Episodes of cyclic Vulcanian explosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat

Transcription

1 Episodes of cyclic Vulcanian eplosive activity with fountain collapse at Soufriere Hills Volcano, Montserrat T. H. DRUITT 1, S. R. YOUNG 2, B. BAPTIE 2, C. BONADONNA 3, E. S. CALDER 3, A. B. CLARKE 4, P. D. COLE 5, C. L. HARFORD 3, R. A, HERD 6, R. LUCKETT 2, G. RYAN 7 & B. VOIGHT 4 1 Laboratoire Magmas et Volcans (UMR 6524 & CNRS), Universite Blaise Pascal, 5 rue Kessler, 6338 Clermont-Ferrand, France ( T.Druitt@opgc.univ-bpclermont.fr) 2 British Geological Survey, Murchison House, Edinburgh EH9 3LA, UK ^Department of Earth Sciences, University of Bristol, Queens Road, Bristol BS8 1RJ, UK 4 Department of Geosciences, Pennsylvania State University, University Park, PA 1682, USA 6 Centre for Volcanic Studies, University of Luton, Park Square, Luton LU1 3JU, UK 6 British Geological Survey, Keyworth, Nottingham NG12 5GG, UK 7 Environmental Science Department, Institute of Environmental and Natural Sciences, University of Lancaster, Lancaster LAI 4YQ, UK Abstract: In 1997 Soufriere Hills Volcano on Montserrat produced 88 Vulcanian eplosions: 13 between 4 and 12 August and 75 between 22 September and 21 October. Each episode was preceded by a large dome collapse that decompressed the conduit and led to the conditions for eplosive fragmentation. The eplosions, which occurred at intervals of 2.5 to 63 hours, with a mean of 1 hours, were transient events, with an initial high-intensity phase lasting a few tens of seconds and a lower-intensity, waning phase lasting 1 to 3 hours. In all but one eplosion, fountain collapse during the first 1-2 seconds generated pyroclastic surges that swept out to 1-2 km before lofting, as well as high-concentration pumiceous pyroclastic flows that travelled up to 6 km down all major drainages around the dome. Buoyant plumes ascended 3-15 km into the atmosphere, where they spread out as umbrella clouds. Most umbrella clouds were blown to the north or NW by high-level (8-18 km) winds, whereas the lower, waning plumes were dispersed to the west or NW by low-level (<5 km) winds. Eit velocities measured from videos ranged from 4 to 14ms -1 and ballistic blocks were thrown as far as 1.7 km from the dome. Each eplosion discharged on average m 3 of magma, about one-third forming fallout and two-thirds forming pyroclastic flows and surges, and emptied the conduit to a depth of.5-2 km or more. Two overlapping components were distinguished in the eplosion seismic signals: a low-frequency (c. 1 Hz) one due to the eplosion itself, and a high-frequency (>2Hz) one due to fountain collapse, ballistic impact and pyroclastic flow. In many eplosions a delay between the eplosion onset and start of the pyroclastic flow signal (typically 1-2 seconds) recorded the time necessary for ballistics and the collapsing fountain to hit the ground. The eplosions in August were accompanied by cyclic patterns of seismicity and edifice deformation due to repeated pressurization of the upper conduit. The angular, tabular forms of many fallout pumices show that they preserve vesicularities and shapes acquired upon fragmentation, and suggest that the eplosions were driven by brittle fragmentation of overpressured magmatic foam with at least 55vol% bubbles present in the upper conduit prior to each event. Vulcanian eplosions are a common feature of andesitic volcanoes (Morrissey & Mastin 2). Eamples include historic eruptions of Arenal (Costa Rica), Ngauruhoe (New Zealand), Fuego (Guatamala) and Augustine (Alaska) volcanoes (Melson & Saenz 1973; Martin & Rose 1981; Nairn & Self 1978; Kienle & Shaw 1979). Recent eamples include eplosions at Lascar (Chile) in the period (Matthews et al 1997), Pinatubo (Philippines) in 1991 (Hoblitt et al 1996) and Galeras (Columbia) in 1992 (Sti et al 1997). Individual Vulcanian eplosions typically discharge between 1 2 and 1 6 m 3 of magma and comminuted accidental debris in cannon-like detonations, generating buoyant plumes mostly between 5 and 2km high. Most of the erupted mass is discharged on a time scale of 1 to 1 3 seconds. Eit velocities ranging from about 5 to at least ms -1 have been observed and blocks up to 2m or more can be thrown ballistically up to several kilometres during the initial vent-clearing phase (Self et al 1979; Fagents & Wilson 1993). Fallback of part of the discharging material (fountain collapse) generates pyroclastic flows. Successions of powerful eplosions occur from some volcanoes, with intervals ranging from 1 2 to 1 7 seconds (Self et al 1979). Vulcanian eplosions are attributed to the interaction of magma with eternal water or to the sudden release of highly pressurized, vesicular magma beneath a cooled or degassed cap (Self et al 1979; Fagents & Wilson 1993; Woods 1995; Sti et al 1997; Sparks 1997). Involvement of eternal water is invoked when there is direct field evidence, or when the gas content implied by models eceeds likely maimum magmatic values of a few per cent (Self et al 1979; Fagents & Wilson 1993). This paper concerns two episodes of Vulcanian eplosions that took place in the second half of 1997 at the lava dome of Soufriere Hills Volcano, Montserrat. Thirteen of these occurred in August and 75 in September and October. A remarkable feature was the repeated and regular nature of the eplosions, intervals ranging from 2.5 to 63 hours with a strong mode at c.1 hours. The activity in August was accompanied by cyclic patterns of edifice deformation and seismic energy release (Voight et al 1998, 1999). The eplosions generated plumes up to 15km and, in all but one, fountain collapse formed pumice-and-ash pyroclastic flows that travelled up to 6 km from the vent. The cyclicity of the eplosions permitted accurate short-term forecasting and hazard assessment over more than a month of intense activity. It also allowed unusually detailed study by a variety of techniques. Eplosions were filmed both during the day and at night, and when possible from multiple locations. Maimum plume heights were estimated, fallout and pyroclastic flow deposits were mapped and sampled, and measurements were made of the sizes and distributions of ballistic blocks. Seismicity before, during and after eplosions was recorded by the Montserrat Volcano Observatory (MVO) broadband system (Neuberg et al 1998), and edifice deformation was measured by a combination of electronic distance measurement, global positioning system (GPS) and tiltmeter (Voight et al 1998, 1999). We describe the 1997 eplosions and their products and make estimates of erupted masses, eit velocities, fragmentation pressures and conduit drawdowns. In some cases only ranges and averages of the parameters are presented, as practical considerations of time and safety limited data acquisition for most individual events. We also describe the eruption seismic signals and recognize two components: one related to the eplosion itself and the other to fountain collapse, ballistic impact and pyroclastic flow transport. The data and analysis in this paper are complemented by numerical DRUITT, T. H. & KOKELAAR, B. P. (eds) 22. The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, /2/S15 The Geological Society of London

2 282 T. H. DRUITT ET AL. modelling of the eplosion plumes by Clarke et al. (22) and of the associated conduit flow by Melnik & Sparks (22b). Bonadonna el al. (22a, b) describe the fallout from the eplosions and develop mathematical models of tephra dispersal. The pyroclastic flows and their deposits are described by Cole et al. (22). Two other periods of eplosive eruption at Soufriere Hills Volcano have been described by Robertson et al. (1998) and Norton et al. (22). During the night of 17/18 September 1996 there occurred a 5-minute sub-plinian eruption that formed a plume at least 11.3 km high (above sea level), but no fountain-collapse pyroclastic flows (Robertson et al. 1998). Multiple short-lived (Vulcanian) eplosions occurred in late 1998 and in 1999, during the period of virtually no magma etrusion (Norton et al. 22). These relatively weak eplosions generated plumes that typically rose to heights of 3-6 km above sea level and fountain-collapse pyroclastic flows that travelled up to 3 km from the lava dome. We begin by summarizing the main features of the eruptive period and, in particular, events of the period from July to October 1997 during which the eplosions described in this paper occurred. Montserrat local time (four hours behind universal time) is used throughout the paper unless noted. Place names are given on Figure 1. All plume heights are given above sea level. The eruption of Soufriere Hills Volcano from 1995 to 1999 Soufriere Hills Volcano is an andesitic lava dome comple situated in southern Montserrat, in the Lesser Antilles island arc. Detailed overviews of the eruptive period have been given by Young et al. (1997, 1998), Kokelaar (22) and Sparks & Young (22). The eruptive vent was situated in an ancient sector-collapse scar (English's Crater) about 1 km across and open to the east (Fig. 1). The western rim of English's Crater is called Gages Wall and the southern rim is called Galway's Wall (Fig. 1). The flanks of Soufriere Hills are scarred by radial valleys (locally called ghauts), which served to channel pyroclastic flows. Initial phreatic eplosions began in July Magma reached the surface in November 1995, and a lava dome began to form. The first dome-collapse pyroclastic flows occurred in March 1996, and flows first reached the sea down the Tar River valley in May of the same year. Major dome collapses occurred in July and August, 1996, and on 17 September 1996 a major collapse of the dome was followed by an eplosive eruption (Robertson et al. 1998). Dome collapses and associated pyroclastic flows continued throughout 1997, with particularly large ones occurring on 25 June, 3 August, 21 September, 4 November and 6 November (Cole et al. Fig. 1. Map of southern Montserrat, showing the dome location inside English's Crater, the principal drainages (ghauts) around the dome, and the area of impact from pyroclastic surges and flows of the entire phase of the eruption.

3 EPISODES OF CYCLIC EPLOSIVE ACTIVITY ). The two episodes of Vulcanian eplosions reported in this paper followed the collapses of 3 August and 21 September On 26 December 1997, Galway's Wall of English's Crater failed, sending m 3 of the wall, lava dome and dome talus down the White River valley as a debris avalanche and high-energy blockand-ash flows (Sparks et al. 22). A pyroclastic density current generated by decompression of gases trapped in the dome interior devastated 1 km 2 of southern Montserrat. Magma etrusion ceased in March 1998, then resumed again in November The total volume of magma discharged over the 28 months of dome formation was.3 km 3 dense-rock equivalent (DRE). Overall magma flu and the sizes of gravitational collapses increased during the period, but with some significant fluctuations. Discharge rates during the first year of etrusion were mostly less than 2 m 3 s -1, but by June 1997 had risen to more than 7-8 m 3 s -1 (Sparks et al. 1998; Sparks & Young 22). The magma is a crystal-rich andesite containing phenocrysts of plagioclase, hornblende, orthopyroene, quartz, and Fe-Ti oides set in a groundmass of microphenocrysts, microlites and rhyolitic glass (Murphy et al. 2). It is believed to have formed by heating and remobilization of a pre-eisting crystal-rich (6-65 vol%) mush. The pre-eruptive liquid phase of the magma was saturated with 4.3±.5wt% water, as determined from glass inclusion analysis and phase equilibria studies (Barclay et al. 1998; Devine et al. 1998#). This corresponds to a water-saturated magma reservoir depth of 5 to 6 km beneath the vent, which is consistent with seismic evidence (Aspinall et al. 1998). Dome growth was accompanied by cyclic patterns of ground deformation and seismicity with periodicities of 3 to 3 hours and attributed to non-linear processes of gas esolution, crystallization, rheological stiffening and pressurization in the conduit beneath the lava dome (Voight et al. 1998, 1999). Ground deformation was measured by tiltmeters installed near the rim of English's Crater. During a typical tilt cycle there was a slow inflation of the edifice (5-3urad), followed by abrupt deflation and associated gas emission, ash-venting, dome collapse or eplosion. Associated seismic swarms built up during inflation, then declined during deflation. Seismicity included volcanotectonic and long-period earthquakes, rockfall, pyroclastic flow and eplosion signals, and tremor. Most seismic signals forming the swarms were of hybrid type, which combined the high frequencies of volcanotectonic earthquakes with low-frequency components (Miller et al. 1998). Two observatory stations were active during the eplosive periods of Prior to early September, the MVO was sited 6km NW of the dome (MVO South, Fig. 1), whereas from then onwards it was located in northern Montserrat (MVO North, not shown on Fig. 1). Eruptive chronology from July to October 1997 Buildup to the August eplosions The scar left by the m 3 (DRE) 25 June dome collapse (Loughlin et al. 22) began to fill in rapidly during the last week of June, about 65% of the void having been filled by 1 July. Between 28 June and 5 July there was intense pyroclastic flow activity. Multiple block-and-ash flows were shed up to 3.5km down Mosquito and Fort Ghauts, up to 1.1 km down Tuitt's Ghaut, and up to.5km down the White River valley. Many of these started with a resounding boom, and a vertical ash column ascended to an altitude of more than 1km. A period of intense ash emissions began on 8 July and persisted through 13 July. Peaks in ash emission often coincided with the peaks of tilt cycles. Some preceded small block-and-ash flows into Mosquito Ghaut and Gages valley; small ash columns reached heights of no more than 3km before dissipating. By 17 July the highest point on the new dome growth nested in the 25 June scar had reached 957m above sea level. The total volume of the dome on 17 July was estimated to be m 3. The level of seismic and eruptive activity in late July was generally low. The 25 June scar had filled in, and the dome had a more or less flat summit 96m above sea level. Activity was characterized by low-amplitude broadband tremor associated with multiple rockfalls and small block-and-ash flows, particularly over Gages Wall. The August eplosions In retrospect, tilt and hybrid cycles related to the impending eplosive activity began on 31 July. Pale, weakly convecting plumes of ash rose almost continuously to heights of between 4.5 and 6km, and small block-and-ash flows travelled as far as 2 km down Gages valley and Tuitt's Ghaut. High levels of long-period and hybrid seismicity continued on 1 August, peaking at one event per minute at the top of the tilt cycle, and a number of detonations were heard coming from the volcano. At least one correlated with an impulsive signal recorded on the broadband seismometers and eplosive activity was surmised. A view of the dome on the same day revealed a small horseshoe-shaped depression in its west side above Gages Wall. Activity over the net three days was characterized by further tilt cycles and hybrid swarms every 9-12 hours. Associated venting became increasingly eplosive and the block-and-ash flows more voluminous. The dome was observed at 14: on 3 August, when an active face of large blocks and spines was present high above Gages Wall. From 18: to 2:3, at the peak of the afternoon tilt cycle on 3 August, a succession of block-and-ash flows travelled down Gages valley. A boom heard at 16:28 at MVO South may have been a small eplosion of the dome or of a gas tank ignited by the flows. Then at 18:1 a m 3 (DRE) block-and-ash flow descended the length of Fort Ghaut to the sea, causing etensive damage in Plymouth. The first clear eplosive activity occurred on 4 August. A blockand-ash flow at 6:3 was accompanied by a loud rumbling, followed by fallout of lithic and crystal lapilli up to 5mm in diameter at MVO South. A second eplosion at 16:43, following a hybrid swarm, sent a dark grey jet inclined at about 6 to the horizontal northwards from the dome. Large blocks were observed to follow ballistic trajectories. Moments later, block-and-ash flows swept 3.5km down Tuitt's Ghaut, 3.5km down the Tar River valley to the sea, 4km down Fort Ghaut to the sea, and an unknown distance down the White River valley. The resulting ash plume rose to 4.5km, and fragments of dome rock and dense pumice as large as 15 mm fell at MVO South. Between the morning of 5 August and the morning of 12 August another 11 eplosions occurred. They are listed in Table 1. Each generated pumice-and-ash pyroclastic flows by fountain collapse and showered the island with pumice-rich fallout. The first nine eplosions occurred regularly every 1 to 12 hours during or immediately after hybrid earthquake swarms. The last two occurred on 11 and 12 August, and were slightly weaker than the others. The August eplosions occurred from a circular crater ecavated in the summit of the dome. Observations of crater development were hampered by cloud cover, but the following sequence is deduced. Rockfalls over Gages Wall at the end of July and on 1 and 2 August generated a summit crater with a low lip to the west. Significant enlargement took place on 3 August during the 18:1 collapse. Further enlargement of the crater took place during the eplosions of 4 August, which discharged large quantities of dense dome rock as well as pumice. By midday on 5 August (after the eplosion of 4:45, but before that of 16:57; Table 1), there eisted a circular, funnel-shaped crater at the top of the dome with a rim diameter of ± 2 m. The crater was seen again clearly at 7:3 on 7 August, when the highest point on the northern rim lay at 94 m above sea level, and that on the southern rim at 98m. There was also a low lip in the crater wall (87 m) above Gages valley, showing that the crater was at least 11m deep. The crater persisted through the eplosion of the morning of 8 August, and was seen from MVO South at 17: on 9 August. When seen again at midday on 1 August, a small new lobe of lava nested within the crater was visible above the western crater lip. Following the final eplosion on 12 August, lava continued to be etruded within the crater.

4 284 T. H. DRUITT ET AL. Table 1. Characteristics of the Vulcanian eplosions of August, September and October, 1997 Date Local time* Intervalf (hrmin) Plume height! (km) Pyroclastic flows Ta W Tu M Ty G Wr Date Local time* Interval! (hr:min) (km) Plume height; Pyroclastic flows Ta W Tu M Ty G Wr 4 Aug. 4 Aug. 5 Aug. 5 Aug. 6 Aug. 6 Aug. 7 Aug. 7 Aug. 7 Aug. 8 Aug. 8 Aug. 1 1 Aug. 12 Aug. 22 Sep. 22 Sep. 22 Sep. 23 Sep. 24 Sep. 24 Sep. 24 Sep. 25 Sep. 25 Sep. 25 Sep. 26 Sep. 26 Sep. 27 Sep. 27 Sep. 27 Sep. 28 Sep. 28 Sep. 28 Sep. 29 Sep. 29 Sep. 29 Sep. 29 Sep. 3 Sep. 3 Sep. 1 Oct. 1 Oct. 1 Oct. 2 Oct. 2 Oct. 2 Oct. 4 Oct. 6:3 16:43 4:45 16:57 4:2 14:36 :34 12:5 21:55 1:32 2:51 11:38 1:12 :57 1:45 2:42 7:23 :34 1:54 17:16 3:54 11:9 2:5 4:25 14:56 :1 9:46 17:15 4:28 1:34 23:3 6:26 11:23 16:48 21:57 4:43 17:44 5: 11:34 17:4 1:5 12:53 22:5 8:33 1:13 12:2 12:12 11:5 1:34 9:58 11:31 9:5 12:37 1:19 62:47 22:35 9:48 9:57 1:41 17:11 1:2 6:22 1:38 7:15 8:56 8:2 1:31 9:5 9:45 7:29 11:13 6:6 12:29 7:23 4:57 5:25 5:9 6:46 13:1 11:16 6:34 6:6 7:25 11:48 9:57 33: > ~ < > <12.2 <<1.7 ~ O O O O O 4 Oct. 5 Oct. 5 Oct. 5 Oct. 6 Oct. 6 Oct. 6 Oct. 7 Oct. 7 Oct. 8 Oct. 8 Oct. 9 Oct. 9 Oct. 1 Oct. 1 Oct. 1 1 Oct. 12 Oct. 12 Oct. 13 Oct. 13 Oct. 14 Oct. 14 Oct. 14 Oct. 15 Oct. 15 Oct. 15 Oct. 15 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 17 Oct. 17 Oct. 17 Oct. 17 Oct. 18 Oct. 18 Oct. 19 Oct. 19 Oct. 2 Oct. 2 Oct. 21 Oct. 21 Oct. 18:27 2:53 1:41 18:41 2:44 1:42 17:5 4:6 16:2 3:47 15:1 3:3 12:32 4:13 18:4 17:57 7:55 22:24 9:32 15:24 1:36 13:48 23:16 5:47 8:33 14:5 22:2 2:51 6:35 9:44 14:2 18:48 4:1 12:35 16:5 23:18 6:48 15:17 5:13 21:27 5:4 15:13 11:39 19:2 9:54 8:26 7:48 8: 8:3 7:58 7:8 1:16 11:56 11:45 11:23 11:53 9:29 15:41 14:27 23:17 13:58 14:29 11:8 5:52 1:12 12:12 9:28 6:31 2:46 6:17 7:3 4:31 3:44 3:9 4:36 4:28 9:13 8:34 3:3 7:13 7:3 8:29 13:56 16:14 7:37 1:9 2:26 7: <12.2 > < ~ > >4.6 > < > O O O O O O O O O O O O O O O O * Local time is 4 hours behind universal time. ftime interval since the previous eplosion. Best estimates of maimum plume heights (above sea level) based on National Oceanic and Atmospheric Administration (NOAA) data, supplemented by Montserrat and Trinidad airport reports and by Abney-level measurements of the plume top made from the MVO. A blank denotes lack of data. Additional height estimates were obtained using plume-top temperatures on GOES satellite images (Bonadonna el al. 22b) for three eplosions: 12:5, 7 Aug. (9.3km); 14.56, 26 Sep. (11.3km); 9:46, 27 Sep. (1.8km). f Presence () or absence (o) of fountain-collapse pumice-and-ash flows. Ta, Tar River valley; W, White's Ghaut; Tu. Tuitfs Ghaut; M. Mosquito Ghaut; Ty, Tyre's Ghaut; G, Gages valley; Wr, White River valley. A blank denotes lack of observations. Following the initiation of Vulcanian activity in early August, major changes were made to the volcanic hazards map of the island, resulting in an enlargement of the evacuated zone and northward displacement of inhabitants under threat from this new style of activity (Kokelaar 22). Renewed dome growth after the August eplosions The new dome lobe in the summit crater developed a large spine, the elevation of which was 95m by 13 August. By 14 August the crater was nearly filled in, and by 19 August the entire dome had reached its pre-3-august volume, growth being concentrated in an area about 15m wide above Gages valley. Initially during this period the seismic activity continued to be characterized by intense hybrid earthquake swarms occurring at approimately 8-hour intervals. The intense hybrid activity that characterized the August eplosive period continued until 19 August, with block-and-ash flows occurring mainly in the upper Gages valley from the growing dome, immediately after peaks in earthquake activity. Throughout late August and early September, rockfall and pyroclastic flow signals dominated the seismic records, with activity confined to the western and northern flanks of the dome. A small dome collapse (about m 3 ) took place on 3 August. A slow

5 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 285 increase in block-and-ash flow activity occurred through mid- September, with material being shed across Farrell's Plain and into Tuitt's Ghaut. Rapid dome growth was occurring at this time, with the unstable active face located above the northern wall of the crater. Earthquake activity remained at a low level, with occasional swarms of hybrid earthquakes, some of which included the largest individual events recorded by the broadband network up to that time. Heightened long-period earthquake activity was also notable, with events often occurring immediately before pyroclastic flow generation. At least one long-period event, on 16 September, correlated with an audible detonation from the dome. By 28 August, the dome volume was more than m 3 and the total volume of magma erupted to form both lava and pyroclastic deposits was m 3 DRE. Dome collapse of 21 September Hybrid earthquake swarms recommenced 24 hours prior to the dome collapse of 3:54 on 21 September and continued after the collapse at the same level. The swarms were neither long nor intense compared with previous ones; individual events were not large and neither were there visual signs of imminent collapse. The onset of increased seismic amplitude related to this collapse is timed at 3:54, although the first 8 minutes of activity was not at a high level, probably registering small dome collapses with pyroclastic flows in the upper and middle parts of Tuitt's Ghaut. A sharp increase in signal amplitude at 4:2 marked the start of sustained high-amplitude signals on all stations, typical of those generated during major dome collapse. Minor pulsing occurred throughout this phase, which lasted until 4:17, a total of 15 minutes. Within this phase, the highest amplitude signals were recorded between 4:11:1 and 4:13:7; the signal amplitude at this time was markedly higher than at any other time, peaking at 4:12:1. The signal had dropped to background level by 4:24, giving a total duration for the dome collapse of 3 minutes. During the collapse, which involved m 3 (DRE) of the dome, block-and-ash flows moved down Tuitt's Ghaut and White's Ghaut to the ocean, spreading out over the area of Spanish Point (Cole et al. 22). Associated pyroclastic surges covered interfluves between the ghauts, causing the burning of Tuitt's and parts of other villages not overrun by the flows. The ash plume associated with the block-and-ash flows reached an altitude of 9-12 km, causing ash fall over much of Montserrat. The post-collapse dome had a deep scallop-shaped scar on its northern flank, etending back about m. There was a prominent opening and chute above the head of Tuitt's Ghaut, eroded by pyroclastic flows as dome collapse proceeded. The September and October eplosions The first eplosion of this episode occurred about 2 hours later at :55 on 22 September and the last at 19:2 on 21 October. A total of 75 eplosions occurred over a 3-day period (Table 1). Intervals varied between 2.5 and 33.5 hours, with an average of 9.5 hours. No systematic variation of plume height with time occurred from 22 September until 11 October, when there occurred a series of at least ten relatively weak eplosions with plume heights of 7 km or less over a period of five days (Table 1). Gaps in the plume-height record at this time mean, however, that it cannot be ecluded that larger eplosions also occurred. After 16 October, plume heights increased again until the end of the eplosive episode on 21 October. The first eplosion produced a crater at the southern edge of the 21 September collapse scar. The shape and size of this crater changed only gradually thereafter, becoming larger in diameter and probably deeper. The volume of the dome at this point was m 3. An estimate of the volume of the crater ecavated during the first few eplosions was m 3 (Fig. 2). Successive eplosions deposited a tephra rampart on the northern side of the crater, effectively completing the near-circular crater wall across the 21 September collapse scar. Minor reaming of the crater walls occurred throughout the eplosion sequence, but good views into the crater were scarce so that accurate estimates of the volume increases were impossible to obtain. Within a day of the cessation of eplosions on 21 October, a lava lobe was seen growing within the crater. Within a few days, lava had filled the crater and overspilled the tephra rampart, and by 3 November had largely filled the entire 21 September collapse scar. Observations also suggest that a small lobe started to grow in the base of the eplosion crater during the longest break between eplosions (2 to 4 October; Table 1). The Vulcanian eplosions The characteristics of the eplosions are listed in Table 1. Of the 88 eplosions, 37 occurred during hours of darkness and about 3 under poor weather conditions. About 2 eplosions were well observed and documented, five of which were at night. The following descriptions are based on field observations and on ongoing Fig. 2. The crater formed early during the eplosion sequence of September and October The photo is taken from the NE. The crater has a rim diameter of about m and opens into Tuitt's Ghaut.

6 286 T. H. DRUITT ET AL. Fig. 3. Sequence of events during a typical Vulcanian eplosion in analysis of video footage. To a first approimation, the eplosions and their products were similar and they are described together. The main features of a typical eplosion are shown schematically in Figure 3. Sequences of photographs of three typical eplosions are shown in Figures 4, 5 and 6. General description of the eplosions Visual precursors to the eplosions were rare, although on several occasions increases in fumarolic activity around the dome were noted in the preceding seconds to minutes. Each eplosion consisted of two fairly well defined phases: (1) an initial, high-intensity phase lasting about 1 minutes and including the main eplosion and peak magma discharge rates (a few tens of seconds), fountain collapse, formation of a buoyant eruption plume, and the ascent of the plume to its neutral buoyancy level in the atmosphere; (2) a drawn-out, much lower-intensity waning phase lasting a few tens of minutes (typically 1 to 3 hours) and characterized by relatively weak, pulsatory venting of gas and ash. Fountain collapse was limited to the first 1-2 seconds of each eplosion. Each eplosion began with the rapid rise of numerous dark-grey finger jets of ash and debris (Fig. 3a). Condensation of atmospheric moisture ahead of the jets, indicative of shock waves, was observed in some eplosions (Clarke el al. 22). There then followed a loud and steady roaring, like an aircraft, punctuated by further detonations as subsequent jets were discharged. The initial jets were relatively weak, but subsequent ones became progressively more vigorous with time over the first few seconds of the eplosion. As the jets rose, they decelerated rapidly. Decimetre- to metre-sized ballistic blocks detached from the leading edges of many jets and were thrown outwards in high, curving arcs as far as 1.7km from the dome. Impact of the blocks with the ground kicked up clouds of ash visible from a distance (Fig. 6). The finger jets, having reached their maimum elevation, then collapsed back towards the ground. Simultaneously, part of the material that had ingested enough air began to rise as one or more buoyant plumes, which then merged into a single large plume. Viewed from a distance, the collapsing fountain took the form of a hemispherical cap, through which the buoyant plume then pierced (Fig. 3b). The fountain-collapse height was typically between and 65 m above the crater rim, although, in some eplosions, later jets remained momentum-driven up to more than 1 m. Impact of the collapsing material with the ground generated highly epanded pyroclastic surges, which swept out from the volcano with frontal velocities of 3-6 ms -1 (Figs 3c, 4, 5 and 6). On some occasions, the surges moved almost as quickly horizontally as the central plume was ascending vertically, so that the eplosion cloud appeared to epand equally in all directions. The surges decelerated rapidly 1-2 km from the dome, then lifted

7 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 287 Fig. 4. Photographs of the eplosion of 12:5 on 7 August The plume from this eplosion ultimately rose to about 13 km. Times after the onset of the seismic signal: (a) 27 s, (b) 36 s, (c) 42 s, (d) 48s, (e) 58s, (f) 76s. The collapsing fountain is visible to the left in (a) and to the right in (b) and (c). Fountain collapse generated pyroclastic surges and flows in all the major drainages around the volcano. The three individual plumes shown in Fig. 15 and discussed in the tet are numbered 1 to 3; g, Gages Mountain; sg, St George's Hill. For scale, the top of the plume in (e) is about 3 km above sea level and 2 km above the lava dome. Photographs taken from NW of the volcano by K. West. off the ground to form buoyant ash plumes (Fig. 3d). Shortly thereafter, highly concentrated pumice-and-ash pyroclastic flows were observed advancing at about 1ms -1 down one or more valleys around the dome as thin (.5-1 m), granular avalanches with associated billowing clouds of elutriated ash. It is surmised that the pyroclastic flows formed by rapid fallout of debris from the collapsing fountain and initial pyroclastic surges. In most of the eplosions, pyroclastic flows occurred in all major ghauts around the dome. Some eplosions in September and October were angled to the north, probably because the horseshoe-shaped dome summit crater opened in that direction (Fig. 2). Runout distances varied between eplosions, but were typically 3-6 km, with the flows taking up to a few hundred seconds to reach their distal limits. Only one eplosion (6:35 on 16 October) did not generate pyroclastic flows, and another generated only very small ones (5:47 on 15 October). Fountain collapse was clearly visible at night. First, a brightly incandescent cloud was seen rising over the dome. Moments later, a ring of coarse, incandescent debris fell back from height along steep, outwardly inclined trajectories onto the slopes surrounding the dome. Fountain collapse was short-lived, no more than 1 to 2 seconds. Incandescence in the initial pyroclastic surges disappeared rapidly over a few seconds as the surges entrained air and cooled. The central eplosion plumes had fast-rising bulbous heads and narrow central stems (Figs 4, 5 and 6) and reached heights of 3-15 km, with an average of about 1 km (Figs 7 and 8). The height

8 288 T. H. DRUITT ET AL. Fig. 4. (continued) estimates (probably ±1%) were supplied by the US National Oceanic and Atmospheric Administration (NOAA) Satellite Analysis Branch in Washington DC, based on ash-cloud movements tied to radiosonde wind data from Puerto Rico and Guadaloupe. Additional ground-based height estimates made on Montserrat using an Abney level were in broad agreement with the NOAA heights. For three eplosions, height estimates were obtained using plume-top temperatures from GOES satellite images (Bonadonna et al. 22b; Table 1). Upon attaining their maimum altitude, which typically took about 1 minutes, the plume heads spread out to form umbrella clouds (Fig. 9), which then detached from their stems and were carried to the north or NW by high-level (8-18 km) winds. Satellite images of one such cloud (14:56 on 26 September) are presented by Bonadonna et al (22/?). Some plumes were richer in steam and thus paler in colour than others and tended not to rise as high. The weaker plumes also had more poorly developed umbrellas, and were more liable to be blown off-centre by the wind. Ash clouds generated by the lofting of the pyroclastic surges and by elutriation from pyroclastic flows were gradually drawn up and incorporated into the central plume by inward-moving currents of air (Fig. 3d). After several minutes, each eplosion settled into a phase of waning, relatively low-intensity discharge, generating a low, bentover plume transported mainly to the west or NW on low-level trade winds (below about 5 km). This decoupling of the high umbrella (north or NE) and low, waning plume (west or NW) was characteristic of many of the eplosions. The waning plume then

9 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 289 Fig. 4. (continued) decreased slowly in height over a period of 1-3 hours. Venting during the waning phase was often pulsatory on a time-scale of 1-2 minutes. Lightning occurred during many of the eplosions, and was particularly evident at night. It appeared in the eruption column a few seconds after the initial eplosion and continued for up to 1 minutes. Cloud-to-cloud strikes accompanied by thunderclaps were most common. Despite the presence of condensed steam in some of the plumes, it is not believed that eternal water played any significant role in the eplosions. The groundwater system of Soufriere Hills Volcano had essentially dried out by mid-1997 after 2 months of lava etrusion, and there was no evidence of a significant source of groundwater. The eplosion of 12:5 on 7 August, 1997 This eplosion was studied in particular detail from video footage. The events and their timing are listed in Table 2, which is eploited later to compare with calculations. The eplosion was filmed by a time-lapse video camera mounted at MVO South (frame interval 1.4 s), two ordinary video cameras at MVO South and at Fleming, and an ordinary video camera aboard the MVO helicopter. Still photographs were taken roughly every 2 s from MVO South until 45 s into the eplosion. Another set of images taken by K. West is shown in Figure 4. Correlation of all image sets permitted detailed reconstruction of the eplosion. The timer on the Fleming video camera had been previously correlated with GPS time to ±.5s

10 Fig. 5. Eplosion at 14.2 on 6 August 1997, showing the typical development of the buoyant plume, which rose to between 9 and 12 km above sea level. Partial fountain collapse during the initial stages of the eplosion sent pyroclastic flows and surges down valleys to the north (left) and west (right). either side of Gages Mountain (g). Photographs taken from MVO South by T. H. Druitt.

11 Fig. 6. Eplosion at 15:13 on 2 October Fountain collapse generated pyroclastic surges and flows visible to the west (right) and north (left) of Gages Mountain (g). Ash was thrown up by the ground impact of ballistic blocks (b). The buoyant plume ultimately rose to about 1km. Note the buildings for scale in the foreground. Photographs taken from the NW by P. Cole.

12 292 T. H. DRUITT ET AL. Fig. 7. Histograms of plume height and eplosion repeat interval for the 1997 Vulcanian eplosions. Fig. 8. Variations of (a) plume height and (b) eplosion repeat interval with time during the episode from 22 September to 21 October The repeat interval is that which preceded the eplosion concerned, which is why none is reported for the first eplosion. The horizontal ais shows the eplosion number during this period (see Table 1). An absence of bars indicates a lack of data. In mid-october there occurred a series of relatively weak eplosions with short repeat intervals. allowing correlation of images with the broadband seismic signal measured by the Galway's Estate seismometer (Fig. 1). Time zero in Table 2 is taken as the onset of the seismic signal (as received at the Galway's Estate seismometer) at 12:4:44. Emergence of the first jet above the crater rim, as estimated by backetrapolation of height-time curves (see below), followed 1 s later. Given that the velocity of long-period seismic waves at Montserrat is 18-19ms- 1 (Neuberg et al. 1998). the travel time for the

13 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 293 Fig. 9. (a, b) Development of the umbrella cloud during a typical eplosion (9:46 on 27 September 1997). The plume height in (b) is about 1.8 km above sea level. Photograph taken from a boat NE of Montserrat by B. Poyer. The boat was retreating from Montserrat, so that (a) was taken from closer than (b). seismic signal from the dome to the seismometer was about 1.5 s. About 2.5 s therefore separated the onset of the seismic signal and emergence of the first jet above the crater rim. Given the vertical eit velocity of this jet (8 ± 1ms -1, see below), this suggests a depth of about 2m for the summit crater at the time of the eplosion, which is compatible with the observed crater diameter ( ± 2 m) and a typical angle of rock stability. The eplosion generated at least three separate plumes (numbered 1 to 3) visible from MVO South and Fleming that merged, after about 1s, into a single, large plume (Fig. 4). There may have been other plumes not visible from these observation points. Plume 1 emerged first over the southern or southwestern flank of the dome, rapidly followed by plume 2 to the north (Fig. 4a). Each decelerated as it rose, part collapsing back to the ground, and part ascending buoyantly. The fallback height of the two plumes was estimated visually as a few hundred metres above the crater rim and the collapsing material first hit the ground behind Gages Mountain about 18s into the eplosion (Table 2). The resulting pyroclastic surge was first visible behind Gages Mountain at 22.8 s. As this surge travelled out from the dome, a thin veil of ash was thrown up all over Gages Mountain, either due to seismic shaking or to a blast of preceding air. The surge then ramped over the north face of Gages Mountain before decelerating abruptly, ceasing forward motion, and lifting buoyantly off the ground at about 45 s (Fig. 4b, c and d). Collapse over the northern flanks lagged a few seconds behind that over Gages. The northern fallback curtain was first observed at 19.1 s and it hit the ground three seconds later (Fig. 4a). The resulting surge was first visible at 27.8 s advancing at about 45 m s -1 down the headwaters of Mosquito Ghaut. Large blocks thrown northwards ahead of plume 2 followed ballistic trajectories. The first blocks were seen to hit the ground at 21.6s, throwing up ash. Impacts then migrated northwards away from the dome, reaching the maimum range of 1.6 km about 5 s later (26.9 s). At 27.9 s, as plumes 1 and 2 became buoyant, plume 3 broke out at high speed at the top of the column (Fig. 4b). It rapidly decelerated and began to rise buoyantly, before separating into two (Fig. 4c and d). Plume 3 remained momentum-driven up to 12m above the crater rim. After a few tens of seconds, thin pumice-and-ash pyroclastic flows were observed travelling slowly down Tuitt's Ghaut, Mosquito Ghaut and the Tar River valley, reaching their maimum limits of 3-6 km about 2 s after the onset of the eplosion. The central plume rose to between 12.2 and 13.7km according to NOAA data, forming a large umbrella. The waning phase of the eplosion lasted an hour. Eplosion products Fallout tephra from the eplosions had three sources (Bonadonna et al. 22b). (1) Pumiceous (and minor lithic) blocks, lapilli and ash from the main central plumes and umbrella clouds. Pumice clasts as large as 1cm in mean diameter fell on St George's Hill, 6.5cm on the South Soufriere Hills, 6.5cm on northern Plymouth, 4.5cm on Windy Hill and at Cork Hill, 4cm at MVO South and 2 cm in northern Montserrat. (2) Ash from plumes generated by the lofting of pyroclastic surges and by elutriation from pyroclastic flows (termed co-pyroclastic-flow plumes). This was mostly drawn

14 294 T. H. DRUITT ET AL. Table 2. The 12:5 eplosion of 7 August 1997 Time (s) Event Start of eplosion seismic signal (phase 1)* 1 Emergence of eplosion jet 1 at 95 ± 1ms -1 7 Emergence of eplosion jet 2 at 95 ± 1 ms Emergence of eplosion jet 3 at > 13ms Fallout visible behind Gages Mountain from MVO South 18 Start of seismic signal from fountain collapse and pyroclastic flows (phase 2) 19.1 Fallout curtain descending over the north flank 21.6 First ballistics hit Farrell's Plain, 1.2km north of the vent 22. First ballistics hit Paradise Plain, 1.2km north of the vent 22.2 Collapsing fountain hits the north flank 22.8 Pyroclastic surge visible behind Gages Mountain 26.9 Ballistics reach maimum range on Paradise Plain, 1.6km north of the vent 27.8 Pyroclastic surge in Mosquito Ghaut, 1.7km from source, travelling at c. 45ms Jet 3 arrives at the top of the plume 34.3 Pyroclastic surge passes Gages soufriere on the west flank 45 Pyroclastic surge ramps over Gage's Mountain and lofts 58 Pyroclastic surge reaches maimum runout on the Farrell's Plain and begins to loft 7 Drop in intensity of the phase 2 seismic signal 18 Pyroclastic flows reach the foot of St George's Hill on the west flank 167 Pyroclastic flow reaches the Paradise River, 3.5km from source, at 1ms Pyroclastic flow level with Harris, 3.4km from source, at 9ms Pyroclastic flow reaches sea on Tar River delta, 3.3 km from source, at 13-25ms -1 End of pyroclastic flow seismic signal; continuing tremor (phase 3) c. 36 End of the eplosive eruption *The seismic signal was measured at the Galway's Estate station (Fig. 1). The time for seismic waves to reach this station from the dome was about 1.5s, so emergence of jet 1 occurred about 2.5s after the onset of the eplosion seismic activity. well defined levees and snouts rich in relatively low-density pumice boulders. Temperature measurements made approimately 2 hours after flow emplacement ranged from 18 to 22 C. This is consistent with the night video footage of the eplosions, which shows that the discharging material lost heat very rapidly during fountain collapse, presumably by entrainment of air, as also seen in numerical models of the eplosions (Clarke et al. 22). Vesicularities of clasts (>3cm) from the eplosion deposits were calculated from density measurements made by the waterimmersion technique. Vesicularities of pumice clasts from both fallout and pyroclastic flows (from all but the first three eplosions in August) ranged from 55 to 75vol% (26 clasts). In contrast, ballistic blocks discharged during the initial vent-clearing phase of each eplosion were mostly dense, with Vesicularities much lower than 55%. Dense clasts also occurred in the pyroclastic flows, but only up to 5% by volume (Cole et al. 22; Clarke et al 22). These are interpreted as fragments of crater rubble, degassed magma plugging the conduit prior to each eplosion, or material picked up from the ground. No systematic differences in pumice vesicularity were observed between different eplosions, with the eception of the first three of the August series, in which the Vesicularities of fallout clasts ranged from to 75 vol%. The ejection of important quantities of dense andesite as fallout could be due to reaming out of the August summit crater by these early eplosions. It is not known if a similar process took place in September. A notable feature of the fallout pumices was the high abundance of tabular clasts with planar or curviplanar surfaces and sharp edges (Fig. 12). Many pumices were lenticular in cross-section, with one flat face and a conve face on the other side. Between 2 and 5% of fallout pumices a few centimetres in diameter had tabular shapes. Many had been slightly deformed and rounded by minor post-fragmentation inflation, showing that their angular shape was not due to ground impact. Neither were they associated on the ground with other pieces of the same block, as epected if impact breakage had occurred. The shapes of the clasts are attributed to brittle fragmentation of an already vesicular magmatic foam with at least 55 vol% bubbles. In contrast, pumice blocks in pyroclastic flows from the eplosions were typically rounded and subequant due to abrasion during fountain collapse and transport. up and mied into the central plumes; however, where observed separately, fallout from this source was often sporadic due to generation from surges and flows in different sectors around the volcano. (3) Ash from the low, waning plume. Ecept within a couple of kilometres from the lava dome, the total fallout from individual eplosions seldom eceeded just a scattering of clasts or a layer of ash no more than a few millimetres thick. Fallout distributions were complicated by miing of material from the three sources and by variable wind directions, wind intensities and plume heights. Western Montserrat was affected by fallout from all three sources, whereas fallout from the umbrellas dominated in the north. A map of the areas impacted by the fountain-collapse pyroclastic flows and surges is given in Figure 1. Pyroclastic flows travelled down the Tar River and White River valleys as far as the sea, and down Fort Ghaut to within a few hundred metres of the shoreline. Pyroclastic flows discharged to the north by the August eplosions travelled down Tuitt's and Mosquito Ghauts, then on as far as the village of Farm (6 km). Those in September and October did the same, but also entered White's Ghaut and spread out over the surface of the 21 September dome-collapse block-and-ash flow deposit (Cole et al. 22), locally reaching the sea. Some pyroclastic flows in September and October were channelled preferentially down Tuitt's Ghaut because the eplosions were angled to the north. Many eplosions sent flows down Tyre's Ghaut, then into Dyer's River valley. The deposits from the pyroclastic flows are described by Cole et al. (22). They consisted of numerous anastomosing lobes with multiple breakouts (Fig. 11). The distal ends of individual lobes were typically 1-5m wide and.5-1 m thick (Fig. 11c,d), with Cyclic patterns of edifice deformation, seismicity and eplosion Cycles of edifice deformation and hybrid seismicity were closely associated with the August eplosive activity (Voight et al. 1998, 1999). A typical cycle consisted of slow inflation of the dome followed by rapid deflation and eplosion, giving a saw-tooth pattern on tiltmeter records (Fig. 13a). Seismic swarms began up to several hours before each eplosion and culminated near the tilt peak or during the deflation phase at between one and four events per minute (Fig. 13b, c). Deformation records ceased during the eplosion of 16:57 on 5 August, when the only operating tiltmeter (Fig. 1) was destroyed, but the cycles continued to be evident from the seismic record. All but one of the August eplosions occurred shortly before or shortly after the peak in seismicity. The close correlation between seismicity and eruption onset permitted accurate prediction of the eplosions during this period. Repeat intervals between the August eplosions ranged from 1 to 14 hours, ecept for the last two of the series, which were preceded by intervals of 63 and 23 hours respectively (Table 1). In contrast, there was much less precursory seismicity before each of the September and October eplosions. Precursor seismic swarms occurred on 14 occasions, although only a few included periods of intense seismicity such as those preceding most of the August eplosions. When a swarm did occur, it tended to continue for a short time after the eplosion. On no occasion did the precursor seismicity develop in such a way as to enable accurate forecasting of eplosions as in August. There was also no tiltmeter functioning at the time, so there are no tilt data from this period.

15 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 295 Fig. 1. Map of the pyroclastic flow and pyroclastic surge deposits from the 1997 eplosions. The locations of the seismometers and tiltmeter operational at the time are also shown. The tiltmeter was destroyed at 16:57 on 5 August and was not replaced until after the eplosions had finished. Eplosion forecasting in September and October relied on the observed periodicity of the eplosions themselves, which ranged from 2.5 to 33.5 hours, with an average of 9.5 hours (Figs 8 and 9). Seismic signals from the eplosions Each eplosion generated a seismic signal captured by the MVO broadband system. A typical eample (16:57 eplosion on 5 August 1997) is shown in Figure 14a, with a blow-up of the first 1.4 min in Figure 14b. Each signal began abruptly and remained at a high level for several minutes, then decayed over a period of a few tens of minutes to 3 hours. Selected signal parameters are listed in Table 3. There were three discrete phases to each signal: (1) a long-period part, typically of 1-2 s duration; (2) a higher amplitude pyroclastic flow signal lasting a few minutes; and (3) harmonic tremor lasting 1-3 hours (average 8 minutes) during the long period of waning discharge that terminated each eplosion. Phases 1 and 3 had similar frequency spectra, with the main energy between.6 and 1.7 Hz (Fig. 14c). Phase 2 contained much more energy distributed over a range of higher frequencies, principally 2 to 2 Hz. Filtering of the raw signals enabled us to separate the signals into low-frequency and high-frequency components (Fig. 14a and b). We applied a time-domain recursive Butterworth filter between.5 and 1 Hz and a high-pass filter at 2 Hz. The low-frequency component was present throughout all three phases of the seismic signal, whereas the high-frequency component occurred only during phase 2 when it was superimposed on the low-frequency one. The low-frequency component is interpreted as the vibrational response of the magmatic conduit to the eplosion itself (Neuberg & O'Gorman 22). During most eplosions, it remained at a high level for about 45-7 s before falling to a much lower level, which agrees with visual observations for the duration of peak discharge. The low-frequency component fluctuated in intensity with time, as visible on Figure 14b for the eplosion at 16:57 on 5 August. This is typical of the amplitude modulation of a longperiod seismic signal (Neuberg & O'Gorman 22), and is not thought to be due to eruption unsteadiness. Video analysis of the 12:5 eplosion of 7 August revealed no obvious correlation between eruptive intensity and the amplitude of the low-frequency seismic component. The high-frequency component of phase 2 was due to a combination of ballistic impact, fountain collapse and pyroclastic flow, and had a frequency spectrum typical of pyroclastic flow signals, for eample at Montserrat (Miller et al. 1998) and Mount Unzen, Japan (Uhira et al. 1994). Its onset coincided with the first impact of the

16 296 T. H. DRUITT ET AL. Fig. 11. Pyroclastic flow deposits from the 1997 Vulcanian eplosion., (a) Buff-coloured pumice-and-ash pyroclastic flow deposits from the August eplosive period overlying grey block-and-ash flow deposits of the 25 June 1997 dome collapse. View looking NNE down Tuitts Ghaut and Pea Ghaut. Paradise River valley joins Tuitt's Ghaut from the left. (b) Buff-coloured pumice-and-ash pyroclastic flow deposits from August 1997 in White River valley. The dome lies hidden in cloud to the right. Road for scale. (c) Pumice-rich snout of a pyroclastic flow lobe from the September/October 1997 eplosions, near Spanish Point. The lobe is about 1 m high and about 12m across. It overlies block-and-ash flow deposits of the 21 September 1997 dome collapse. (d) Pyroclastic flow lobes from the September/October 1997 eplosions overlying block-and-ash flow deposits of the 21 September 1997 dome collapse. Houses of the community of Spanish Point on the left. The lobes have well defined levees and snouts rich in coarse pumice boulders. collapsing fountain and ballistic blocks with the ground. This is confirmed by analysis of the 12:5 eplosion of 7 August, in which phase 1 lasted 18s, and the end of phase 1 corresponded approimately with the collapsing fountain hitting the flank of the volcano behind Gages Mountain (Table 2). Ballistics were first observed to hit the north flank of the volcano 21.6 s into this eplosion and the collapsing fountain touched down.6 s later. The 1-2 s duration of seismic phase 1 is therefore the transit time for fountain collapse, i.e. the time for the momentum-dominated eruption jets and ballistic blocks to reach their maimum height, then fall to the ground. The duration of the high-frequency signal (i.e. of phase 2) corresponded with the time necessary for the pyroclastic flows to reach their distal limits (2-5 s). The abrupt drop in intensity of the phase 2 signal about 55 s after the eplosion onset (arrow, Fig. 14a) is believed to be due to pyroclastic flows nearest the seismometer concerned ceasing movement, while those in other valleys further away were still in motion, resulting in a drop in apparent seismic energy production.

17 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 297 Fig. 12. Fallout pumices from the August eplosions. The fragments are tabular with angular edges and curviplanar surfaces, and were formed by brittle fragmentation of a pressurized magmatic foam resident in the conduit prior to each eplosion. The scale is in centimetres. Fig. 13. (a) Tiltmeter and (b, c) seismic data for the period 3 July to 14 August The tiltmeter (Fig. 1) was destroyed by the eplosion at 16:57 on 5 August. Cycles of slow inflation of the dome, followed by rapid deflation, are evident from the tilt data. Seismic data are from the St George's Hill seismometer (Fig. 1) and show cyclic variations in total seismic amplitude (RSAM) and number of triggers per 1 minutes that are in phase with the tilt cycles. Eplosions are marked by the letter E and the dome collapse of 18:1 on 3 August by the letter C. The events marked P were the two relatively weak initial eplosions of 4 August (Table 1). Eruptive volumes We now estimate the volumes of magma discharged during the eplosions and the partitioning of material between fallout tephra and pyroclastic flows. The fallout includes that from (1) the main, central plume and umbrella, (2) the co-pyroclastic-flow plumes, and (3) the low, waning plume. We use the following mean densities based on field and laboratory measurements: airfall ash 11 kgm -3, pumice-and-ash flow deposit 135 kgm -3, and dense juvenile andesite26kgm -3. The uncompacted volume of fallout tephra from the months of August, September and October has been estimated by Bonadonna et al (22b) at m 3. All but 1 6 m 3 of this ( m 3 or m 3 DRE) is attributable to the 88 Vulcanian eplosions. The estimate is based on etrapolating isopach data to infinity using an eponential decay model. It is thought to be a minimum estimate, at least 1% too low, because study of GOES satellite images shows that considerable quantities of very fine (2-2 um) ash in fact travel further than predicted by the eponential model (Bonadonna et al. 22b). Raising the figure by 1% and dividing

18 298 T. H. DRUITT ET AL. by 88 gives an average of at least m 3 DRE of fallout tephra per eplosion. Another estimate of fallout volumes can be obtained from plume height data (Table 1). The maimum ascent heights of the eplosion plumes ranged from 3 to 15km. Since the duration of peak discharge (c. 5 s) was an order of magnitude shorter than the time for the plumes to reach their maimum altitude (c. 5 s), the plumes can be treated as discrete thermals of a given initial mass. This agrees with visual observations that the plumes developed the bulbous heads characteristic of thermals (Figs 4, 5 and 6). The ascent height of a volcanic thermal in the atmosphere can be epressed approimately by: H=1.89( Mc[T- To]).25 (1) where M is the mass of solids, T is magma temperature (about 11K; Barclay et al. 1998), T is the atmospheric temperature at vent level (about K), c is the specific heat of the solids (typically 11Jkg -1 K -1 ), and is the fraction of particles contributing to the thermal mass of the plume (Morton et al. 1956; Woods & Kienle 1994). This relationship has been shown to be appropriate for fine-grained ash clouds with a major part of their ascent below the tropopause (Woods & Kienle 1994). On the timescale (t) of plume ascent in the atmosphere, particles of radius r can attain thermal equilibrium with the gas phase only if: r < (2) (Woods 1995) where K is the thermal diffusivity of the particles (c. 1-6 m 2 s -1 ). The Montserrat plumes took on the order of 5s to reach their maimum heights, so only particles smaller than 1 cm or so attained thermal equilibrium with the gas during plume ascent. These typically constitute about 8% of particles discharged during eplosive eruptions (e.g. Druitt 1992; Woods & Bursik 1991), so is taken as.8. Equation 1 yields individual plume masses ranging over three orders of magnitude, equivalent to m 3 DRE of magma, reflecting the wide range of plume heights. The heights used were the mean values of the ranges given in Table 1. They were corrected by first subtracting the height of the dome during the eplosions (c. 1m), since the buoyant thermals formed above the dome, not at sea level. The average DRE volume calculated in this way was m 3, which is over three times that estimated from the field measurements given above ( m 3 ). One reason for the discrepancy may be the parameter, which is poorly constrained. Another reason may lie in the plume height estimates. The calculated volumes are a strong function of plume height due to the one-quarter-power dependence in Equation 1. Independent estimates of plume heights for three eplosions made from GOES images (Bonadonna et al. 22b) yield lower values than those provided by NOAA (Table 1), the differences ranging from 1 to 3%. If we reduce all the plume height estimates by 2% and reapply equation 1 while retaining.8, we obtain a range Fig. 14. (a) Seismic signal from the 16:57 eplosion of 5 August 1997, recorded on the seismograph at Galway's Estate (Fig. 1). The unfiltered signal shows phase 1, phase 2 and the first minute of phase 3. The abrupt drop in intensity of the phase 2 signal (arrow) may be due to pyroclastic flows nearest the seismometer coming to rest. The low-frequency (.5-1 Hz) filtered component is attributed largely to fragmentation and conduit flow. The high-frequency (>2 Hz) component is due to fountain collapse, ballistic impact, and pyroclastic flow. (b) Enlargement of the first 1.4 minutes of the same signal. Phase 1 is interpreted as the time for the ballistic blocks and collapsing fountain to first hit the ground. Pulsing of the.5-1 Hz component is attributed to resonance of seismic waves in the conduit. (c) Frequency spectra for the different phases of the eplosion signal, as well as for the background seismicity. Two spectra are shown for phase 2: one prior to the intensity drop (arrow in (a)) and one after it. The lowfrequency component of the eplosion signal is present throughout the eplosion, whereas the high-frequency component occurs only in phase 2.

19 Table 3. Characteristics of the eplosion seismic signals EPISODES OF CYCLIC EPLOSIVE ACTIVITY 299 Date Time (local) Duration phase 1 (s) Ma. vertical velocity eplosion (nms -1 )* Duration phase 2 (s) Ma vertical velocity py flows (nms-')t Date Time (local) Duration phase 1 (s) Ma. vertical velocity eplosion (nms- 1 )* Duration phase 2 (s) Ma. vertical velocity py flows (nms- 1 )t 4 Aug. 4 Aug. 5 Aug. 5 Aug. 6 Aug. 6 Aug. 7 Aug. 7 Aug. 7 Aug. 8 Aug. 8 Aug. 1 1 Aug. 12 Aug. 22 Sep. 22 Sep. 22 Sep. 23 Sep. 24 Sep. 24 Sep. 24 Sep. 25 Sep. 25 Sep. 25 Sep. 26 Sep. 26 Sep. 27 Sep. 27 Sep. 27 Sep. 28 Sep. 28 Sep. 28 Sep. 29 Sep. 29 Sep. 29 Sep. 29 Sep. 3 Sep. 3 Sep. 1 Oct. 1 Oct. 1 Oct. 2 Oct. 2 Oct. 2 Oct. 4 Oct. 6:3 16:43 4:45 16:57 4:2 14:36 :34 12:5 21:55 1:32 2:51 11:38 1:12 :57 1:45 2:42 7:23 :34 1:54 17:16 3:54 11:9 2:5 4:25 14:56 :1 9:46 17:15 4:28 1:34 23:3 6:26 11:23 16:48 21:57 4:43 17:44 5: 11:34 17:4 1:5 12:53 22:5 8: Oct. 5 Oct. 5 Oct. 5 Oct. 6 Oct. 6 Oct. 6 Oct. 7 Oct. 7 Oct. 8 Oct. 8 Oct. 9 Oct. 9 Oct. 1 Oct. 1 Oct. 1 1 Oct. 12 Oct. 12 Oct. 13 Oct. 13 Oct. 14 Oct. 14 Oct. 14 Oct. 15 Oct. 15 Oct. 15 Oct. 15 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 16 Oct. 17 Oct. 17 Oct. 17 Oct. 17 Oct. 18 Oct. 18 Oct. 19 Oct. 19 Oct. 2 Oct. 2 Oct. 21 Oct. 21 Oct. 18:27 2:53 1:41 18:41 2:44 1:42 17:5 4:6 16:2 3:47 15:1 3:3 12:32 4:13 18:4 17:57 7:55 22:24 9:32 15:24 1:36 13:48 23:16 5:47 8:33 14:5 22:2 2:51 6:35 9:44 14:2 18:48 4:1 12:35 16:5 23:18 6:48 15:17 5:13 21:27 5:4 15:13 11:39 19: * Maimum vertical component of the velocity spectrum for the eplosion component of the seismogram. Values for August were measured from the seismometer at Galway's Estate. Those for September and October were measured on the Windy Hill seismometer. The two data sets are therefore not directly comparable. t Maimum vertical component of the velocity spectrum for the pyroclastic flow component of the seismogram, measured at Windy Hill. of DRE volumes from.1 to m 3, with an average of m 3, which is more consistent with the field estimate. The total volume of pumice-and-ash pyroclastic flow deposits generated during the two episodes of Vulcanian eplosions was m 3, equivalent to an average of m 3 DRE per eplosion. This was calculated from volume surveys of the main ghauts carried out throughout the Soufriere Hills eruption. The method involved surveying of the ground surface by helicopter using rangefinder binoculars and GPS (Sparks et al. 1998). Only for one eplosion (15:17 on 18 October) was a detailed survey carried out of the pyroclastic flow deposits from a single event in enough detail to calculate a volume. The resulting map is given in Cole et al. (22). This eplosion was average in magnitude (NOAAestimated plume height 9.1 km), and the calculated pyroclastic flow volume ( m 3 DRE) agrees broadly with the overall average given above. The variation of pyroclastic flow volume with eplosion magnitude cannot be determined. There are indications from the data in Table 1 that flows from larger eplosions (as indicated by higher plumes) entered more ghauts around the dome, and thus were perhaps more voluminous, although this is not possible to quantify owing to the incompleteness of the observations. On the other hand, there is no obvious correlation between pyroclastic flow runout and plume height. For eample, flows from the relatively small eplosions at 11:34 on 1 October (plume height 4.6 km) and at 17:57 on 11 October (plume height 6.9 km) had runouts down Tuitt's Ghaut of 4.5 and 5.5 km respectively, which is comparable to, or greater than, those from eplosions with higher plumes. Neither is there any

20 T. H. DRUITT ET AL. correlation between plume height and the amplitude or duration of the high-frequency (pyroclastic flow) component of the eplosion seismic signals that might suggest systematic variation of pyroclastic flow volume with eplosion magnitude. We conclude that the average Vulcanian eplosion during August, September and October 1997 discharged a total of about m 3 DRE of magma m 3 as fallout and m 3 as pyroclastic flows - although there was a large variation from the smallest to the largest eplosions. Since the volume of ash in the co-pyroclastic-flow plumes was c. 1% of the total fallout (Bonaonna et al. 22b), the partitioning of magma during an average eplosion is estimated to have been as follows: central plume, umbrella cloud and waning plume m 3, pyroclastic flows m 3, and co-pyroclastic-flow ash plumes m 3. About two-thirds of the material ejected during an average eplosion underwent fountain collapse to form pyroclastic flows. Eit velocities during the eplosions Eit velocities during the eplosions (the velocity at which the material left the dome summit crater) were estimated from analysis of video footage, distributions of ballistic blocks and, more crudely, from the observed fountain-collapse heights and durations of phase 1 of the eplosion seismic signals. Observed fountain-collapse height and duration of seismic phase 1 A crude estimate of eit velocity (u) at the start of each eplosion can be made from the estimated fountain collapse heights (h = -65 m above the summit crater rim) during the initial 1-2s of the eplosions. Numerical modelling has shown that fountain-collapse height can be approimated by h u 2 /2g (Dobran et al. 1993; Clarke et al. 22), which implies a vertical eit velocity at crater-rim level of 8-115m s -1. For such velocities, the transit time (t) of the collapsing fountain from initiation to ground impact would be approimately t 2u/g, or s, which is consistent with the observed durations of phase 1 of the eplosion seismic signals (1-2 s). Analysis of video footage More accurate estimates of vertical ascent velocities were made for the three main plumes of the eplosion at 12:5 on 7 August, using the northern crater rim (94m) as reference level. Video footage from two sites (MVO South and Fleming) was analysed in order to construct height-time curves and thus to estimate ascent velocities. The height of each of the three plumes was measured as a function of time and corrected for perspective effects using the equations of Sparks & Wilson (1982). Distortions were small because the plumes were filmed from distances of several kilometres and scaling corrections were approimately linear. The vertical and horizontal fields of view for the two cameras were determined by filming a known landscape using the same settings as during the eplosion. Tracings of the three plumes are shown in Figure 15, and the height-time data are shown in Figure 16a. Plume ascent velocities were calculated using best-fit polynomials to smooth the heighttime curves, then differentiating the polynomials (Fig. 16b). For all three plumes the velocity first decreased, reached a minimum in the range 15 to 3ms -1, then increased again. The velocity minimum is interpreted as the transition from momentum-driven to buoyancydriven behaviour and occurred at heights of m (plumes 1 and 2) and 12m (plume 3) above the crater rim. Eit velocities for the three plumes were estimated by backetrapolation of the velocity curves to crater-rim level. Plume 2 left the crater 7 s into the eplosion with a vertical velocity of 8 ± 1ms -1 ; but, since it was initially inclined at about 6 to the horizontal, the absolute eit velocity was about 95 ± 1ms -1. Fig. 15. Tracings of plumes 1 to 3 of the eplosion of 12:5 on 7 August (Fig. 4). The data were measured from video footage taken from Fleming and corrected for perspective effects using the equations of Sparks & Wilson (1982). Numbers are the time in seconds after the onset of the eplosion and refer to the overlying plume front. The tracing intervals are either 2 or 3s. The star marks the vent location. The data for plume 1 do not allow accurate etrapolation, but the velocity and emergence angle were about the same as for plume 2. Plume 3 left the vent vertically 17 s into the eplosion at a velocity of at least 13ms -1. Clarke et al. (22) have analysed the same video footage of the 12:5 eplosion on 7 August using a similar method, but without distinguishing the three individual plumes recognized here (i.e. by tracing the progress of the front of the entire plume). Their data are consisted with an initial eit velocity of 11±9ms -1 and a velocity-minimum height of about 45m above the crater rim. They also analysed footage of the 14:36 eplosion on 6 August, for which the eit velocity and velocity-minimum height were estimated to be 13±7ms -1 and 65m respectively. A more detailed analysis of video footage of three eplosions in October 1997 (17:5 on 6 October, 16:2 on 7 October and 12:32 on 9 October) has been carried out by Formenti & Druitt (in prep.). The analysis involved plotting height-time curves for individual finger jets and etrapolating velocities back to crater-rim level using a fluid dynamic model for a jet. The calculated eit velocities range from 4 to 14m s -1 and in all three eplosions increased with time as fragmentation progressed to deeper levels in the conduit. Thus the slowest jets emerged at the start of the eplosion and the fastest ones about 1s later. Higher eit velocities may have occurred subsequently, but any such jets were hidden by the billowing clouds of ash. Analysis of ballistic trajectories The locations and sizes of ballistic blocks also served to constrain eit velocities. Ballistic crater fields were mapped by helicopter following the August eplosions using onboard GPS, and block diameters were estimated to within ±1 cm (Fig. 17). No accurate measurements were made in September and October due to safety considerations, but observations indicate that they were similar. The August eplosions threw blocks out to 1.7 km from the crater centre. The distribution was approimately symmetrical around the dome. Blocks with the largest ranges had diameters of.7m (to the north)

21 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 31 Fig. 16. (a) Heights and (b) velocities as a function of time for the three plumes of the 12:5 eplosion on 7 August (Fig. 15). The lines simply connect the data points. and 1.2m (to the south). Blocks smaller than.4m are not shown in Figure 17, but became increasingly abundant nearer the dome, reflecting the stronger air drag eperienced by small blocks (Bower & Woods 1996). Two models (Self et al 198; Waitt et al 1995) have been used to estimate launch velocities for the ballistic blocks. In the Self et al. (198) model the maimum range for a block several decimetres in size is achieved for an optimum launch angle of about 35. Launch angles as low as 3 are observed on video footage, so the optimum angle was assumed in the calculations. Selected launch velocities and travel times calculated using the model are shown on Figure 17. These take into account the elevation difference between the crater rim and landing site. Blocks.7m large on Farrell's Plain require launch speeds of about 16ms -1 to reach their range of 1.6km, 54m below the crater rim. The 1.2m block launched over Galway's Wall requires a velocity of 135ms -1. The highest velocities (25 ms -1 ) are required by.4m blocks that landed 1.7km from the vent. However, it was unclear in the field whether these were fragments of larger blocks that had broken on impact, so the result is ambiguous. Launch velocities up to 16m s -1 are therefore required by the Self et al. (198) model to eplain the ballistic data, which is higher than plume eit velocities measured from video footage (up to 14ms -1 ). Calculated flight times for the ballistics, which range from about 17 to 27 s (Fig. 17), are, however, broadly consistent with observations of the 12:5 eplosion of 7 August, in which the first ballistic impacts on Farrell's Plain were observed at 21.6 s (Table 2). The same ballistic data have been modelled by Clarke et al. (22) using the numerical scheme of Waitt et al. (1995). This

22 32 T. H. DRUITT ET AL. Fig. 17. Sizes and ranges of ballistic blocks from the 13 eplosions in August The numbers are the launch velocities and flight times calculated using the model of Self el al. (198), taking into account air drag and the elevation differences between the impact sites and the crater rim. The calculations assume an optimum eruption angle of 35 2 to the horizontal. Locations shown by squares are south of the dome, dots between Tar River valley and White's Ghaut, upwardpointing triangles between White's Ghaut and Tuitt's Ghaut, and downward-pointing triangles between Tuitt's Ghaut and Mosquito Ghaut. Fig. 18. Summary of a single eplosive cycle in 1997: (a) immediately prior to eplosion onset; (b) during the eplosion; (c) during the interval between eplosions. See the tet for discussion. requires lower initial velocities for the same 35 launch angle as used above (< 125ms -1 as opposed to < 16ms -1 ), since the drag coefficients assumed in this model result in lower form drag than in the model of Self et al. (198). The result is in better agreement with the eit velocities observed. Discussion Eplosion cyclicity Activity of Soufriere Hills Volcano in 1997 involved a regime of cyclic eplosive behaviour, with an average interval of about 1 hours. Repetitive Vulcanian eplosions have been reported from other volcanoes. For eample, si eplosions occurred regularly at Mount Ngauruhoe, New Zealand, on 19 February 1975 at intervals of between.5 and 1 hour (Nairn & Self 1978). Over the period 12 to 14 June 1991, a sequence of four eplosions occurred at Mount Pinatubo, Philippines, at intervals ranging from 1 to 28 hours, culminating in the climatic eruption (Hoblitt et al 1996). At Volcan Galeras, Colombia, si eplosions took place at intervals of between 8 and 181 days in (Sti et al. 1997). Twenty-three eplosions occurred at Tokachi-dake, Hokkaido, between 16 December 1988 and 5 March 1989, an average of three to four days apart (Katsui et al. 199). Repetitive eplosive behaviour at Montserrat is attributed to the cyclic build-up and release of magmatic pressure beneath a rheologically stiffened plug of degassed magma at shallow levels in the conduit below the dome (Voight et al. 1998, 1999). A type of stickslip effect has been invoked to eplain cyclic conduit pressurization at Montserrat, resulting in cyclic deformation of the dome and surrounding terrain (Voight et al. 1999; Denlinger & Hoblitt 1999; Wylie et al. 1999). Hybrid earthquakes are attributed to hydrofracturing and associated gas flow in rock or crystal-rich magma at

23 Table 4. Estimates of physical parameters for the 1997 eplosions EPISODES OF CYCLIC EPLOSIVE ACTIVITY 33 Parameter Plume height (km) Ascent duration of plume (s) Total magma volume discharged (m 3 DRE) Duration of peak discharge (s) Eit velocity (ms -1 ) Fountain-collapse height (m) Mass fraction entering collapse fountain Initial velocity of pyroclastic surges (ms -1 ) Runout of pyroclastic flows (km) Runout duration of pyroclastic flows (s) Typical velocity of pyroclastic flows (ms -1 ) Fragmentation pressure (MPa) Velocity of fragmentation wave (ms -1 ) Conduit withdrawal depth (km) Duration of waning phase of eplosion (min) Eplosion interval (min) Magma ascent velocity between eplosions (ms -1 ) Value* c. 1(3 to 15) c c.2/ c to >2 c. 6 (2-19) c. 6 (1-3 7) >.2 * Average values, with ranges in brackets. the peak of each pressurization cycle (Neuberg et al. 1998; Voight et al. 1999). Synchronized tilt cycles and hybrid swarms during the August eplosive episode provided accurate indicators of the pressurization state of the system, enabling MVO volcanologists to anticipate many of the eplosions successfully and to reduce the threat to the population. They also facilitated study of the eplosions and their products. Subsequently, in September and October, when there was no tiltmeter and hybrid swarms were weak or absent, the strong periodicity of the eplosions themselves played this role. Eplosion mechanisms Figure 18 summarizes schematically the events throughout one cycle of 1997 eplosive activity at Soufriere Hills Volcano. Our best estimates of the physical parameters are given in Table 4. Eplosive eruption commenced when the conduit overpressure eceeded the strength of the cap of degassed crystal-rich magma. During the initial few seconds, crater rubble and fragments of disrupted, degassed plug were thrown out, forming ballistic showers. A fragmentation wave then descended the conduit into the region of pressurized magma, resulting in a rapid escalation of eit velocities from about 4 to 14ms -1. Eruption was highly unsteady, peak discharge lasting just a few tens of seconds with the highest intensity over the first 1-2 s. Each eplosion discharged on average about m 3 DRE of magma. Since the conduit diameter during the phase of the eruption is estimated at 25 to 3m from spine dimensions and the widths of early vents (Watts et al. 22), and is not believed to have varied greatly with time (Voight et al. 1999), the conduit drawdown during an average eplosion was about 5 m below the crater floor (which itself was about 8m above sea level). This is a DRE drawdown; the actual average drawdown of vesicular magma would have been greater, but not more than 1 km. The largest eplosions must have emptied the conduit to depths of 2km or more. Since peak discharge lasted a few tens of seconds, the velocity of the fragmentation wave down the conduit is constrained to have been of the order of l-5ms -1, although the initial value could have been greater. High-intensity eruption probably ceased once the wave reached a level in which the magma pressure was not sufficiently large to drive fragmentation. Ash then continued to be discharged for 1-3 hours, but at a greatly reduced rate. Numerical models of the plume dynamics (Clarke et al. 22) and conduit flow (Melnik & Sparks 22b) reproduce several key features of the eplosions, including their highly transient nature, peak eit velocities in the range 8-14 ms -1, and discharge durations and conduit drawdowns comparable to those observed. The models are based on the rapid decompression and epansion of Fig. 19. Eplosion magnitude (as measured by plume height) as a function of the time interval between eplosions: (a) interval preceding the eplosion, and (b) interval following the eplosion. See the tet for discussion. gas-rich, pressurized magma beneath a degassed plug, thus supporting this interpretation of the eruption dynamics. The eruption columns were partially unstable in all but one eplosion. On average about two-thirds of the erupted material collapsed back to form pyroclastic surges and flows, while the other third, including probably a large proportion of smaller particles, was carried up into the plume. However, these proportions may have varied considerably between individual eplosions. Fountain collapse occurred in the first 1-2s of each eplosion from a few hundred metres above the crater rim. Vertical velocity profiles in the plumes reveal velocity minima corresponding to the transition from momentum-driven to buoyancy-driven behaviour. This is analogous to the superbuoyant regime of sustained eruption columns, which is intermediate between fully stable (convective) and fully unstable (collapsed) regimes (Bursik & Woods 1991) and is seen in the eplosion simulation of Clarke et al. (22). Once each eplosion was over, magma rose in the conduit by viscous flow at a couple of centimetres per second or more (1-2 km in 1 hours). This eceeds the critical ascent speed of about l^cms -1 for amphibole breakdown and eplains the presence of hornblende phenocrysts lacking breakdown rims in the eplosion pumices (Devine et al. 1998b). In at least some cases the conduit was totally refilled prior to the net eplosion and a small dome appeared in the crater. Repressurization of the conduit then occurred until conditions were right for the net eplosion, although the eact mechanism is not well understood. Figure 19 shows that a weak correlation eists between plume height and the intervals between eplosions for both the August and September-October episodes. Positive correlations eist between (1) plume height and the interval prior to a given eplosion (Fig. 19a) and (2) plume height and the interval following a given eplosion (Fig. 19b). Correlation 1 would suggest that large eplosions result from long preceding intervals, perhaps allowing the build-up of larger magma pressures in the conduit. Correlation 2 is consistent with a scenario in which large eplosions drain the conduit to deeper levels, so that longer

24 34 T. H. DRUITT ET AL. intervals are then required to refill the conduit prior to the net eplosion. The data do not distinguish between these two mechanisms, although correlation 2 appears visually to be slightly better than correlation 1. Conduit pressurization and eplosive fragmentation Pressurization of the magmatic conduit at Montserrat is attributed to non-linear vertical pressure gradients caused by large viscosity variations that accompany esolution of water from magma (Sparks 1997; Massol & Jaupart 1999). Magma viscosity is a strong function of water content, particularly at low pressures (Hess & Dingwell 1996). The estimated viscosity of non-degassed magma at Montserrat is about 1 6 Pas and that of completely degassed magma about 1 14 Pas (Voight et al. 1999). This is believed to have generated large magma overpressures (magma pressure minus lithostatic pressure) at shallow levels in the conduit. Another effect is the development of high gas pore pressures in the ascending magma due to (1) viscous resistance to vesicle epansion, which increases as the liquid esolves gas (Massol & Jaupart 1999), and (2) growth of microlites in the degassed, undercooled liquid, which forces further gas into vesicles (Sti et al 1997; Sparks 1997). Tilt amplitudes and far-field deformation measurements at Montserrat are consistent with maimum magma overpressures of about ten to a few tens of megapascals a few hundred metres below the base of the dome (Shepherd et al 1998; Voight et al. 1999). The conduit flow modelling of Melnik & Sparks (22a) predicts steep pressure gradients and overpressures up to lmpa in the upper conduit. The angular, platy shapes of many of the 1997 fallout pumices with vol% vesicles are consistent with brittle fragmentation of a pressurized magmatic foam present in the upper conduit prior to each eplosion. Brittle fragmentation of magma requires steep pressure gradients and fast decompression rates in order to drive the magma through the glass transition limit (Dingwell 1996). This has been observed eperimentally by Alidibirov & Dingwell (1996), who showed that pressure differentials across the fragmentation interface of a few megapascals can be sufficient to drive brittle failure, generating platy fragments with shapes very much like those at Montserrat. Recent eperimental work has shown that the tensile strength of crystal-rich magma like that at Montserrat may be of the order of 2MPa or more (Martel et al 21). As the fragmentation wave descended the conduit during each eplosion, the pressurized foam broke up into tabular fragments that were then accelerated to the surface. Magma fragments erupting from the vent apparently had sufficiently high viscosities due to gas esolution to suppress post-fragmentation epansion, enabling pumices to retain vesicularities and angular shapes close to those acquired at fragmentation (the viscosity quench effect; Thomas et al 1994). Pumice incorporated into pyroclastic flows were subsequently rounded by abrasion during transport. The presence of magmatic foam with at least 55% bubbles in the upper conduit can be used to provide an independent estimate of magma pressure prior to each eplosion. The bulk water content of the magma prior to ascent was about 1.6±.3wt%, based on the water content of glass inclusions (4.3±.5wt%; Barclay et al 1998; Devine et al 1998a) and the estimated crystal content in the magma reservoir at 5-6km depth (6-65vol%; Murphy et al 2). In the Appendi we show that the total confining pressures required for magmatic foam with 1.6 ±.3 wt% water to have vesicularities of 55-75% are 5-15MPa. One key feature of the eplosions is that the products are dominantly pumiceous, with dense clasts making up no more than 5% of those erupted (Clarke et al 22). Given that an average eplosion emptied the conduit to about 5 m (DRE) depth, the plug of degassed magma present in the conduit prior to each eplosion can have been no more than about 25m thick, corresponding to an overburden of less than 1 MPa. Given that pressures of 5-15 MPa are required in the magmatic foam just below this cap, this suggests that the foam must have been very significantly overpressured (by at least a few megapascals) relative to the overlying plug and surrounding conduit walls. The presence of pressurized, gas-charged magma at very high levels in the conduit immediately prior to each eplosion is consistent with the observation that eit velocities in ecess of 1 m s -1 or more were achieved only a few seconds after each eplosion began (Clarke et al 22; Melnik & Sparks 22b). Initiation of episodes of eplosive activity on Montserrat Each of the two episodes of eplosive activity in 1997 was triggered by a major dome collapse (3 August and 21 September), as was the eplosive eruption on 17 September 1996 (Robertson et al 1998). In each case, sudden removal of part of the dome led to the conditions for eplosive fragmentation. This was not immediate, the delays being 2.5 hours (17 September 1996), 1 hours (3 August 1997) and 2 hours (22 September 1997), showing perhaps that time was necessary for the build-up of sufficient conduit pressure for this to occur. Many dome collapses occurred during the period, but only three are known to have triggered major vertical eplosions. We eclude here the relatively weak eplosions of late 1998 and 1999, which may have been triggered by slow pressure build-up in the slowly crystallizing lava dome and conduit during the period of virtually no magma etrusion (Norton et al 22). One important factor was probably that the 17 September 1996 and 21 September 1997 collapses resulted in two of the largest height reductions of the active dome-growth area during the period (13m and 23m respectively), causing large decompressions of the conduit (at least 3.5 and 6 MPa). The height reduction from the 3 August 1997 collapse was not observed clearly, but it is inferred to have been at least 11m (3 MPa) from the form of the crater observed four days later. Sudden decompression of at least 3 MPa therefore appears necessary to trigger eplosive fragmentation at Montserrat. Conduit flow beneath lava domes involves comple feedback effects and sudden decompressions can force systems from effusive to eplosive behaviour (Jaupart & Allegre 1991; Woods & Koyaguchi 1994). An additional effect in 1997 may have been the high magma discharge rate. The time-averaged magma discharge rate increased throughout the period, and by August 1997 had reached 7-8 m 3 s -1 (Sparks et al 1998; Sparks & Young 22). High discharge rates favour eplosive fragmentation by limiting the time available for magma degassing during ascent (Jaupart & Allegre 1991; Melnik & Sparks 22a). High magma flu during August, September and October of 1997 may have helped to prime the conduit for eplosive activity once a suitably large dome collapse occurred. Strangely, the largest dome collapse of the period (26 December 1997; Sparks et al 22) decompressed the conduit by at least 8 MPa. giving rise to a violent lateral blast and pyroclastic density current, but triggered no vertical eplosion from the conduit and produced little pumice. This highlights the compleity of the system and the eistence of important effects not considered here. Conclusions Two episodes of cyclic eplosive activity occurred at Soufriere Hills Volcano in Thirteen eplosions took place in August and another 75 in September and October. The activity had a major impact on southern Montserrat and triggered northward enlargement of the evacuation zone in mid-august. Like the eplosive eruption of 17 September both episodes in 1997 were preceded by major dome collapses that decompressed the conduit by 3 MPa or more. Delays of 3 to 2 hours then followed before eplosive activity commenced. Large gravitational collapses are a prerequisite for vertical eplosive eruption at Montserrat. The eplosions were highly unsteady, with the most intense phase lasting only a few tens of seconds. Peak discharge was accompanied by ballistic showers, eit velocities up to 14 m s -l, and (in all but one event) fountain collapse from a few hundred metres above the crater rim over the first 1-2s of each eplosion. Pyroclastic flows travelled up to 6km down all major drainages around the

25 EPISODES OF CYCLIC EPLOSIVE ACTIVITY 35 dome and entered the sea on the south and east coasts. Buoyant eruption plumes with large, bulbous heads rose to 3-15 km in the atmosphere, then spread out as umbrella clouds. After 1 minutes or so, each eplosion settled into a waning phase that typically lasted an hour and generated a low, bent-over ash plume. Fallout and pyroclastic flows/surges from the eplosions accounted on average for one-third and two-thirds of the magma discharged, respectively. The eplosions emptied the conduit to a depth of.5-2 km, perhaps more in some cases. Filtering of eplosion seismic signals permitted distinction of a low-frequency (c. 1 Hz) component due to the eplosion itself and a high-frequency (>2 Hz) component due to ballistic impact, fountain collapse and pyroclastic flow. Relative timing of the onsets of the two components provided information on the flight durations of ballistic blocks and on the transit time for fountain collapse, from inception to first ground impact. Eplosions in August were accompanied by cyclic patterns of seismicity and edifice deformation. Repeated slow inflation, followed by rapid deflation, of the volcano recorded cycles of build-up, then release, of pressure beneath the dome. The eplosions were driven by rapid decompression and brittle fragmentation of overpressured magmatic foam in the upper conduit and occurred at intervals of 2.5 to 63 hours, with a mean of 1 hours. Synchronized tilt cycles and hybrid earthquake swarms during the August eplosions provided accurate indicators of the pressurization state of the system, enabling volcanologists to anticipate many of the eplosions and reduce the threat to the population. In September and October, when there was no tiltmeter and hybrid swarms were weak or absent, the strong periodicity of the eplosions themselves played this role. We thank the staff of the M VO for their very important contributions in the study of the 1997 eplosions. D. Lea, M. Sagot and D. Williams kindly provided us with video footage of the eplosions and allowed us to study it. D. Williams helped us in the analysis of video footage. B. Poyer kindly provided the photographs in Figure 7. Careful reviews by T. Koyaguchi, L. Wilson and P. Kokelaar are gratefully acknowledged. Appendi Estimation of fragmentation pressures during the eplosions We estimate the pressure necessary for the magma with 6-65 vol% crystals (Murphy et al. 2) and a bulk water content of 1.6±.3wt% to have vol% vesicularity. Consider a unit volume of crystal-bearing magmatic foam immediately prior to fragmentation. The volume fraction of bubbles is and the volume fraction of crystals in the liquid phase is F. The masses of gas, liquid, and crystals are given by: M g = p g M 1 = A(1-A-)(1-F) M c = Pc(l - )F (Al) where M is mass, p is density and the subscripts g, 1 and c stand for gas, liquid and crystals respectively. Given the solubility law for water in magmatic liquid, n = P 1 / 2, where n is a mass fraction and a is approimately Pa 1/2 for rhyolite (the composition of interstitial glass in the pumices), we can write the mass balance equation for water in the foam M g + M 1 ap 1/2 = N(M g + M 1 + M c ) (A2) where N is the bulk mass fraction of water in the magma. The density of the gas is given by: ( A3 ) where T is temperature (about 86 C or 1133 K; Barclay et al. 1998) and r is the gas constant (462 J kg -1 K -1 for water). Given a bulk water content N, we can use Equation A2 to estimate the vesicularity of the foam as a function of pressure P. For a bulk water content of 1.6±.3wt%, bubble contents of vol% require pressures in the range 5-15MPa. References ALIDIBIROV, M. A. & DINGWELL, D. B Magma fragmentation by rapid decompression. Nature, 38, ASPINALL, W. P., MILLER, A. D., LYNCH, L. L., LATCHMAN, J. L., STEWART, R. C., WHITE, R. A. & POWER, J. A Soufriere Hills eruption, Montserrat : volcanic earthquake locations and fault plane solutions. Geophysical Research Letters, 25, BARCLAY, J., RUTHERFORD, M. J., CARROLL, M. R., MURPHY, M. D., DEVINE, J. D., GARDNER, J. & SPARKS, R. S. J Eperimental phase equilibria constraints on pre-emptive storage conditions of the Soufriere Hills magma. Geophysical Research Letters, 25, BONADONNA, C., MACEDONIO, G. & SPARKS, R. S. J. 22a. Numerical modelling of tephra fallout associated with dome collapses and Vulcanian eplosions: application to hazard assessment on Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, BONADONNA, C., MAYBERRY, G. C., CALDER, E. S., SPARKS, R. S. J., CHOU, C., JACKSON, P., LEJEUNE, A. M., LOUGHLIN, S. C., NORTON, G. E., ROSE, W. L, RYAN, G. & YOUNG, S. R. 22b. Tephra fallout in the eruption of Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, BOWER, S. M. & WOODS, A. W On the dispersal of clasts from volcanic craters during small eplosive eruptions. Journal of Volcanology and Geothermal Research, BURSIK, M. I. & WOODS, A. W Buoyant, superbuoyant and collapsing eruption columns. Journal of Volcanology and Geothermal Research, 45, CLARKE, A. B., NERI, A., VOIGHT, B., MACEDONIO, G. & DRUITT, T. H. 22. Computational modelling of the transient dynamics of the August 1997 Vulcanian eplosions at Soufriere Hills Volcano, Montserrat: influence of initial conduit conditions on near-vent pyroclastic dispersal. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, COLE, P. D., CALDER, E. S., SPARKS, R. S. J. ET AL. 22. Deposits from dome-collapse and fountain-collapse pyroclastic flows at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoir. DENLINGER, R. P. & HOBLITT, R. P Cyclic behavior of silicic volcanoes. Geology, 27, DEVINE, J. D., MURPHY, M. D., RUTHERFORD, M. J. ET AL. 1998a. Petrologic evidence for pre-eruptive pressure-temperature conditions, and recent heating, of andesitic magma erupting at the Soufriere Hills Volcano, MONTSERRAT, W. I. Geophysical Research Letters, 25, DEVINE, J. D., RUTHERFORD, M. J. & GARDNER, J. C. 1998b. Petrologic determination of ascent rates for the Soufriere Hills Volcano andesite magma. Geophysical Research Letters, 25, DINGWELL, D. B Volcanic dilemma: Flow or blow? Science, 273, DOBRAN, F., NERI, A. & MACEDONIO, G Numerical simulations of collapsing eruption columns. Journal of Geophysical Research, 98, DRUITT, T. H Emplacement of the 18 May 198 lateral blast deposit ENE of Mount St. Helens, Washington. Bulletin of Volcanology, 54, FAGENTS, S. A. & WILSON, L Eplosive volcanic eruptions - VII. The ranges of pyroclasts ejected in transient volcanic eplosions. Geophysical Journal International, 113, HESS, K. U. & DINGWELL, D. B Viscosities of hydrous leucogranite melts: a non-arrhenian model. American Mineralogist, 81, HOBLITT, R. P., WOLFE, E. W., SCOTT, W. E., COUCHMAN, M. R., PAL- LISTER, J. S. & JAVIER, D The pre-climactic eruptions of Mount Pinatubo, June In: NEWHALL, C. G. & PUNONGBAYAN, R. S. (eds) Fire and Mud: Eruptions and Lahars of Mount Pinatubo, Philippines. University of Washington Press, JAUPART, C. & ALLEGRE, C. J Gas content, eruption rate and instabilities of eruption regime in silicic volcanoes. Earth and Planetary Science Letters, 12,

26 36 T. H. DRUITT ET AL. KATSUI, Y., KAWACHI, S.. KONDO, Y. ET AL The eplosive eruption of Tokachi-dake, Central Hokkaido, its sequence and mode. Bulletin of the Volcanological Society of Japan, Series 2, 35, KIENLE, J. & SHAW, G. E Plume dynamics, thermal energy and long-distance transport of vulcanian eruption clouds from Augustine volcano, Alaska. Journal of Volcanology and Geothermal Research KOKELAAR, B. P. 22. Setting, chronology and consequences of the eruption of Soufriere Hills Volcano, Montserrat ( ). In: DRUITT. T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London. Memoirs, 21, LOUGHLIN, S. C, CALDER, E. S., CLARKE, A. B. ET AL. 22. Pyroclastic flows and surges generated by the 25 June 1997 dome collapse, Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs. 21, MARTEL, C., DINGWELL, D. B., SPIELER, O., PICHEVANT, M. & WILKE. M. 21. Eperimental fragmentation of crystal- and vesicle-bearing silicic melts. Bulletin of Volcanology, 63, MARTIN, D. P. & ROSE, W. I., JR Behavioural patterns of Fuego volcano, Guatemala. Journal of Volcanology and Geothermal Research, 1, MASSOL, H. & JAUPART, C The generation of gas overpressure in volcanic eruptions. Earth and Planetary Science Letters, MATTHEWS, S. J., GARDEWEG, M. C. & SPARKS, R. S. J The cyclic activity of Lascar Volcano, Northern Chile: cycles of dome growth, dome subsidence, degassing and eplosive eruptions. Bulletin of Volcanology, 59, MELNIK, O. & SPARKS, R. S. J. 22a. Dynamics of magma ascent and lava etrusion at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, MELNIK, O. & SPARKS, R. S. J. 22b. Modelling of conduit flow dynamics during eplosive activity at Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London. Memoirs, 21, MELSON, W. & SAENZ, R Volume, energy and cyclicity of eruptions at Arenal volcano, Costa Rica. Bulletin of Volcanology, 37, MILLER, A. D., STEWART, R. C., WHITE, R. A. ET AL Seismicity associated with dome growth and collapse at the Soufriere Hills Volcano, Montserrat. Geophysical Research Letters, 25, MORRISSEY, M. M. & MASTIN, L. G. 2. Vulcanian eruptions. In: SIGURDSSON, H. (ed.) Encyclopedia of Volcanoes. Academic Press. San Diego, MORTON, B., TAYLOR, G. I. & TURNER, J. S Turbulent gravitational convection from maintained and instantaneous sources. Proceedings of the Royal Society, A234, MURPHY, M. D., SPARKS, R. S. J., BARCLAY, J., CARROLL, M. R. & BREWER, T. S. 2. Remobilization of andesite magma by intrusion of mafic magma at the Soufriere Hills Volcano, Montserrat, West Indies. Journal of Petrology, 41, NAIRN, I. & SELF, S Eplosive avalanches and pyroclastic flows from Ngauruhoe Journal of Volcanology and Geothermal Research, 3, NEUBERG, J. & O'GORMAN, C. 22. A model of the seismic wavefield in gas-charged magma: application to Soufriere Hills Volcano, Montserrat. In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society, London, Memoirs, 21, NEUBERG, J., BAPTIE, B., LUCKETT, R. & STEWART, R Results from the broadband seismic network on Montserrat. Geophysical Research Letters, 25, NORTON, G. E., WATTS, R. B., VOIGHT, B. ET AL. 22. Pyroclastic flow and eplosive activity at Soufriere Hills Volcano, Montserrat during a period of virtually no magma etrusion (March 1998 to November 1999). In: DRUITT, T. H. & KOKELAAR, B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London, Memoirs, 21, ROBERTSON, R., COLE, P., SPARKS, R. S. J. ET AL The eplosive eruption of Soufriere Hills Volcano, Montserrat, West Indies, 17 September, Geophysical Research Letters, 25, SELF. S.. WILSON. L. & NAIRN. I. A Vulcanian eruption mechanisms. Nature SELF, S., KEINLE. J. & HUOT. J. P Ukinrek Maars. Alaska: II. Deposits and formation of the 1977 craters. Journal of Volcanology and Geothermal Research SHEPHERD. J. B.. HERD. R. A.. JACKSON. P. & WATTS. R Ground deformation measurements at the Soufriere Hills Volcano. Montserrat: II: Rapid static GPS measurements June 1996-June Geophysical Research Letters SPARKS. R. S. J Causes and consequences of pressurization in lava dome eruptions. Earth and Planetary Science Letters, SPARKS. R. S. J. & WILSON. L Eplosive volcanic eruptions - V. Observations of plume dynamics during the 1979 Soufriere eruption. St. Vincent. Geophysical Journal of the Royal Astronomical Society SPARKS. R. S. J. & YOUNG. S. R. 22. The eruption of Soufriere Hills Volcano. Montserrat ( ): overview of scientific results. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London. Memoirs. 21, SPARKS, R. S. J., YOUNG. S. R.. BARCLAY. J. ET AL Magma production and growth of the lava dome of the Soufriere Hills Volcano. Montserrat, West Indies: November 1995 to December Geophysical Research Letters SPARKS. R. S. J.. BARCLAY. J.. CALDER. E. S. ET AL. 22. Generation of a debris avalanche and violent pyroclastic density current on 26 December (Boing Day) 1997 at Soufriere Hills Volcano. Montserrat. In: DRUITT, T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London. Memoirs STI. J.. TORRES. R. C.. NARVAEZ. M. L.. CORTES. G. P.. RAIGOSA. J. A.. GOMEZ. D. M. & CASTONGUAY. R A model of Vulcanian eruptions at Galeras Volcano. Columbia. Journal of Volcanology and Geothermal Research THOMAS. N.. JAUPART. C. & VERGNIOLLE. S On the vesicularity of pumice. Journal of Geophysical Research UHIRA, K., YAMASATO. H & TAKEO. M Source mechanism of seismic waves ecited by pyroclastic flows observed at Unzen volcano. Japan. Journal of Geophysical Research VOIGHT. B.. HOBLITT. R. P.. CLARKE. A. B.. LOCKHART. A. B.. MILLER. A. D., LYNCH. L. & MCMAHON. J Remarkable cyclic ground deformation monitored in real-time on Montserrat. and its use in eruption forecasting. Geophysical Research Letters VOIGHT. B.. SPARKS. R. S. J.. MILLER. R. C. ET AL Magma flow instability and cyclic activity at Soufriere Hills Volcano. Montserrat. British West Indies. Science WAITT, R. B.. MASTIN. L. G. & MILLER. T. P Ballistic showers during Crater Peak eruptions of Mount Spurr Volcano, summer The 1992 Eruptions of Crater Peak Vent, Mount Spurr Volcano, Alaska. US Geological Survey. Bulletin WATTS. R. B.. HERD. R. A.. SPARKS. R. S. J. & YOUNG. S. R. 22. Growth patterns and emplacement of the andesitic lava dome at Soufriere Hills Volcano. Montserrat. In: DRUITT. T. H. & KOKELAAR. B. P. (eds) The Eruption of Soufriere Hills Volcano, Montserrat, from 1995 to Geological Society. London. Memoirs WOODS. A. W A model of vulcanian eplosions, Nuclear Engineering and Design WOODS. A. W. & BURSIK. M. I Particle fallout, thermal equilibrium and volcanic plumes. Bulletin of Volcanology WOODS, A. W. & KIENLE. J The dynamics and thermodynamics of volcanic clouds: theory and observations from the April 15 and April eruptions of Redoubt Volcano. Alaska. Journal of Volcanology and Geothermal Research WOODS, A. W. & KOYAGUCHI. T Transitions between eplosive and effusive eruption of silicic magmas. Nature. 37, WYLIE, J. J.. VOIGHT. B. & WHITEHEAD. J. A Instability of magma flow from volatile-dependent viscosity. Science YOUNG. S.. SPARKS. R. S. J.. ROBERTSON. R.. LYNCH. L. & ASPINALL. W. P Eruption of Soufriere Hills Volcano in Montserrat continues. EOS, Transactions, American Geophysical Union YOUNG. S. R., SPARKS. R. S. J.. ASPINALL. W. P.. LYNCH. L. L.. MILLER. A. D.. ROBERTSON. R. E. A. & SHEPHERD. J. B Overview of the eruption of Soufriere Hills volcano. Montserrat. 18 July 1995 to December Geophysical Research Letters