Journal of Volcanology and Geothermal Research

Transcription

1 VOLGEO-4787; No of Pages 15 Journal of Volcanology and Geothermal Research xxx (211) xxx xxx Contents lists available at SciVerse ScienceDirect Journal of Volcanology and Geothermal Research journal homepage: Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption David Fee a,, Stephen R. McNutt b, Taryn M. Lopez b, Kenneth M. Arnoult a, Curt A.L. Szuberla a, John V. Olson a a Wilson Infrasound Observatories, Geophysical Institute, University of Alaska Fairbanks, 93 Koyukuk Drive, Fairbanks, AK , United States b Alaska Volcano Observatory, Geophysical Institute, University of Alaska Fairbanks, 93 Koyukuk Drive, Fairbanks, AK , United States article info abstract Article history: Received 19 May 211 Accepted 14 September 211 Available online xxxx The explosive phase of the 29 Redoubt Volcano eruption produced predominantly short duration, highamplitude infrasound signals recorded up to 45 km away. All 19 numbered explosive events were recorded at a local microphone (DFR, 12 km), as well as at an infrasound array in Fairbanks, Alaska (I53US, 547 km), most with high signal to noise ratios. The local microphone provides an estimate of the source parameters, and comparison between the two datasets allows the unique opportunity to evaluate acoustic source term estimation at a remote array. High waveform similarity between DFR and I53US occurs during much of the explosive phase due to strong stratospheric ducting, permitting accurate source constraints inferred from I53US data. Cross-correlation analysis after applying a Hilbert transform to the I53US data shows how the acoustic energy has passed through a single caustic, as predicted by ray theory. Similar to previous studies, significant low-frequency infrasound from Redoubt recorded at I53US is coincident with high-altitude ash emissions. The largest events also produced considerable energy at greater than 5 s periods, likely related to the initial oscillations of the volcanic plume or jet. Many of the explosive events have emergent onsets, somewhat unusual for explosive, short-duration eruptions. Comparison of the satellite-derived SO 2 emissions with the relative amount of acoustic energy at I53US shows a very high, statistically significant correlation. This study reiterates the utility of using remote infrasound arrays for detection of hazardous emissions and characterization of large volcanic eruptions, and demonstrates how, under typical meteorological conditions, remote infrasound arrays can provide an accurate representation of the acoustic source. 211 Elsevier B.V. All rights reserved. 1. Introduction Low latency detection and characterization of hazardous volcanic eruptions remains a challenging task in remote areas, particularly when seismic data coverage is sparse or weather clouds and/or sampling frequency diminish the effectiveness of remote sensing. The Comprehensive Nuclear-Test-Ban Treaty Organization has developed a global network of infrasound arrays as part of the International Monitoring System (IMS) with the goal of detecting clandestine nuclear tests. Recent work has shown that this network can also be utilized to detect and constrain large volcanic eruption parameter estimates. Eruption characterization using the global infrasound network is particularly promising for remote volcanic eruptions, such as the recent Kasatochi and Sarychev Peak eruptions, as the existing monitoring networks in those regions were sparse and cloud cover (Prata, 29) and icing of ash particles (Rose et al., 1995) often hampers remote sensing's ability to detect volcanic ash and gas. Fee et al. (21b) combined IMS infrasound data with remote sensing analysis Corresponding author. Tel.: address: dfee@gi.alaska.edu (D. Fee). to determine eruption source parameters of the 28 plinian eruptions of Kasatochi and Okmok volcanoes. They observed at least three main eruption pulses and related their remote infrasound data (N2 km) to eruption source processes at Kasatochi. Arnoult et al. (21) also analyzed IMS data for these eruptions and identified at least seven arrays that detected the Kasatochi eruption. Matoza et al. (211a) provided the most detailed eruption chronology available for the 29 eruption of Sarychev Peak, Kurile Islands using IMS data, as this remote volcano had no seismometers nearby and the satellite sampling frequency was 15 min. Estimates of the source location were also made for the Sarychev Peak and Kasatochi eruptions (Arnoult et al., 21; Matoza et al., 211a). The recent eruption of Eyjafjallajökull, Iceland was detected at 14 infrasound arrays and demonstrated that even moderate sized eruptions can produce significant infrasound at remote distances (Matoza et al., 211b). These eruptions all presented a significant hazard to aviation and caused extensive flight delays and loss of airline revenue. These studies clearly show the IMS infrasound network can assist with the detection and understanding of large eruptions and help mitigate volcanic hazards. The goal of remote acoustic monitoring of volcanic eruptions is to provide constraints on various eruption parameters, such as location, /$ see front matter 211 Elsevier B.V. All rights reserved. doi:1.116/j.jvolgeores Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

2 2 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx timing, duration, and changes in intensity. These constraints can be used to detect and notify civil and aviation authorities of an ongoing eruption (Fee et al., 21a), as well as provide input for volcanic ash transport and dispersion models (e.g. Mastin et al., 29). The 29 eruption of Redoubt Volcano, Alaska provides a unique opportunity to further assess the utility of using infrasound to characterize volcanic eruptions. This eruption had numerous explosive events that produced prodigious infrasound detected by both a local pressure sensor (DFR, 12 km) and remote IMS arrays km away. Unlike previous remote volcano infrasound studies, the availability of local infrasound data provides a representation of the acoustic source waveform. In this manuscript we use the IMS array I53US, located in Fairbanks, Alaska (547 km), to derive source constraints on the explosive events themselves, and compare the local and remote acoustic data to determine the accuracy of these constraints. A thorough understanding of acoustic propagation from source to receiver is necessary for remote volcano infrasound studies, as the atmospheric structure will affect how the sound propagates to the array, thus influencing the acoustic travel times, propagation path, transmission loss, etc. For example, Fee et al. (21b) showed that uncertainties in the propagation path from Kasatochi volcano to the I53US infrasound array could lead to source timing discrepancies of up to 15 min. For this study we perform basic propagation modeling and analysis of the atmospheric models to help understand the remote data and comparisons to DFR. Further, we compare the satellite-derived SO 2 emissions with the relative acoustic energy at I53US and discuss the implications for using infrasound to detect hazardous volcanic emissions. This paper will serve as a companion paper to McNutt et al. (this issue), which discusses in greater detail the local infrasound data and its relation to other datasets. 2. Eruption overview Redoubt Volcano ( N, W) is a 3.1 km high active stratovolcano located in the upper Cook Inlet of southern Alaska (Fig. 1). The volcano's relatively high level of activity and proximity to Anchorage, AK (17 km), the Drift River Oil Terminal, and numerous North Pacific air routes make it a hazard to both air and ground civil activity. The most recent eruptive activity at Redoubt occurred in and is split into three phases: precursory, explosive, and effusive (Bull et al., this issue-b). The precursory phase began with increases in gas emissions noted in July 28, soon followed by elevated thermal activity at the summit. Seismicity and gas emissions increased substantially in January 29, followed by the first explosive eruption of ash and steam on 15 March 29 that signaled the onset of the explosive phase. Nineteen magmatic explosive eruptions were detected and classified by the Alaska Volcano Observatory (AVO) (Table 1) between 23 March and 4 April, with many of these events producing hazardous ash plumes to stratospheric altitudes that disrupted local air traffic and deposited ash on local communities. Melting of the Drift River Glacier during the eruptions produced voluminous lahars that inundated the Drift River Valley (Bull et al., this issue-b; Schaefer et al., 211). Petrological studies indicate crystal-rich andesites were erupted with a range of compositions between 57.5 and 62.5 wt.% SiO 2 (Coombs et al., this issue). Seismicity prior to the explosive phase consisted of swarms of repeating events and each of the explosive events coincided with strong co-eruptive tremor (Buurman et al., this issue). Significant SO 2 emissions were also detected during the explosive phase (Lopez et al., this issue). Four periods of lava dome growth were detected during the eruption, with three being destroyed during the explosive phase while the final dome grew after Event 19 and persisted during the effusive phase (Bull et al., this issue-a). Additional reanalysis events were determined by AVO after the eruption had ceased (McNutt et al., this issue). For this study we focus on the 19 primary events as the reanalysis events were smaller and not recorded as well at the remote infrasound arrays. 3. Data and methods 3.1. Infrasound data This study utilizes infrasound array data from the IMS network and a single local microphone deployed by AVO (station DFR). Fig. 1a shows the locations of the arrays relative to Redoubt, while Table 2 lists the stations and arrays within 5 km that detected at least one eruptive event, as well as their azimuth and range from the volcano. The DFR infrasound station consists of a single Chaparral a) b) N I18DK 6.6 DFR 75 N 6 I44RU N 45 N I53US Redoubt I1CA I56US 6.5 Redoubt W I57US 12 W km 15 N I59US Fig. 1. Map of study area. a) IMS stations (black circles) within 46 km of Redoubt Volcano, Alaska (black triangle). The closest array is I53US at 547 km NE of Redoubt. b) Redoubt Volcano and local infrasound station DFR, 12.2 km NE of the summit. Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

3 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx 3 Table 1 List of AVO defined events and detection summary at I53US. Onset indicates either emergent (E) or impulsive (I). Event times and DFR parameters are used from McNutt et al. (this issue), while plume heights are from Schneider and Hoblitt (this issue). Event 29 date Time DFR pressure DFR duration I53US I53US pressure I53US energy I53US duration Correlation Plume height Onset ULP (Pa) (min) Onset (Pa) (Pa 2 s) (min) coefficient (km) energy 1 23 Mar 6:35: :9: NaN 5.5 E N 2 23 Mar 7:1: :3: I Y? 3 23 Mar 8:14: :44: E? 4 23 Mar 9:38: :8: E? 5 23 Mar 12:3: :57: I Y 6 24 Mar 3:4: :13: E? 7 26 Mar 16:34: :1: N/A N 8 26 Mar 17:24: :53: E Y 9 27 Mar 7:47: :14: E N 1 27 Mar 8:28: :54: I Y Mar 16:38: :7: I Y Mar 1:34: :1: I Y Mar 3:24: :52: I Y Mar 7:19: :47: I Y Mar 9:19: :48: I Y Mar 21:4: :9: I N Mar 23:29: :57: I Y Mar 3:23: :52: E Y 19 4 Apr 13:57: :29: E Y Physics Model 25 microphone connected to multiple porous hoses for wind noise reduction, deployed 12.2 km NE of Redoubt's summit (Fig. 1b). The Model 25 has a flat frequency response between.1 and 5 Hz. I53US is the nearest array to the volcano (547 km) and is composed of 8 infrasound elements deployed as a pentagon enclosing a small triangle. Each element consists of a Chaparral Physics Model 5 microphone (flat frequency response between.2 and 5 Hz) connected to a rosette pipe array to reduce wind noise. Seven other infrasound arrays detected parts of the eruption as well: I56US (Newport, Washington), I44RU (Petropavlovsk-Kamchatsky, Russia), I18DK (Qaanaaq, Greenland), NVIAR (Mina, Nevada), I1CA (Lac Du Bonet, Canada), I57US (Pinon Flat, California), and I59US (Kona, Hawaii). For this study, we focus on data from DFR and I53US, as they are the two stations closest to the volcano and have the highest signalto-noise ratio (S/N) and lowest atmospheric perturbation effects. All of the stations are sampled at 2 Hz, except NVIAR (4 Hz) and DFR (1 Hz). All times listed are in UTC Signal processing and detection methods Array processing is performed on all array data to detect signals from Redoubt. Data are divided into 15 s windows with 8% overlap, then band-pass filtered between.75 and 2 Hz. This frequency band is where the majority of acoustic signals are concentrated for the explosive eruptions. Lower frequency energy is apparent for some explosions, but signal contamination from wind is more common at these low frequencies. After filtering, the trace velocity (component Table 2 Table of infrasound arrays that recorded the Redoubt eruption. Redoubt is located at N, W. Azimuth is the bearing from the station to the volcano, in degrees from north. Station Lat. Lon. Azimuth Range ( N) (km) DFR I53US I56US I44RU I18DK NVIAR I1CA I57US I59US of the signal velocity in the plane of the array) and azimuth (geographic bearing opposite the propagation direction across the array) are determined for each time window using a least-squares solution for plane waves traversing the array (Szuberla and Olson, 24). The Fisher statistic (F-stat) (Melton and Bailey, 1957), a common signal processing detector for infrasound array data (Olson and Szuberla, 28), is used as the signal detector. F-stat processing performs a comparison of the variances of both signal and uncorrelated noise, and effectively estimates the signal-to-noise ratio for spatiotemporally correlated signals. First the array data are time shifted for an incoming acoustic wave, which corresponds to the dominant azimuth and trace velocity found using the aforementioned least squares approach (Szuberla and Olson, 24). Then the F-stat is found by computing the variances within each sensor record and between the array sensors using the following relationship: F ¼ V b= ðm 1Þ V w = ðmn 1 ð ÞÞ where V b is the variation between the sensor recordings, V w is the variation within a single sensor recording, M is the number of samples, and n is the number of sensors (Melton and Bailey, 1957; Olson and Szuberla, 28). Melton and Bailey (1957) show that the S/N (P S/N ) can then be derived from the F-stat and number of sensors: P S=N ¼ F 1 n : ð2þ Data segments with an F-stat above a threshold are then counted as detections (Olson and Szuberla, 28). Here we choose a P S/N of 1 as the detection threshold, corresponding to F=9 for an 8-element array. We follow the method of Blandford (1974) and apply the F- stat method to overlapping data segments. Redoubt detections are further restricted to segments originating from ±15 of the theoretical azimuth to the volcano and must have an acoustic trace velocity ( km/s). Fig. 2 shows an example of signal detection processing for one of the Redoubt's explosive eruptions (Event 13). The relative acoustic energies of the eruptions are also estimated. For a spherical source in free space the acoustic energy (E a ) can be calculated by: E a ¼ 4πr2 ρc T Δp 2 ðþdt t ð1þ ð3þ Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

4 4 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx a) b) c) d) Azimuth ( ) Vel (km/s) F stat Pressure (Pa) Event 13, 28 March 29 3:51 3:53 3:55 3:57 3:59 4:1 4:3 UTC Time (HH:MM) Fig. 2. Infrasound detections at I53US for Event 13. a) The array data is beamformed and filtered, then divided into 15 s windows for array processing and detection which produces the b) F-stat, c) trace velocity, and d) azimuth. Data windows exceeding the F-stat threshold (F=9), having acoustic trace velocities, and within ±15 of the theoretical azimuth to the volcano (dashed red line) are considered a detection and are denoted in red. Event 13 begins with an impulsive onset and has two primary energy pulses with high F- stat. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) where r is the source receiver distance, ρ the air density, c the sound speed, Δp pressure perturbation, and T the source duration (Pierce, 1981). Acoustic energy calculations have proven effective in detecting changes in eruptive vigor and ash cloud height (Fee et al., 21a; Steffke et al., 21), discriminating between types of volcanic activity (Woulff and McGetchin, 1976; Marchetti et al., 29), and comparing the relative seismic and acoustic energy release from explosions (Johnson and Aster, 25). However, Eq. (3) assumes purely spherical spreading of the waveforms in free space, which is incorrect at remote distances. Further it does not account for absorption (Sutherland and Bass, 24) or energy loss through ground reflection (Attenborough et al., 26), and is thus not valid at remote distances. Here we use a modified form of Eq. (3) to calculate the relative acoustic energy (E ar ) from Redoubt by disregarding the propagation (4πr 2 ) and acoustic impedance (ρc) terms, and simply integrating the squared pressure over the time interval: Celerity is an often used term that describes the great-circle distance between the source receiver divided by the travel time. A lag of 1 s in the I53US data is allowed to align the waveforms. All signals were filtered using a 4-pole, acausal Butterworth filter. Power spectral density (PSD) estimates are made using Welch's modified periodogram method. Delay and sum beamforming was performed for all waveforms (where indicated) to increase the S/N (Johnson and Dudgeon, 1992). For the signals with periods between 5 and 2 s, the instrument response at I53US has been removed. This band is referred to as the Ultra Long Period (ULP) band (Chouet, 1996), and corresponds to long-duration atmospheric oscillations. ULP signals at I53US were distinguished from noise by comparing waveforms between all eight array elements. Acoustic signal durations are determined using the aforementioned detection thresholds. The acoustic onsets were derived by McNutt et al. (this issue) to the nearest second using data from the local station DFR. E ar ¼ T Δp 2 ðþdt: t ð4þ 3.3. Propagation modeling The relative acoustic energies presented here are thus useful in comparing the eruptive vigor between explosions at Redoubt but not to actual acoustic energies for other volcanoes. The units of E ar are Pa 2 s, and it is calculated only for time segments that meet the detection thresholds previously detailed. Cross-correlation is performed between DFR and I53US waveforms. DFR waveform segments are selected beginning one minute prior to the acoustic onset and ending five minutes after the acoustic signal has ceased. Onsets and durations from McNutt et al. (this issue) are used. The longer event duration is used to fully capture the I53US signals. DFR data are also resampled to 2 Hz to match the I53US sample rate. The corresponding I53US waveform for crosscorrelation is selected by looking 1823 s after the event onset, which corresponds to a typical stratospheric celerity of.3 km/s. Infrasound may propagate long distances due to the relatively low amount of attenuation at infrasonic frequencies and multiple atmospheric waveguides (ducts). The propagation of acoustic energy in the atmosphere is predominantly governed by vertical wind and temperature gradients. The effective sound speed in the atmosphere can be estimated by: p c eff ¼ ffiffiffiffiffiffiffiffiffi γrt þ v n ð5þ where γ is the specific heat ratio, R the universal gas constant, T the temperature, v the horizontal wind vector, and n the ray normal (Whitaker and Norris, 28). The latter term is necessary due to the translational effects of winds in the atmosphere (advection of sound). The sound speed for a typical windless atmosphere of 2 C at sea level is 343 m/s. Following Snell's Law and Eq. (5), sound Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

5 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx 5 waves propagating from a source near the earth's surface will generally be refracted upward as the temperature (and thus sound speed) often decreases with height (Fig. 3b). However, strong winds and/or temperature gradients may cause the sound speed at altitude to exceed that at the source, causing the sound to refract downward, creating a waveguide. In this study we compare the maximum c eff in the stratosphere (4 7 km) to that at the source (3 km height- Redoubt's summit) to obtain an estimate of the strength of the stratospheric duct (c eff ratio). A c eff ratio N1 indicates that a stratospheric duct should be present, causing rays to refract back to the ground. The two most common long-range atmospheric ducts occur in the stratosphere (~5 km) and thermosphere (~1 km) (Fig. 3b). Thus, the two most common long-range atmospheric returns are stratospheric and thermospheric, termed Is and It, respectively. Stratospheric arrivals usually have higher amplitudes due to the shorter propagation paths and the higher attenuation levels above ~6 km. Tropospheric (~b1 km height) arrivals are unlikely at distances N25 km. Both Is and It arrivals will likely have different characteristics at the array. Thermospheric returns arrive with higher incidence angles due to refraction at higher altitudes. Trace velocity (apparent horizontal phase velocity) at the array, v, is related to the incidence angle, θ, through v=c/cos(θ), where c is the sound speed. It arrivals should therefore have higher trace velocities. Azimuthal deviations will also vary between Is and It arrivals, as the two propagation paths will experience different horizontal wind translations. Typical celerities for Is phases range between ~.28 and.31 km/s, while It phases have lower celerities between ~.22 and.26 km/s due to the longer propagation path through the thermosphere. We perform basic propagation modeling to interpret the remote infrasound recordings. Travel times and propagation paths are estimated using ray theory, a type of geometric model which relies on a high frequency approximation to represent propagation paths as rays. This method is useful for estimating travel times and identifying propagation paths. However, it does not account for diffraction and scattering and thus often incorrectly predicts shadow zones where no sound should propagate. We utilize the InfraMAP software program developed and maintained by BBN Technologies (Gibson and Norris, 22) to run the 3-D Hamiltonian Ray Tracing Program for Acoustic Waves in the Atmosphere (HARPA), modified from Jones et al. (1986). HARPA accounts for horizontal and vertical translation of the acoustic wave by winds. For this study, theoretical rays are launched from a 3 km source height (Redoubt summit) between 1 and 6 at 2 intervals. Ray tracing modeling used here is rangedependent, meaning updated atmospheric variables are used along the propagation path, compared to range-independent modeling where only the source profile is considered. Atmospheric specifications are provided by the Naval Research Laboratory (NRL) ground to space (G2S) models (Drob et al., 23). The G2S models provide temperature, wind, and atmospheric composition estimates from the earth's surface to 14 km every 6 h at 1 1 intervals, and include a recent revision to the upper atmosphere (Drob et al., 28). 4. Results 4.1. Infrasound array observations and eruption constraints We detected all 19 numbered Redoubt explosive events at I53US and most have a high S/N. Table 1 lists the explosive events and the detection characteristics for DFR and I53US, as well as the plume heights determined from radar measurements (Schneider and Hoblitt, this issue). Fig. 4 shows the beamed,.2 to 2 Hz waveforms for the events at I53US, time aligned according to the onset at the array. Coherent time periods are outlined in black, while uncorrelated noise is shown in gray. At I53US, most events have relatively short durations (b15 min) and high amplitudes (N2 Pa peak pressure) considering the distance (547 km). Durations range from 2.8 to 31 min, although the durations (as determined here) are likely overestimates of the source duration due to multi-pathing (multiple ray paths arriving at different times). Peak amplitudes vary from.28 to 6.76 Pa and relative acoustic energies range between.18 and 47 Pa 2 s. Fig. 5 is similar to Fig. 4, except that I53US data are band-pass filtered between.5 and.2 Hz. The 29 Redoubt eruption explosive phase consists of four main groups of explosive events, classified solely on the similarity of acoustic characteristics (Table 1, Figs. 4 and 5): (1) Events 1 and 9; (2) Events 2 6, and 8; (3) Events 1 18; and (4) Event 19. Group 1 consists of two events (1 and 9) with relatively long durations (N16 min) and multiple pulses/explosions. Both of these events have low relative acoustic energies (1.37 and 29.7 Pa 2 s), low ash clouds (b11 km) and no ULP energy. Group 2 comprises Events 2 6 and 8. Events 2 6 occur over a span of 21 h on March and have similar acoustic signals. Each event has a relatively long duration (N1 min) and sustained infrasound where the amplitude does not vary significantly during the event. All Group 2 events produced high acoustic energies and ash clouds (N13 km). ULP energy is also clear for Events 2 and 5, and possibly for Events 3, 4, and 6. Event 5 a) b) c) 12 1 Zonal Meri GCP c c eff Altitude (km) Velocity (m/s) Sound Speed (m/s) Redoubt Range (km) IS53 Fig. 3. Environmental profiles and ray tracing from Redoubt to I53US for Event 13. a) Winds above Redoubt for 28 March 6:. Zonal winds (solid black line) show a maximum in the stratosphere between 4 and 65 km height and are minor throughout the rest of the profile. Meridional winds (gray line) are weak up to 12 km. Great circle path (GCP) winds (dotted black line) between Redoubt and I53US have a peak in the stratosphere (~4 65 km). b) Sound speed (c, solid black line) above Redoubt shows a typical high-latitude shape. The effective sound speed (c eff, dotted black line) exceeds that of the source at ~4 km, and has a broad maximum in the stratosphere, primarily due to the zonal winds. c) Ray tracing from Redoubt to I53US, with the majority of the rays being refracted in the stratosphere. The eigenray (dark black line) is refracted around 45 km height and experiences a single ground reflection around 225 km. Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

6 6 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx Mar 6: Mar 7: Mar 8: Mar 9: Mar 12: Mar 3: Mar 16: Mar 17: Mar 7: Mar 8: Mar 16: Mar 1: Mar 3: Mar 7: Mar 9: Mar 21: Mar 23: Mar 3: Apr 13:57 Pressure (Pa) Time past event onset (mins) Fig. 4. I53US.2 2 Hz beamed waveforms for all 19 explosive events. Waveforms are aligned relative to the detected onset ( min), with the next 31 min being plotted. Amplitude scale is the same for all events. Event number is listed on the left and event time denoted on the right. Black waveform portions correspond to segments that meet the Redoubt detection criteria. Most events have relatively short durations (b2 min) and relatively high amplitudes considering the distance. The events are divided into four main groups based on similar infrasound characteristics: 1) Events 1 and 9; 2) Events 2 6, and 8; 3) Events 1 18; and 4) Event 19. Event 7 is only weakly detected. (23 March 12:3) produced the most acoustic energy of any event and highest amplitude at DFR, clipping the microphone at N173 Pa. This corresponds to N Pa at 1 m from the vent (assuming spherical spreading), or roughly 2 atm overpressure. After a lull in activity for ~ 15 h, Event 6 (24 March 3:4) begins with an emergent onset and moderate acoustic energy. This event produced an ash cloud greater than 18 km altitude. No infrasound is detected from Redoubt over the next two days. Event 7 (26 March 16:34) is only weakly detected at I53US, consistent with the low pressure amplitudes for this event at DFR (11 Pa). Due to the low S/N and weak detection, Event 7 is not assigned to any group. Event 8 (26 March 17:24) signals the return of explosive activity, with a large amplitude, high energy eruption that produces the highest ash cloud of the eruptive period (18.9 km). Fig. 6 shows the detections for Event 8 at I53US where a) is the beamed and band-pass filtered waveform (.75 2 Hz), b) F-stat, c) trace velocity, and d) azimuth. Red circles indicate time segments that meet the detection criteria. Event 8 is characterized by high amplitude infrasound for ~ 1 min, with low amplitude infrasound at the beginning and end of the event. The acoustic onset is emergent and has noticeable ULP energy (Fig. 5). Group 3 consists of Events 1 18, which represent a significant shift in acoustic activity at Redoubt from the first nine events. Events 1 18 have relatively short durations, high acoustic energies, impulsive onsets, and peak frequencies of ~.1 Hz. Events 1 15 in particular share similar characteristics, and occur over a span of ~25 h. These six events have high amplitudes (N3.5 Pa), relatively short durations (b1 min), significant ULP energy, and relatively high ash clouds (N12 km). Fig. 2 shows the detections for Event 13 at I53US. The event begins with low amplitude signal for ~2 min, followed by energetic infrasound and very high F-stat for the next ~4 min. Lower level infrasound is detected for another 2 min. These characteristics are similar for Events Events have some similar characteristics to Events 1 15, primarily duration and frequency content, but are of lower amplitude and energy. Event 16 is the only event of Group 3 without noticeable ULP energy. After Event 18 (29 March 3:23), there is a lull in explosive activity for the next 4.5 days and no large explosive events occur. During this period a new lava dome developed and grew extensively (Bull et al., this issue-a). On 4 April 13:57, the final explosive event began (Event 19Group 4), and consists of two pulses of infrasound. The first eruptive pulse has an emergent onset and lasts for ~ 15 min, producing an ash plume to 15.2 km. The second pulse begins ~2 min after the first and has a similar duration, but higher amplitudes and significant ULP energy (Fig. 5). No infrasound from Redoubt is detected by the remote arrays after Event 19, although some minor signals were recorded at DFR during the effusive phase (McNutt et al., this issue). Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

7 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx Mar 6:35 23 Mar 7:1 23 Mar 8: Mar 9: Mar 12: Mar 3: Mar 16: Mar 17:24 Pressure (Pa) Mar 7:47 27 Mar 8:28 27 Mar 16:38 28 Mar 1: Mar 3: Mar 7: Mar 9: Mar 21: Mar 23: Mar 3: Apr 13: Time past event onset (mins) Fig. 5. I53US.5.2 Hz (2 5 s) beamed waveforms for all 19 explosive events. The format is similar to Fig. 4. This frequency band is referred to as Ultra Long Period (ULP) and likely corresponds to relatively long duration oscillations of the volcanic jet or plume. All events with noticeable ULP energy (Events 2,5,8,1 15, and 17 19) also have significant ash plumes (N11 km). The majority of the explosive events are detected at multiple arrays. Fig. 7 shows the waveforms and detections at all IMS arrays within 4 km of Redoubt for 28 March : 15:, corresponding to Events The dotted lines indicate acoustic propagation for.3 km/s celerity, typical of stratospheric ducting. Waveform segments highlighted in red indicate coherent acoustic detections at the array corresponding to Redoubt azimuths. Amplitudes for each array are normalized by the maximum over the time period. I53US (547 km, 29 ) records all four explosive events with high S/N. I56US (2625 km, 314 ), NVIAR (3416 km, 327 ), and I1CA (3628 km, 31 ) clearly record the four events as well. I18DK (3385 km, 285 ) records Event but not 15, likely due to slightly higher noise levels during this period. I44RU (341 km, 55 ) did not record any of the events during this interval. I57US and I59US did not record any 28 March events, but did record at least one event Propagation modeling and atmospheric profiles First-order propagation modeling and analysis of the relevant atmospheric structure is performed to aid in the interpretation of the remote recordings. Fig. 3 shows the wind, sound speed, and ray tracing modeling at Redoubt for 28 March 6:, coincident with a high level of explosive activity and the closest atmospheric profile for Events The winds are defined as: zonal (east west, positive east), meridional (north south, positive north), and the vector wind component along the great circle path (GCP) between Redoubt and I53US. The GCP winds are added to the sound speed in Eq. (5) to obtain the effective sound speed (c eff ). Winds above Redoubt for this time period consist of an easterly zonal wind jet in the stratosphere of ~55 m/s at 6 km. This wind jet is common in the stratosphere, and for northern latitudes, is predominantly positive in the winter (easterly) and negative (westerly) in the summer (Drob et al., 23). Numerous studies have shown the significant influence of the stratospheric wind jet on global infrasound propagation (e.g. Le Pichon et al., 29). Meridional winds are mostly minor, with a slight northerly component. The GCP winds from Redoubt to I53US (bearing ~24 N) are mostly positive, with a peak in the stratosphere around 4 km/s at 55 km and thermosphere around 11 km. The sound speed and c eff above Redoubt in Fig. 3b show a typical profile for early spring at high latitudes. Sound speed decreases with height in the troposphere (~ 1 km), slowly increases through the stratosphere (~1 5 km), decreases again in the mesosphere (~5 9 km), and increases in the thermosphere (N9 km). The c eff profile shows a significant stratospheric duct that begins at ~4 km height, where c eff first exceeds c eff at the source. The c eff ratio for this time and location is ~1.14, indicating a strong stratospheric duct. Of note is the relatively low sound speed at the source (~32 km/s), primarily due to the cold temperatures at this latitude and the elevated source Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

8 8 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx a) b) c) Pressure (Pa) F stat Event 8, 26 March 29 d) Vel (km/s) Azimuth ( ) :51 17:54 17:57 18: 18:2 18:5 18:8 UTC Time (HH:MM) Fig. 6. Infrasound detections at I53US for Event 8. The figure format is similar to Fig. 2, with a) the filtered, beamed waveform, b) F-stat, c) trace velocity, and d) azimuth. Note how the trace velocities transition from moderate values ( km/s) at the beginning of the event to higher values (~.4 km/s) at the end of the event. This likely represents a transition from stratospheric to thermospheric arrivals at the array. height. Matoza et al. (211a) discussed how an elevated source at Sarychev Peak (and potentially other volcanoes) decreases the effective sound speed at the source and increases the likelihood of stratospheric ducting. We note that the cold temperatures often present at high latitudes will also decrease the effective sound speed at the source and enhance stratospheric ducting. Distance from Redoubt (km) /3/28 UTC Hour I59US I57US I1CA NVIAR I18DK I44RU I56US I53US DFR Fig. 7. Infrasound detections for the Redoubt events on 28 March 29 : 15: UTC (Events 12 15). For the array data, red waveform segments indicate coherent energy surpassing the aforementioned detection thresholds. DFR onsets and durations are from McNutt et al. (this issue). Amplitudes are normalized and the dotted lines denote the travel time for an acoustic wave propagating from Redoubt at.3 km/s. I53US detects all four events with high S/N. I56US, NVIAR, and I1CA also detect all four events but with lower S/N. I18DK detects Events but not 15, likely due to increased noise levels. I44RU does not detect any events due to its location to the west of the volcano. I57US and I59US do not detect any of these events, but detected at least one event on another day. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) Fig. 3c shows numerous rays being turned by the stratospheric duct, with higher launch angle rays either being turned in the thermosphere or not refracting back to the ground. The combination of an easterly zonal wind jet, minor meridional winds, and low sound speed at the surface all contribute to the deep stratospheric duct between Redoubt and I53US for this time period. The ray outlined in Fig. 3c is the ray which best connects the source and receiver within a specified tolerance (5 km), and is thus termed the eigenray. This ray has a turning height of ~45 km, travel time of 1834 s (celerity.298 km/s), and a single ground reflection (bounce). The majority of the acoustic energy for Event 13 arrives at I53US at 28 March 3:55:12 (Fig. 2), corresponding to a travel time of 1856 s (calculated celerity.295 km/s). The modeled travel of 1834 s (celerity of.298 km/s) for the eigenray is consistent with the observed travel time. Faster arrivals for Event 13 are possible from other propagation paths, in particular diffracted arrivals from faster stratospheric returns. The predicted azimuthal deviation at I53US for the eigenray at I53US is 4.7, primarily due to the easterly zonal stratospheric winds deflecting the sound. The measured azimuthal deviation for the stratospheric arrivals for Event 13 is 3 to 5 (Fig. 2d), in good agreement with the model predictions. Atmospheric variables above Redoubt for the explosive phase are shown in Fig. 8, including (a) zonal winds, (b) meridional winds, (c) c eff, and (d) the c eff ratio for propagation to I53US. Explosive events are identified by black arrows above (a). The easterly stratospheric zonal wind jet peaks around 26 March and decreases with time. Meridional winds are mostly minor and slightly positive (northerly) during the study period. A moderate-strong stratospheric duct is present (c eff ratio N1) during the entire explosive period, particularly between 23 March and 3 April. Significant stratospheric ducting of infrasound to I53US and other arrays east of the volcano is therefore predicted for most of the explosive period. Westerly propagation from Redoubt (e.g. I44RU) should be inhibited by the stratospheric jet, with only thermospheric arrivals predicted. Diurnal variations in the thermosphere are due to solar tides (Drob et al., 23). These results should be fairly typical of wintertime propagation from volcanoes in southwest Alaska and the Aleutian arc. Summertime propagation will be enhanced to the west. Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

9 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx 9 Fig. 8. Atmospheric profiles for the duration of the explosive phase (23 March 6 April) including a) zonal winds, b) meridional winds, c) effective sound speed, and d) c eff ratio for Redoubt to I53US. The c eff ratio is derived by dividing the maximum c eff in the stratosphere (4 7 km) to that at the source to estimate of the strength of the stratospheric duct. Stratospheric zonal winds are moderately strong and blow to the east (positive values) for the entire experiment. Meridional winds are mostly minor with predominantly positive values in the stratosphere (northerly). The effective sound speed and c eff ratio are high for the majority of the explosive phase, and are likely responsible for the strong atmospheric ducting observed at I53US. Black arrows along the top of the figure denote explosive event times. The detections for 28 March at other IMS arrays (Fig. 7) are also consistent with strong stratospheric ducting to the east. Events are all detected at I53US, I56US, I18DK, NVIAR, and I1CA (highlighted waveform portions in Fig. 7), with the timing consistent with stratospheric arrivals. I44RU is the third closest IMS array (341 km), but did not detect the 28 March events, likely due to its location to the west of the volcano. These results are consistent for the events during the entire Redoubt explosive phase and illustrate the importance of the stratospheric duct. Trace velocities for numerous events at I53US show a transition from moderate (~.33 km/s) values at the onset to high values (~.4 km/s) later in the wave train. Event 8 infrasound and detections are shown in Fig. 6. Trace velocities for the onset and majority of the event lie between.33 and.35 km/s (Fig. 6c). However, trace velocities for the final ~2 min of the event transition to.4 km/s, likely indicating the presence of thermospheric arrivals. Ray tracing for this event predicts primarily stratospheric arrivals (similar to that for Event 13). However, thermospheric arrivals with longer propagation paths and travel times are also predicted. Thermospheric arrivals should produce lower signal levels due to increased attenuation from longer propagation paths and higher absorption in the thermosphere (Sutherland and Bass, 24), which is consistent with the observations in Fig. 6a. Further, azimuthal deviations should also differ for thermospheric arrivals due to the thermospheric acoustic energy encountering different winds. Estimated azimuths (and hence azimuthal deviations) for the end of Event 8 detections differ from the earlier, likely stratospheric arrivals (Fig. 6d). The observation of trace velocities transitioning from moderate to high values was also observed at I53US for signals from the 26 Augustine eruption (Wilson et al., 26), and was interpreted to represent the transition from stratospheric to thermospheric arrivals. Similar observations were also made for the 29 Sarychev Peak eruption (Matoza et al., 211a). Note multiple ducted arrivals will increase the measured duration estimate Local and remote infrasound data comparison Nearly all major features of the DFR waveform (and hence acoustic source) are apparent at I53US, including the impulsive signal onset. This is not a common occurrence given the distance between source (DFR) and I53US (e.g. Herrin et al., 28). Fig. 9 displays the.75 2 Hz filtered waveforms and PSD estimates for DFR (red) and I53US (black) for Event 13 (28 March 3:24:18). The I53US waveforms have been time-aligned and beamed to permit comparison. The Event 13 infrasound signal consists of two pulses of activity, the first lasting ~1.5 min and peaking about 1 min after the onset (~138 Pa at DFR, 4.2 Pa at I53US). The second infrasound pulse is of lower energy and lasts ~3 min. A PSD comparison (Fig. 9c) shows similar frequency content between both stations, consisting of a single broad peak frequency at ~.1 Hz with a gradual roll-off at higher frequencies. The spectral shapes differ above ~3 Hz, where higher frequency source energy likely experiences greater attenuation along the propagation path (Sutherland and Bass, 24). Note the DFR microphone response begins to roll-off below.1 Hz, so comparisons below this frequency must be taken with care. The energy for this event is well above the median IMS noise model (Bowman et al., 29) at I53US above 2 Hz, and well above the high noise model up to 7 Hz, illustrating the high S/N of the event. Cross-correlation between the waveforms is now presented to quantitatively examine the similarity between DFR and I53US. Fig. 1a shows the time-aligned, filtered DFR (red) and I53US (black) waveforms from Event 13. The cross-correlation coefficient for this segment is.83, a high value considering the distance between the stations. Note although the cross-correlation is relatively high, the waveforms appear out of phase. Geometric acoustics theory predicts that as closely spaced propagating rays progressively refract, they will eventually focus into regions where they intersect, termed caustics (Fig. 11a). Here some of the assumptions of ray theory are no longer valid. The wave passing through the caustic will undergo a 9 phase shift, equivalent to taking the imaginary part of the Hilbert transform of the signal (Fig. 11). Geometric acoustics states that a wave will encounter a caustic after each ground bounce (reflection) and its subsequent turning point (Kinney and Pierce, 198). Although primarily theoretical, phase shifts due to acoustic waves passing through caustics have also been confirmed by experiment (Talmadge et al., 28). Propagation from Redoubt to I53US is predicted to encounter a single ground bounce at ~225 km (Figs. 3c, 11), corresponding to passage through a single caustic after the Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores

10 1 D. Fee et al. / Journal of Volcanology and Geothermal Research xxx (211) xxx xxx a) b) Pressure (Pa) 5 Pressure (Pa) Event 13, 28 Mar 29 3:24:26 DFR 12 km 3:24:3 3:25: 3:25:3 3:26: 3:26:3 3:27: 3:27:3 3:28: IS km 3:54:3 3:55: 3:55:3 3:56: 3:56:3 3:57: 3:57:3 3:58: UTC Time c) Power (db//2e 6 Pa 2 /Hz) PSD for Event 13 IS53 DFR Frequency (Hz) Fig. 9. Event 13 local and remote infrasound comparison. Waveforms for a) DFR (12 km, red) and b) I53US (547 km, black) show very similar features. c) PSD comparison for the event shows a single, broad frequency peak at ~.1 Hz and roll-off at higher frequencies for both stations. The spectral shape is similar between the two stations except for at higher frequencies where absorption is greater for I53US. The dashed spectra in c) represent the high, median, and low noise models of the IMS network (Bowman et al., 29). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) ground reflection (and subsequent 9 phase shift). We Hilbert transform the I53US waveform and repeat cross-correlation with DFR in Fig. 1b. The cross-correlation value increases to.89, and visual inspection reveals all principal DFR waveform features (and phases) are now apparent at I53US. Note that high waveform similarity is uncommon in long range infrasound propagation, and likely occurs due to the deep stratospheric duct and only one ground bounce. Increased attenuation is also predicted due to the ground bounce, although this effect is relatively minor at low frequencies (Attenborough et al., 26). Propagation to the other arrays will encounter numerous ground bounces and increased multi-pathing, leading to much lower waveform correlations at greater distances. Waveform similarity varies considerably from event to event (Table 1). However, even when correlation values are relatively low, principal source features are still apparent at I53US. Fig. 12 shows the aligned DFR (red) and I53US (black) waveforms and PSD for Event 8 (26 March 17:24). Although the cross-correlation value is only.54, the main source waveform features are apparent at I53US, particularly the acoustic energy arriving as a function of time. PSD comparisons for the explosive events show similar energy distribution as well. The lower cross-correlation value is likely due to a more complex, extended source waveform Acoustic energy and hazardous emissions In this section we compare the infrasound energy with satellite derived SO 2 mass, and use the acoustic energy integral as a proxy for the eruptive vigor. Previous studies have shown a broad correlation between infrasound energy and ash cloud height for continuous emissions (Fee et al., 21a; Steffke et al., 21); however, the shorter duration explosive events here do not permit a similar detailed correlation, as the satellite sampling interval is often greater than the event duration. We do note that all events with significant low frequency infrasound (b.5 Hz) produced hazardous ash clouds, consistent with previous infrasound studies of large eruptions (Fee et al., 21a; Fee et al., 21b; Steffke et al., 21). McNutt et al. (this issue) provide a detailed look at the relationship between infrasound parameters at DFR versus plume heights and seismic data, and similar correlations using the remote infrasound data should yield comparable results due to the high waveform similarity for most of the events. Daily SO 2 mass estimates presented here are made using data from the Ozone Monitoring Instrument (OMI). Due to OMI's temporal resolution (~1 3 passes per 24 h at high latitudes) it is difficult to differentiate the SO 2 release between the relatively short duration explosive events, thus the SO 2 mass produced by Redoubt is estimated on a daily basis. All SO 2 estimates used here are from Lopez et al. (this issue), and the reader is referred to that manuscript for a more detailed explanation of the SO 2 estimation methods and interpretations. Relative acoustic energies at I53US are calculated on an hourly basis following the methods outlined in Section 3.2. There is a high correlation between the cumulative amounts of infrasound energy and SO 2 produced during the 29 Redoubt eruption explosive phase between 22 March and 6 April. Fig. 13 shows the cumulative daily SO 2 mass estimates (red) and cumulative relative infrasound energy (E ar black) for the explosive phase. Black arrows at the top of the figure indicate numbered explosive events. The relative amounts of daily SO 2 mass detected agree well with the E ar recorded at I53US. Events 2 5 produced extensive infrasound on 23 March (total E ar =1357 Pa 2 s), compared to the ~54 kilotonnes (kt) of SO 2. A similar amount of 6 kt SO 2 is detected the next day, presumably erupted mostly during Event 6 (23 March 3:4, E ar =186.7 Pa 2 s). Note the amount of measured SO 2 for this day is greater than for the previous day. Rather than being a more SO 2 rich eruption, it is likely that some of the measured SO 2 from 24 March is residual SO 2 from the previous day, and essentially double counted, as the shape of the SO 2 plume on 24 March is similar to 23 March and the bulk of the plume is far from Redoubt (Lopez et al., this issue). After a lull in activity over the next day and a half in which no infrasound energy or SO 2 emissions were detected, 12 explosive events occurred between 26 March 16:34 and ~29 March 4:. Event 7 is quite small (E ar =.18 Pa 2 s, plume height 6.7 km) and likely did not produce significant emissions. Events 8 15 have relatively high acoustic energies, with events 1 15 having similar characteristics (Section 4.1). Overall relatively similar amounts of cumulative SO 2 (~72 kt) and E ar (~86 Pa 2 s) were generated by Events Similarly, Events (28 March 21:4 29 March 3:23) produced low (and comparable) amounts of both acoustic energy (~17 Pa 2 s) and SO 2 (b2 t on 29 March). Minor but significant amounts of SO 2 (~1.5 5 kt) were detected between 3 March and 4 April, coincident with low E ar and lava dome growth (Bull et al., this issue-a). The final explosive event Please cite this article as: Fee, D., et al., Combining local and remote infrasound recordings from the 29 Redoubt Volcano eruption, J. Volcanol. Geotherm. Res. (211), doi:1.116/j.jvolgeores