satellite imagery beginning late on 5 June (Fig. 8c) (Coombs et al. 2019; Waythomas et al. 2019); thus, eruptions during this period are presumed to be lava dome related and have significant subaerial components. The dome appears to have been destroyed by the June 10 eruption (Fig. 8c). Figure 8a and b shows the seismo-acoustic data for the June 7 and June 10 eruptions. Similar to other events during these few days, these eruptions have high FI (up to and slightly above 0 with a low around − 2) (Fig. 2). The June 10 eruption differs somewhat from the other June 6–10 eruptions. The last hour has a gradual decrease in FI to below − 3, potentially corresponding to a change from subaerial to submarine after destruction of the dome earlier in the eruption. Durations of the events on June 6–9 are short (< 200 s) in contrast to June 10 which is much longer (5040 s). VASR* values on June 10 show a gradual increase and generally follow the acoustic amplitudes, peaking when the FI peaks. VASR* values peak between 0.0006 and 0.0007 Pa/μ m/s. Plume height was low for the June 6–9 events (not detected or < 3.2 km) and quite a bit higher (9.5 km) on June 10. Similarly, no lightning or SO2 were detected from the June 6–9 events but numerous strokes and SO2 were detected on June 10 (Table 1).
The August 7, 2017 eruption (event 63) has a progression from low to high FI and VASR*, with peak FI values around − 0.5 (Fig. 9a). Maximum infrasound amplitudes are \(\sim \)1.5 Pa and VASR* of \(\sim \)0.0002 Pa/m/s and the duration is relatively long at 1860 s. Prior to this eruption, the volcano experienced a lull in activity (last eruption July 11) and the bay in the center of the island was open to the ocean (Fig. 9b). Similar to the January 31 eruption, a Worldview-2 image after the August 7 eruption (Fig. 9b) shows the now enclosed crater lake has been enlarged and water is draining back into the below sea level lake. Plume height is estimated 10.8 km, with 4 lightning strokes and a relatively high 5.8 kt of SO2. Clear harmonics are also apparent in the infrasound for the June 7, June 10, and August 7 eruptions.