The Minoan Deposits: a Review of their Characteristics and Interpretation

The maximum discharge rate is estimated at 1.4-4.2 10⁸ x kg/s and the plinian phase lasted several hours. The vent for the plinian phase opened up about 1 to 2 km south-west of Phira in a subaerial environment only slightly above sea-level. Phase 2 involved access of sea water either by enlargement of the vent or by new vents opening up. Phreatomagmatic activity generated base surges and well-stratified fall-out deposits. Variable emplacement temperatures of the deposits and contemporaneous plinian activity indicate highly variable magma-water interaction during phase 2. Phase 3 deposits are predominantly massive and poorly sorted, consisting of many individual units. They adhered to steep slopes and carried large lithics and intraclasts (several metres in diameter). The deposits are interpreted as poorly fluidized, low temperature pyroclastic flows, some of which contained liquid water. The flows are envisaged to form by very large phreatomagmatic explosions from a wide funnel-shaped vent filled with a pyroclastic slurry. Phase 3 disrupted a young dacitic lava shield which had been constructed in a pre-Minoan caldera. Phase 4 produced the thick ignimbrite fans seen around the present islands, composed of numerous flow units. The pyroclastic flows of phase 4 were hot (250-400° C), lithic-rich and fluidized, and could not adhere to steep slopes. Thick pyroclastic accumulations around the widening vent system of phase 3 may have formed a barrier to the sea and enabled the eruption to return to high temperature magmatic activity. There is no clear evidence that major caldera subsidence had begun at the end of phase 4. Phase 5 consisted of flood events forming alluvial fans on top of the ignimbrite. Avalanches of unstable caldera walls may have generated repeated flood waves following subsidence.

INTRODUCTION

The pyroclastic deposits formed in the Minoan eruption have been interpreted in a diversity of ways. These differences are partly attributable to advances in understanding of volcanic processes and development of new techniques. They are also related to genuine uncertainties in the knowledge of how pyroclastic deposits form and to some extent a lack of detailed data. This paper is partly a review of the current state of knowledge, but also emphasizes problems of interpretation. Some new interpretations and speculations are offered which can be tested by future research. The paper is organized in terms of the four successive phases and a fluvial/erosion phase recognized by most previous studies (Pichler and Kussmaul 1972; Bond and Sparks 1976; Pichler and Friedrich 1978; Heiken and McCoy 1984).

PHASE 1

A thin ash-fall deposit (up to 10 cm thick) underlies the plinian deposit in the south-south-east of the island and has been described in detail by Heiken and McCoy (1990). The deposit consists of fine-grained pumiceous ash impacted by the first plinian pumices. This deposit is similar to phreatomagmatic ashes found in many other sequences (Self and Sparks 1978; Wilson and Walker 1985) and suggests a short-lived initial influence of water.

The eruption began with the formation of a plinian pumice fall deposit dispersed to the east and south-east (Bond and Sparks 1976). Vertical changes in the deposit include: a slight decrease in pumice density (Wilson and Houghton 1990), coarsening of vesicle size, incoming of grey mafic clasts with moderate to poor vesicularity, abundant hydrothermally bright red to orange altered lithics in the upper parts of the deposit, and an overall reverse grading which is most pronounced in more distal localities.

The maximum lithic clast data (Fig. 1d of Bond and Sparks 1976) can be used to estimate column height and eruption rate using the method of Carey and Sparks (1986). The maximum column height estimate is 36 km and discharge-rate estimate is 2.5 x 10⁸ kg/s. These estimates are somewhat model-dependent and may be as much as 5 km in error, giving a range of inferred maximum discharge rate of 1.4 x 10⁸ to 4.2 x 10⁸ kg/s. The reverse gradings of the deposit suggests that the plinian eruption increased in intensity with time (Wilson et al. 1980; Pescatore et al. 1987) and a more detailed study of lithic size variation must be made to evaluate how the discharge rate might have varied.

A new method of calculating volumes from air-fall deposits has been applied by Pyle (1988; 1990) and indicates that the volume is less than previously thought; at between 0.45 and 1.5 km³ (DRE). Combining this volume with the discharge-rate estimates indicates a duration of somewhere between 1 and 8 hours.

The interpretation of the plinian deposit is not controversial. The deposit implies that the eruption took place from a vent somewhere between the present Nea Kameni island and Phira. The eruption represents a continuous discharge of magma in a dry eruption driven only by magmatic gas.

PHASE 2

The second-phase deposits are characterized by the development of prominent stratification defined by centimetric to decimetric variations in grain size and sorting. The deposits are generally more poorly sorted than the plinian deposit and many individual beds contain substantial proportions of ash (up to 90% < 1 mm). In contrast to the angular to irregular shapes of the plinian clasts, these deposits commonly contain well-rounded pumice. The most conspicuous feature of the deposits is the development of a large-scale cross-layering with spectacular undirectional bed-forms. These deposits were interpreted as base surge in origin by Pichler (1973). Bond and Sparks (1976) recognized that many of the dunes were asymmetric climbing mega-ripples. Heiken and McCoy (1984) showed that the dunes, with wavelengths of several metres to tens of metres and dune heights of up to 2 metres, indicate radial flow from the centre of volcano modified locally by topography. Prominent fine-ash beds and bomb-sag horizons mantle individual dune structures showing the intermittent character of the activity. The bomb dimensions are greater than any found in the plinian deposits at the same locality. For example bombs exceed 1 metre in diameter in the vicinity of Phira. Large ballistic clasts are often composed of fresh lava, including a prominent black glassy dacite absent from the plinian deposit. Hydrothermally altered lithics are common as well and several horizons of phase 2 deposits contain red to brown lithics which have a halo of red-to pink-stained ash adhering to them, indicating that these lithics were derived from an active hydrothermal system.

The phase 2 deposits show considerable variation in detail, which can be illustrated in Phira quarry (see stratigraphic section in Wilson and Houghton 1990). In the quarry phase 2 can be divided into three units by ballistic bomb horizons, each of which can be further subdivided on the basis of fluctuating matrix and lithic contents. Some deposits are fine-grained and poorly sorted, containing 70% to 90% ash less than 1 mm. Such deposits contain vesiculated ash bands and fragile accretionary lapilli. The inference is drawn that these deposits were relatively wet, although the absence of soft-sediment slump structures and faults even on steep slopes suggests that they were not generally water-logged. Other deposits are fines-poor with some units containing less than 10% ash smaller than 1 mm. These deposits are assumed to be somewhat hotter and drier. In Phira quarry bands of coarse angular pumice occur instratified with the obvious surge deposits. One band for example consists of 2.4 m of plane-bedded alternations of pumice and fine, often vesiculated ash with no internal bed-forms. These observations imply that the dry plinian eruption continued for part of phase 2. The bands of more angular pumice were often reworked by current action causing variations in bed thickness. Poorly sorted beds of massive pumice and ash containing suspended blocks occur as a minor facies of phase 2, and are interpreted as pyroclastic flow deposits similar to those in phase 3 (see below).

An important feature of the phase 2 deposits is that the transition from the plinian to phreatomagmatic styles of deposition is gradational. The uppermost part of the plinian deposit is interrupted by a fine-grained cross-layered deposit of phase 2 type. Heiken and McCoy (1984) described this horizon as the phreatomagmatic break. At Phira quarry this interval is represented by a thin ash bed which thickens northwards reaching a maximum thickness of 70 cm at Oia. The deposit is overlain by plinian fall deposits. The deposit represents a phreatomagmatic break heralding the onset of phase 2 while the plinian eruption continued (Heiken and McCoy 1984). As indicated above, plinian activity continued once phase 2 had fully developed. The dispersal of this deposit to the north-north-west implies either the development of a more northerly vent or some kind of topographic control directing the base surge currents.

An interesting feature of the phase 2 deposits is the development of occasional vertical degassing pipes originating at unusually large pumice clasts. In one example at Phira quarry, a fines-depleted pipe was traced 1.7 m up through stratified phase 2 deposits from a 20 cm diameter pink pumice. The pipe could have been generated from a hot pumice buried in cooler wet phreatomagmatic deposits. These observations place some constraints on the accumulation rate during phase 2. A sphere takes approximately R²/Kπ to cool down to a central temperature of 1/e of the initial temperature where R is the radius and K the thermal diffusivity. The degassing pipe must have formed in a time comparable to or shorter than this time. Substituting appropriate values (K =8 x 10^-7m²s^-1) yields a time of about 1 hour, yielding an accumulation rate of 3.0 cm/minute. The pipe was likely formed much quicker and this accumulation rate is very much a minimum.

There is a broad consensus that phase 2 represents the products of phreatomagmatic explosive activity. The prominence of fine ash represents some combination of increased fragmentation due to magma-water interaction and the premature flushing of fine ash due to the steam or liquid water rich character of the eruption column ash (see Brazier et al. 1982). The deposits have been cited as an example of a phreato-plinian deposit (Self and Sparks 1978). Interaction with water was evidently at a high level, because pumice clasts are as vesicular in the phase 2 deposits as in the phase 1 plinian deposit (Wilson and Houghton 1990).

Detailed inspection of the deposits indicates wide variation in degree of magma-water interaction and emplacement temperature. While some features such as vesiculated ash beds imply presence of liquid water, other horizons were clearly much hotter, with some large pumices developing pink interiors suggesting temperatures of several hundred degrees centigrade. Downey and Tarling (1984) record temperatures of 300^o C using thermal remanent magnetic properties, although it is not clear how the deposits were sampled. New data presented by McClelland and Thomas (1990) indicate temperatures between 100 and 200^o C. The implication is of episodic eruptions involving variable interaction with water either in a single evolving vent system or a line of vents as envisaged by Heiken and McCoy (1984).

PHASE 3

The origin of the massive deposits that form most of the upper parts of the Minoan succession around the caldera wall is controversial. Pichler and colleagues (Pichler and Kussmaul 1973; Pichler and Friedrich 1978) interpreted them as ash-flow deposits and considered that they represented the initiation of caldera formation. Bond and Sparks (1976) presented arguments for believing that the deposits were cold and interpreted them as pumiceous mud-flows. Heiken and McCoy (1984) considered the deposits to represent a phreatomagmatic deposit with multiple facies.

The deposits are massive, poorly sorted mixtures of pumice and ash containing abundant lithic blocks, some reaching several metres in diameter (see Fig. 6 in Pichler and Friedrich 1978). Glassy dacite blocks are a prominent lithic type and large intraclasts of dacite hyaloclastite occur. Lava breccia and altered tuff up to 10 metres across are occasionally found. These intraclasts are often sheared, with a sense of strain consistent with laminar flow away from the caldera rim. The deposits crudely mantle the topography, occurring even on the upper flanks of the Mikros Profitis Ilias stratocone and Megalo Vouno. However, they thicken noticeably into topographic lows and reach their greatest thickness at low points along the caldera rim. The deposits adhere on slopes as great as 20°.

Certain aspects of the deposits further constrain the possible emplacement mechanisms. At Oia and in Thera quarry they are divisible into weakly defined flow units, typically from 1 to 5 m thick, by variations in the abundance of lithic and pumice blocks. Inverse graded layer 2a subdivisions (Sparks et al. 1973) are occasionally observed and the lithic blocks often reach a maximum in size and abundance on the layer 2a and 2b boundary. Concentrations of lithics and basal layers form distinct horizons which can be traced a few hundred metres laterally. Pumice concentration zones, reverse grading of pumice and fines-poor degassing pipes are rare, and the deposits would be classified as type 1 (poorly to non-fluidized; Wilson 1980). Heiken and McCoy (1984) describe impact structures implying ballistic emplacement for the largest blocks. However, we have only observed unequivocal impact structures on the transitional boundaries between phase 2 and phase 3, and between phase 3 and phase 4.

The matrix of phase 3 deposits varies from fines-poor, loose and friable to moderately fines-rich and cohesive. In the Phira quarry where the fine ash matrix is abundant, it is typically vesicular and the lithic blocks often have coatings of very fine ash. Very thin (1 to 5 mm wide) degassing fissures are also observed. These features imply that some of the deposits were emplaced with three components: pyroclasts, liquid water and gas (steam).

Palaeomagnetic studies have yielded apparently conflicting results for the temperatures of emplacement. Wright (1978) found random magnetizations of blocks consistent with low temperature emplacement. Downey and Tarling (1984) reported results that they interpreted as high emplacement temperatures (300-400° C) though it is not clear what kind of material was sampled. McClelland and Thomas (1990) document a more complex situation in which lithic clasts varied from being completely cold to over 250° C. Their preliminary data also indicate that the ash matrix of some phase 3 deposits was cold.

The contact between phase 2 and 3 is transitional and is not always easy to define precisely. In Phira quarry (Wilson and Houghton 1990, Fig. 3) the phase 2 deposits become more massive and poorly sorted and the stratification becomes weaker. Approximately 2 metres of transitional material occurs and contains a vesiculated ashy matrix. The transitional material shows cross-bedding and a paucity of large (10-40 cm) lithic blocks.

The phase 3 deposits predominantly have a flow origin. The absence of fluidization structures, pumice grading and pink oxidized pumices, together with the palaeomagnetic evidence, are consistent with low temperatures. The adherence of the flows to steep slopes and capacity to carry large intraclasts implies high viscosity and/or yield strength. The presence of flow units with vesiculated matrices and degassing fissures indicates that at least some contained liquid water. Those with basal layers, lithic concentration zones and fines-poor matrices could have been somewhat higher in temperature.

At the present state of knowledge an interpretation of these deposits combines aspects of all three previous views. The deposits appear to be emplaced as low grade (low temperature) primary pyroclastic flows perhaps ranging from less than 100° C to perhaps a few hundred degrees (up to 300° C?). A full evaluation of the emplacement temperatures must await a more thorough study of their thermal remanent magnetic properties. Those with temperatures less than 100° C and which contained liquid water would have behaved effectively as mud-flows. It is a moot point whether they should be called lahars rather than pyroclastic flow deposits. However, we interpret them to be of primary volcanic origin. The mantling character of the deposits implies that the flows were quite energetic and reached almost all parts of the presently preserved topography. The low temperatures of these pyroclastic flow deposits can be attributed to their phreatomagmatic origin in agreement with Heiken and McCoy (1984). The eruptive conditions required to generate large volumes of energetic low temperature flows are discussed further below.

PHASE 4

The deposits of phase 4 are mostly confined to the coastal plains (Bond and Sparks 1976; Heiken and McCoy 1984) where they form sea cliffs of massive to stratified pumice and ash-rich deposits. Along the caldera rim thin veneer deposits of phase 4 material can be seen overlying phase 3 deposits. The deposits are distinguished by a subtle colour change from white (phase 3) to creamy-white (phase 4) best seen when the sun is at a low angle just before sunset (Fig. 1). At Oia phase 4 deposits can be seen to overlie phase 3 with a rapid transition.

There are many contrasts with phase 3 deposits, described in detail by Bond and Sparks (1976) and Sparks (1976). The phase 4 deposits are finer grained and usually do not contain large pumice clasts greater than 10 cm or lithic clasts greater than 5 cm in diameter. However, concentration zones of larger pumice clasts at the top of flow units, normal grading of lithics, and basal 2a layers are observed. In addition, gas segregation pipes and fines-depleted layers around many lithic clasts indicate that these deposits were formed from strongly fluidized pyroclastic flows. Large numbers of flow unit boundaries are usually recognizable, including stacked sequences of basal layers (5 to 25 cm thick) and ignimbrite stratified on a centimetric to decimetric scale. The phase 4 deposits contain an unusually high proportion of lithic clasts (35% to 50 wt%). This is not obvious in the field as much of the lithic material is ash and its lithic-rich nature is only apparent after sieve and component analyses (Bond and Sparks 1976). The deposits contain a wide diversity of lithic rock types (Bond and Sparks 1976; Heiken and McCoy 1984). The glassy dacitic lithics are much less prominent than in phase 3 deposits and basement (limestone and schist) clasts are often present.

Phase 4 deposits are generally deposited on low slopes and form a series of aprons around the islands sloping away at angles of 1 to 3°. Most of the flows detached almost completely from the steep inner slopes though patchy thin veneer deposits survive. Palaeomagnetic studies (Wright 1978; Downey and Tarling 1984; McClelland and Thomas 1990) show that the deposits were emplaced hot, from 200 to over 400° C. Occasional co-ignimbrite ash fall deposits enriched in vitric components are found interbedded with flow units (Sparks and Walker 1977).

Phase 4 deposits have an uncontroversial origin as products of numerous high temperature pyroclastic flows which progressively accumulated to form thick successions of ignimbrite. An unexplained feature is the high lithic content and high degree of lithic fragmentation, which is considerably greater than in other documented ignimbrites (e.g. Sparks 1976). An important facies of the phase 4 ignimbrites is the occurrence of intercalated horizons of very coarse lithic breccia (Fig. 2). Bond and Sparks (1976) interpreted these breccias as the result of catastrophic floods. This interpretation was based on the clearly erosive contacts of some breccia horizons with ignimbrite and the presence of strong fluidization features in some breccias. Undoubted alluvial fan deposits, including lithic breccia, overlie the ignimbrite and belong to the post-eruption phase 5. However, doubt must now be cast on the alluvial origin of the intercalated breccias. The breccias range from those with an ignimbrite matrix to fines-poor varieties rich in fluidization pipes and pods (Fig. 3).

They are particularly well-developed along the north coast near Oia (Fig. 2) where breccia beds show remnants of ignimbrite preserved locally within fines-poor deposits (Fig. 3). Such 'two-phase' textures are typical of strong fluidization in proximal ignimbrite facies (Wilson 1980; 1985). They are similar in appearance to lag-breccias described elsewhere (Druitt and Sparks 1982; Walker 1985; Freundt and Schmincke 1986). In general the breccias become more prominent close to the steep inner slopes of the islands indicating that their deposition was associated with the break in slope. If they are lag breccias they may originate as a consequence of a hydraulic jump (Freundt and Schmincke 1986) in which the pyroclastic flows decelerate abruptly and deposit heavy components. Another possibility is that they represent discrete eruptive events (Walker 1985). Strong fluidization features (Fig. 3) might also reflect interaction with sea water at the coast. Possible interpretations are discussed further below.

PHASE 5

The alluvial origin of the stratified lithic-rich deposits overlying the ignimbrites is generally accepted. However, the cause of widespread flooding on islands with very small catchment areas is enigmatic.

TIME GAP

Many workers have suggested that the Minoan eruption involved a large time-gap of many years or tens of years. A gap was principally proposed to explain differences in pottery styles between Akrotiri and the main destruction level on Crete. Recent discoveries on Rhodes have shown that the ash layer occurs beneath the Minoan IB pottery level and so the archaeological case for a time-gap is no longer relevant. Downey and Tarling (1984) have proposed that a large time-gap between phase 1 and 2 is evident in palaeomagnetic data. However, the transitional nature of the contact with the early phreatomagmatic break, the continuing plinian activity within phases 2 and the lack of erosion of the plinian deposit are conclusive that no such time-gap exists. Other arguments are presented by Pichler and Friedrich (1978) and Sparks (1985) and other explanations must be sought for the palaeomagnetic observations.

THE SANTORINI CALDERA

Heiken and McCoy (1984) demonstrated that the Santorini caldera depression is a multiple structure formed by several collapse events. Druitt et al. (1989) have documented twelve major pyroclastic deposits of the caldera wall, each representing an eruption with volumes in excess of 1 km³, several of which should have produced caldera collapse. At least two earlier calderas are evident from unconformities and lava-fill sequences in the caldera wall (Heiken and McCoy 1984; Druitt et al. 1989). Heiken and McCoy (1984) proposed that the Minoan collapse was confined to the northern depression and involved about 19 km³ of collapse.

Friedrich et al. (1988) have recently documented compelling evidence for a pre-Minoan caldera in the north. They describe stromatolite clasts in the phase 3 deposits in northern Thera and Therasia with ages of about 17,000 years BP. The stromatolites imply a water-filled depression in the northern area. Francaviglia (1989) has discovered original plinian deposits on a ledge halfway down the caldera wall near Phira. Subsequent geochemical analyses have confirmed that this is the Minoan deposit (Druitt and Francaviglia 1990).

These observations demonstrate that a substantial caldera depression already existed prior to the Minoan eruption, perhaps related to the 18,500 years BP Cape Riva event (Druitt 1985).

DISCUSSION 

From existing published data and our observations a synthesis of the eruption is attempted. Fig. 4 shows a schematic cross-section at the beginning of the eruption across from Phira to Therasia. The section shows an older caldera depression that had been partly filled in by a lava shield though some of the margins of the shield may have still been beneath sea-level. This situation is similar to today except that the post-18,500 year BP equivalent of the Kameni islands substantially filled the submerged portion of the caldera. Several of the main features of the Minoan eruption are explained by this geometry.

The plinian phase 1 began on the eastern side of the caldera within a triangle defined by the present Kameni islands, Phira and centre of the northern depression. The initial thin phreatomagmatic fall unit indicates that water had access to the initial vent. The lithic population dominantly consists of hydrothermally altered rock and older lava, perhaps implying either that the vent developed close to the older caldera edge or that part of the older volcanic sequence formed land above sea-level within the caldera. The plinian activity was dry implying a vent isolated from external water.

Phase 2 saw the onset of a significant influx of water into the vent (Wilson and Houghton 1990) consistent with an initial vent close to above sea-level. Had the pre-Minoan shield completely filled up the old caldera to 200 or 300 m height then it is not easy to envisage how water could have had such a dominant influence through much of the eruption. There are two possibilities. First a fissure system could have extended from the subaerial vent to a submerged part of the shield. As the dominant tectonic trend is NE-SW, extension from the plinian vent would have to have been to the south-west through the postulated shield. Second, the plinian vent could simply have been breached by the sea, allowing access of abundant water. This latter situation is illustrated in Fig. 4. Both possibilities are consistent with the observations.

The increasing abundance of glassy dacitic clasts upwards through phase 2 suggests either that the vent began to engulf lavas of the post-18,500 yr BP shield (Fig. 4b) or that new vents opened up along a fissure propagating into the shield. The complexities of phase 2 activity include alternating or simultaneous fall and surge activity together with fluctuating temperatures and water-magma mixing ratios. Again, variable ingress of water into a single widening vent or multiple vents in different environments are both plausible. We interpret phases 2 and 3 as large scale silicic analogues to Surtseyan activity as postulated by Kokelaar (1983). He envisages a funnel-shaped vent in which fragmenting vesicular magma is mixed with water and a pyroclastic slurry to generate repeated explosions. On Santorini each explosion may also be accompanied by ballistic ejection of large lithics which forms a basal horizon before the emplacement of a surge unit. The large size of ballistic blocks, the ability of surges to deposit on the summit of Profitis Ilias, and the large scale of the dune structures all point to very energetic but episodic explosions (Fig. 4b).

The transition into phase 3 is gradational and involves an increasing dominance of low temperature pyroclastic flows, some of which are thought to have contained liquid water. The abundance of large glassy dacitic clasts suggests that the vent system had further enlarged and the eruption began to excavate into the young lava shield. Large clasts of glassy dacite hyaloclastite in the flow deposits are consistent with a funnel-shaped vent which had started to disrupt the subaqueous interior of the lava shield complex (Fig. 4c). The deposits contain numerous flow units and drape the topography although they are thickest at low points along the rim, suggesting high energy. If, as this scenario demands, the vent was now at or below sea-level then the explosions would have to have been vigorous enough to generate flows capable of surmounting the 200 to 300 m high caldera rim.

We suggest that phase 3 represents the evolution of the vent system to a large funnel full of a slurry of ash, pumice, steam and water. We envisage repeated violent explosions ejecting juvenile clasts and large batches of this slurry, and generating low temperature but high energy pyroclastic flows from substantial pyroclastic fountains. Slumping of the funnel walls generated huge blocks up to several metres in diameter which were incorporated into the slurry and ejected with the flows. A corollary of this interpretation is that a proportion of ejecta would have not cleared the old caldera wall and would have accumulated in the depression. It is not easy to envisage what kind of constructional feature might have been formed, one possibility being a giant tuff-ring (Fig. 4). In one sense this concept is analogous to that proposed by Heiken and McCoy (1984) for phase 3, except that we believe the deposits preserved on the outer flanks are predominantly flow rather than fall-out in origin.

The change to phase 4 activity is marked by a substantial increase in temperature of the erupted material and an increasing abundance of lithics (Bond and Sparks 1976). The pyroclastic flows were more strongly fluidized and did not adhere much to the steeper slopes. The lithics are increasingly dominated by old volcanic clasts and basement, which could be interpreted as either a shift to new vents or the enlargement of the vent system to the point where slumping of the old caldera wall became important. The upward increasing emplacement temperatures suggest that the postulated tuff-cone or tuff-ring and/or accumulation of pyroclastic flows within the old caldera progressively excluded water from the vent, allowing a reversion to magmatic activity (Fig. 4d). The large number of flow units in the phase 4 ignimbrite implies that activity continued to be episodic, accumulating ignimbrite fans around the coasts. Similar deposits formed in the Mount St Helens eruptions of 1980 (Rowley et al. 1982). The differences between phase 3 and 4 deposits are interpreted to reflect the temperature, rheology and fluidization states of the flows rather than their velocities. The high lithic content and high degree of fragmentation of lithics in phase 4 ignimbrite is consistent with continued slumping of the caldera walls (Fig. 4) and perhaps implies that a significant amount of caldera enlargement involved explosive ejection of country rock.

The lack of magma and water interaction during phase 4 implies that final caldera subsidence had still not occurred. The caldera floor is over 300 m below sea-level and it is hard to envisage how hot ignimbrites could be formed if the vent system were submerged to that depth. We therefore suggest that final subsidence occurred after phase 4. There are few data on rates of caldera collapse. However, the collapse of Fernandina caldera in Galapagos by 300-400 m took place over 10 days. Major explosive eruptions like the Minoan have often only taken a few tens of hours, such as Tambora in 1815, Krakatoa in 1883 and Katmai in 1912. The discharge rate in the plinian phase probably represents a lower limit to the rates in the later phases, which implies that the entire eruption need only have taken three or four days to erupt over 30 km³ of magma. While speculative, it is plausible that the magma discharge was too rapid for subsidence to initiate or to keep pace.

Flooding and erosion comprise the final episode. One possibility is that the floods are related to subsidence and post-eruption instability of the caldera walls. Caldera walls just after collapse are highly unstable and frequent rock avalanches occur. Large embayments are thought to mark the major landslides. Avalanches in enclosed bodies of water such as fjords are known to generate huge waves or surges which travel hundreds of metres uphill. We therefore speculate that sudden floods were generated by repeated avalanches of the unstable walls of the new calderas. This hypothesis explains how major alluvial deposits could be formed on small islands with a small catchment area. We acknowledge that much of this model of the eruption remains speculative. We hope that it provides stimulation to further investigation of these fascinating if enigmatic deposits.

-------------------------------------------

--------------------------------------------