In November 1989, a swarm of earthquakes deep beneath Tachibana Bay, Kyushu Island, Japan, heralded the inexorable rise of magma toward the summit of Unzen Volcano, some 15 km upward and 15 km eastward, on the Shimabara Peninsula. When the "magma head" emerged in Jigokuato Crater on May 20, 1991, a beautiful but tragic drama began. It started peacefully as a budding flower unfolding lava petals. But by the time lava stopped flowing in February 1995, it had cost the City of Shimabara and surrounding towns over $2 billion in damage and 44 human lives. At its height, the crisis required the prolonged evacuation of 11,000 citizens. Amid this tragedy, however, volcanologists were able to make unprecedented visual and geophysical observations of processes of magma ascent, dome growth anddome-fed pyroclastic flows.

In May 26-29, 1997, about 75 scientists from Japan, USA, Germany, UK, Belgium, France, and Israel gathered in Shimabara to consider what had been learned during the Unzen episode. The twin themes of the meeting were in applying the lessons learned at this Decade Volcano to interpreting volcano behavior and mitigating hazards elsewhere, and in considering the addition of a bold new step, subsurface exploration through scientific drilling, to the techniques that have been applied to understanding the Unzen system. Because of the recency of activity, the "type system" character of Unzen as a dacite dome complex, the quantity and quality of the eruption data, the accessibility of the volcano, and the extent of the disaster, Unzen is an appropriate place to consider both themes.

Lying within the western end of the east-west striking Beppu-Shimabara graben and 70 km behind the volcanic front, Unzen Volcano consists of thick lava flows and domes, mainly of dacite, and their associated debris. The currently active dome complex, Fugendake, became active starting about 20 ka following a major sector collapse. The recent eruption began atop the the previous (13 and 4 ka) domes, and was likely similar to those dome-building events, though far bigger than lava eruptions of 1663 and 1792. Precursory activity involved progressive shoaling and eastward migration of seismicity, and finally the onset of phreatic eruptions six months before the appearance of lava. The phreatic activity may correlate with the top of the magma system traversing a conductive and presumably water-saturated zone at 2.5 km to 1.0 km below the summit. An increasing content of fresh vesiculated glass shards was noted in the ejecta during this period. Because the survival of fresh glass in a hydrothermal system would be short, it was concluded at the time that magma was approaching the surface. The final thrust to the surface was signaled by swarms of high frequency earthquakes at less then 1 km beneath the summit beginning 8 days before the dome appeared, as well as bulging of the high slopes of the volcano and opening of a small graben across the area of phreatic craters. Recognition of juvenile glass in the tephra, deformation of the summit, and intense shallow seismicity permitted Japanese scientists to correctly predict the arrival of lava. However, any joy at such success was soon tempered by concern for the villages below.

Abundant data on effusion rate define two, two-year-long pulses ofactivity. The first peaked early at 4 x 10^5 m3/day and declined to almost zero before rising abruptly to 2.5 x 10^5 m3/day and tailing off again. SO2 gas release and satellite-detected infrared emission correlate roughly with effusion rate. The more subtle chemical variations in lava may be related to effusion rate as well, through variable tapping of different zones of the chamber during times of different overpressures. High rates of effusion favored exogenous dome growth and low rates favored endogenous growth. Because the dome was growing at the precipitous edge of the volcano's summit, repeated collapse generated frequent block-and-ash pyroclastic flows. Both exogenous and endogenous growth fed such events, whose initiation was documented in detail from a vantage near their source. Over four years, the volcano extruded 13 exogenous lava lobes, left a large new endogenous summit dome, and produced 9,400 seismically detected pyroclastic flows. The latter were the cause of all the deaths and much of the destruction. A particularly large dome collapse on June 8, 1991 triggered vulcanian eruptions on that day and on June 11, with ejection of conduit-derived bombs at 100 m/s. Inferred propellant pressures, Mogi deformation models, and locations of the explosion earthquakes indicate a source at 500-800 m beneath the vent.

As dome growth ceased, a solid, dike-like spine rose through the carapace of the endogenous dome. At this point, the newly erupted volume had returned the volcano to its time-averaged eruption rate. It was concluded, correctly it appears, that the eruption was over. Another eruption of this magnitude should not be expected for at least a few thousand years.

Geodetically, the magma path can be modeled as 3 pressure sources which, like the seismicity, shoal to the east. The shallow conduit is now expressed as a narrow, highly conductive chimney. With the seismic, geodetic, structural, and electrical evidence taken together, the system is envisaged as a vertical, east-west striking dike with eastward ascending flow, narrowing to a pipe-like conduit just below the surface (FIGURE ?)*.

Prominent features of the newly erupted hornblende dacite are disequilibrium phenocryst assemblages and mafic enclaves. These were variously explained as the result of tapping of different zones of the magma chamber or of intruding of mafic magma into an initially quiescent and well-crystallized crustal reservoir. Concentration profiles in Fe-Ti oxides and experimental replication of resorption zones on plagioclase suggest syn-eruptive, rather than pre-eruptive, mixing. The rocks and activity show a remarkable resemblance to those of dacite domes elsewhere, such as Dutton Volcano in Alaska and Lassen Peak in California, and, significantly, at the currently active Soufriere Hills of Montserrat. Perhaps the only noteworthy difference is the somewhat greater explosivity of Soufriere Hills and somewhat higher peak effusion rate. Many meeting participants suspected that these two differences are related: that higher ascent rate diminishes degassing and therefore increases explosivity.

Interpretation of petrologic indicators of ascent rate such as decompression breakdown rims on hornblende is not yet clear. Although rims appear thin or absent, experimental work indicates that Unzen amphiboles may be stable at about half the pressure, or down to 50 MPa, of Mount St. Helens amphiboles, meaning that the same ascent rate would produce a narrower rim at Unzen. Rim thickness on Mount St. Helens amphiboles suggest ascent rates of 40 m/h for late domes. Rim thickness at Montserrat are found to decrease steadily from early to mid 1996, correlating well with effusion rate and suggesting that ascent rates increased steadily from 7m/h to 50m/h. Although this is two orders of magnitude faster than the rate of shoaling of seismicity at Unzen, it should be noted that the latter observation may reflect migration of the top of the magma system, whereas reaction rim width reflects the time that a single parcel of magma has spent out of the amphibole stability field during ascent. This rate may well be much higher. Crystallinity is another likely indicator of ascent rate. Much of the crystallinity of this type of lava may result from water loss during decompression, and so greater crystallinity reflects slower decompression.

Water loss is, in fact, central to the issue of Unzen's behavior. There was complete agreement with the concept that Unzen magma began its upward journey with 4-5 wt.% water and ended it with almost none - though still enough to facilitate fragmentation during the few 0.1 MPa decompressions of dome collapse events. Further, there was agreement that this water must be lost at less than 1-2 km depth in the conduit by some combination of vertical and/or horizontal flow from magma with interconnected bubbles (permeable foam) and/or cracks. Textural analysis of Unzen eruption products suggests that Unzen magma did not reach the degree of inflation postulated for permeable foam degassing, and so some other though no less effective mechanism may be called for. Although a visco-elastic conduit model can reproduce some aspects of Unzen behavior, a complete treatment of conduit flow must take into account the rapid vertical change in physical properties accompanying degassing. An interesting consequence is that most of the viscosity increase, and hence most of the pressure drop, occurs just below the growing dome.

A number of lines of investigation are now being brought to bear on the degassing problem. Physico-chemical models describe the growth of bubbles in rhyolitic melt upon decompression, and suggest that vapor pressure sufficient to initiate a pyroclastic flow could be maintained in a lava dome for some time. Hydrogen isotopes trace complex processes of water loss and readsorption in silicic lavas. Large-volume samples are subjected to controlled decompression from magmatic conditions in the laboratory to simulate conduit unloading and dome collapse conditions. This permits exploration of the critical conditions for fragmentation, and hence for explosive eruption.

Unzen Volcano presents an excellent case for evaluating some of the newly suspected relationships among eruptive behavior, vapor transport, and conduit flow. Indeed, the reason for existence of conduits themselves remains somewhat of a mystery. They represent a preferred path of magma, required for the development of towering volcanoes. Yet conduits do not remain "open", and so do not appear to always represent the weakest path, particularly in long-repose-time systems.

At Unzen, nearly complete degassing of magma occurred during ascent through a conduit whose position is well defined. The regime where degassing is expected to have occurred lies within reach of relatively modest drilling. Such holes would provide observations of stress, strain, temperature, and pressure conditions and samples of fluids, wallrock and intruded magma immediately after an eruption. They would also provide the means to determine the bulk gas permeability of the conduit environment, a site for tomography of the conduit structure, and an in situ borehole observatory for detecting signs of renewed activity.

With these goals in mind, recent accomplishments in high temperature directional drilling in Japan were reviewed, and the logistics of drilling at Unzen discussed. Three kinds of holes were deemed worthy of detailed consideration: (1) shallow core holes in the dome to investigate emplacement and cooling, pyroclastic flow formation, and dome stability; (2) a slant core hole intersecting the conduit at as shallow a depth as geometry allows, ideally at 500 m to 600 m, to address the degassing problem; and (3) a pair of 1500 m to 2000 m holes to seismically image and then sample the feeder dike within the water-saturated domain.

A final drilling plan requires resolution of issues of site access and engineering design, and, perhaps most critically, how to deal with a mountain that is host to catastrophes, religious shrines, and flower gardens. The meeting ended on a note of excitement for what had been and what could be accomplished.

The Unzen International Workshop was cosponsored by the International Continental Scientific Drilling Program (ICDP), Earthquake Research Institute of the University of Tokyo, and the National Committee of Volcanology and Chemistry of the Earth's Interior (Science Council of Japan). It was also supported by the Volcanological Society of Japan and by the City of Shimabara. Development of a science plan and proposal for a drilling project at Unzen is in progress. This process will allow for other interested scientists to join the effort.

On behalf of the workshop participants,

Setsuya Nakada. University of Tokyo
John Eichelberger, University of Alaska
Hiroshi Shimizu, University of Kyushu