Geological setting

Aso volcano is situated in the Hohi Volcanic Zone (HVZ) in Central Kyushu, SW Japan, and forms part of a chain of five large Quaternary caldera complexes and 23 smaller volcanoes that extend NE-SW from SW Honshu to Southern Kyushu (Fig. 1a). It lies within the transitional Pigeonitic-Hypersthenic Rock Series zone of Kuno, (1959), and is situated ~20 km behind the volcanic front [above the 140 km Benioff zone contour (Kubotera, 1981)], and ~275 km inland from the Ryukyu Trench. Volcanic activity at Aso has been subdivided into Precaldera, Caldera and Post-caldera stages, in which both tholeiitic and calc-alkaline magmas have been recognized; only the caldera stage will be discussed in this paper.

Figure 1. (a) Map of Japan illustrating the location of Aso within the Hohi Volcanic Zone, Central Kyushu, SW Japan [adapted from Kubotera (1981)]. (b) Simplified geological map of Aso caldera [adapted from Geological Survey of Japan Map series 1:50000, Ono & Watanabe (1985)].

The following summary of the geological history of Aso is based on a synthesis of previously published work (Lipman, 1967; Watanabe, 1978; Ono et al., 1981), combined with additional field data collected during this study. Descriptions of Aso 1-4 are based on field evidence collected during this study, whereas ages for the four eruptive units are from Ono et al., (1981).

Basement geology

The basement in Central Kyushu consists of Palaeozoic to Mesozoic metamorphic rocks, intruded by Cretaceous granites and small Miocene volcanic and plutonic complexes. These units are sporadically covered by Cretaceous to Palaeogene sediments (Watanabe, 1978; Ono et al., 1981). Based on evidence obtained from two drill sites located close to the northern edge of Aso caldera (Watanabe, 1978; Ono et al., 1981), the pre-Neogene basement is known to exist close to the present-day surface, with Cretaceous granites encountered at shallow depths (150-470 m).

Precaldera activity (3-0·4 Ma)

Volcanic activity before the caldera stage occurred from the Late Pliocene to the Early Pleistocene (3-0·4 Ma), characterized by subaerial intermediate to acidic composition lavas and pyroclastic deposits (Watanabe, 1978; Fig. 1b). The most common rock type is pyroxene andesite, followed by subordinate hornblende andesite, biotite rhyolite, and basaltic to dacitic pyroclastic material. The majority of the units are subhorizontal, and are known to have originated from several volcanic centres (Watanabe, 1978).

Caldera activity: the Aso Pyroclastic Flow Deposit (APFD ~370-70 ka)

Activity during the caldera stage commenced in the Mid to Late Pleistocene (370-70 ka; Matsumoto, 1963; Watanabe, 1978; Ono et al., 1981; Aramaki, 1984), with the 18 km * 25 km caldera formed incrementally by four major eruptive events (Aso 1-4), collectively referred to as the Aso Pyroclastic Flow Deposit (APFD; Fig. 1b). The APFD consists of voluminous eruptions of basaltic to rhyolitic lavas and pyroclastic fall and flow deposits, with total volume estimates ranging from 175 km³ (Matsumoto, 1963; Aramaki, 1984) to >300 km³ (Ono, 1970). Each unit is distributed around the edge of the caldera, forming either shallow-sloping plateaux (1-2°) or flowing into radial valleys incised into the basement outside the caldera (Ono et al., 1981). In addition to the main flow units, two small flank eruption centres (Akai and Omine) consisting of andesitic lava and scoria deposits (not shown as separate units on Fig. 1b), also occur several kilometres to the west and south-west of the caldera.

Petrological, geochemical and stratigraphical relationships within Aso 1-4 suggest that they represent the eruptive product of a zoned magma chamber, with the most evolved magmas erupted first (Ono et al., 1981; Hunter, 1993). Replenishment rates were of the order of a few tens of thousands of years (Ono et al., 1981), with eruption volumes increasing from Aso 1 to 4 (Table 1). Aso 1-3 are compound cooling units, consisting of a range of welded to non-welded basaltic to dacitic pyroclastic material (Hunter, 1993), whereas Aso 4 is more complex, consisting of eight subunits of non-welded to welded dacitic ash and pumice, grading up into basaltic scoria (Lipman, 1967; Watanabe, 1978; Hunter, 1993).

Table 1. Comparison of the eruptive ages of Aso 1-4 and volume of material erupted (Aramaki, 1984).

Aso 1 (370-270 ka)

This is the first and smallest of the pyroclastic flows, made up of several compound cooling units consisting of moderately to densely welded andesitic to dacitic pyroxene-bearing tuff. It is characterized by a grey to red coloration and blocky structure, with the extent of welding dependent on height in the section. The tuff is sporadically topped by an oxidized soil horizon and frequently contains lenses of obsidian with microphenocrysts of plagioclase. Exposures of this unit are limited to the east and north-west caldera walls. Mixed magma clasts ( <= 2 mm) containing xenocrystic plagioclase + clinopyroxene + titanomagnetite, as well as pumice clasts ( <= 1·25 cm) occur throughout the unit, whereas at its base, <= 15 cm juvenile glassy clasts are abundant. Overall, Aso 1 is predominantly tholeiitic in nature, but becomes calc-alkaline with time.

Aso 4 (80-70 ka)

This is the largest and youngest of the four main units of the APFD. It consists of airfall and ignimbritic deposits, displaying complex variations in texture, thickness, extent of welding, and phenocryst and groundmass compositions (Lipman, 1967). It has been subdivided into eight subunits (Lipman, 1967; Watanabe, 1978), with the description of each unit in Table 2 based on a combination of field evidence (this study) and previous work (Lipman, 1967; Watanabe, 1978). Depositional mixing in the form of glass lenses and pumice fragments is present in some units, with Aso 4 characterized by an overall calc-alkaline signature (Fig. 2; Table 3).

Table 2. Summary of eruptive units in Aso 4 based on previously published work (Lipman, 1967; Watanabe, 1978) and field evidence (this study).

Table 3. Essential phenocryst and groundmass mineralogy for tholeiitic and calc-alkaline parental magmas and rock series, following the petrological classification scheme devised by Kuno, (1959); adapted from Kuno, (1959) and Kawano et al., (1961).

Figure 2. (a) Classification of tholeiitic and calc-alkaline series within the APFD using the geochemical system devised by Miyashiro, (1974). The dividing line between the two series is SiO₂ wt % = 6·4(FeO*/MgO) - 42·8, where FeO* = (FeO + 0·9Fe₂O₃). Filled symbols represent the tholeiitic (Th) series; open symbols represent the calc-alkaline (Ca) series. (b) Schematic sketch of hypothetical trends displayed by the data from the APFD. In this model (A), data are treated as four separate trends (i.e. Aso 1-4), which originate from a common starting composition. By changing conditions in the magma chamber, the concentrations of specific elements have varied between the four eruptive units at constant SiO₂ wt % values. (c) Schematic sketch of hypothetical trends displayed by the data from the APFD. In this model (B), data are treated as a single evolutionary trend (i.e. the APFD), with changes in the fractionating assemblage and amount of fractional crystallization producing changes in inflections in the trend with time between Aso 1 and 4.

Throughout the APFD, two subtle petrochemical changes are apparent as the caldera-forming stage evolved with time, with: (1) the most abundant rock type changing from basaltic andesites to dacites from Aso 1 to 4, and (2) the magmatic series changing from dominantly tholeiitic (Aso 1 and 2) to dominantly calc-alkaline (Aso 3 and 4).

Post-caldera activity

Post-caldera activity is characterized by eruptions from 15 bimodal cones (basalt to basaltic andesite and dacite), situated along an east-west trend dividing the caldera into two valleys (Fig. 1b). At present, only one of the central cones is still active (Nakadake), with the latest phase of phreatic activity, in September 1994, consisting of periodic eruptions of small primary and xenolithic clasts, accompanied by episodic ash plumes up to 1 km above the active crater (Japan Meteorological Agency, 1994).

Review of petrological classification schemes

The petrological classification scheme devised by Kuno, (1959) is based on the groundmass mineral assemblage, where the tholeiitic rock series is characterized by clinopyroxene as the only groundmass pyroxene phase [referred to as the Pigeonitic Rock Series (PRS) by Kuno], whereas rocks from the calc-alkaline series contain orthopyroxene ± clinopyroxene as groundmass pyroxene phases [referred to as the Hypersthenic Rock Series (HRS) by Kuno]. To facilitate the classification of rocks using this scheme, Kuno, (1959) established a system of recognizable essential phenocryst and groundmass mineralogies for the tholeiitic and calc-alkaline series, based on the appearance and disappearance of phases within discontinuous reaction series (Table 3). From these reactions, Kuno, (1959, , 1960) implied that evolution of the tholeiitic rock series was controlled by fractional crystallization of an olivine basaltic parent, characterized by a discontinuous reaction series of olivine, followed by augite and hypersthene, with this producing a progressively more siliceous liquid. Kuno, (1959, , 1960) also noted that contamination of this parental magma and/or a related differentiate by crustal material, however, would cause the crystallization temperature to decrease, which in turn would favour hypersthene (± hornblende) formation, to create a calc-alkaline (HRS) rock series in preference to a tholeiitic (PRS) rock series, as well as changing the bulk chemistry of the system towards higher SiO₂ values.

Kuno, (1960) found that continued separation of augite ± hypersthene from the parental magma produces a relative enrichment in the iron content, with increasing water contents changing the oxidation state of iron from Fe²⁺ to Fe³⁺. This change in oxidation potential not only permits titanomagnetite (± haematite) to crystallize, but also mobilizes the alkalis, leaving excess Al and P cations to complex with Ca cations to produce anorthite and apatite, both of which are conspicuous within the tholeiitic series (Kuno, 1959, , 1960). The removal of Ca to form anorthite and apatite results in successive depletion in the wollastonite component of pyroxene, producing a progressive evolutionary trend from clinopyroxene to orthopyroxene as the primary pyroxene groundmass phase, and hence an associated shift from the tholeiitic to calc-alkaline series, i.e. magma series formation is controlled by fractionation.

A second mineralogical classification scheme to distinguish tholeiitic and calc-alkaline magma series was devised by Sakuyama, (1981), based on the presence or absence of reversely zoned mafic phenocrysts (Table 4). He defined the two series as: (1) R-type, characterized by olivine and pyroxene phenocrysts, which are both reversely and normally zoned, with disequilibrium mineral assemblages implying magma mixing to be the main control on magmatic evolution (mineralogically equivalent to calc-alkaline or hypersthene rock series), and (2) N-type, characterized by normally zoned mafic phenocrysts exhibiting no signs of disequilibrium, with magma evolution controlled by differentiation (i.e. tholeiitic or pigeonite rock series).

From his study of the Myoko and Kurohime volcanoes in Central Honshu, Sakuyama, (1981) illustrated that the fundamental differences in the phenocryst phases between the two magma series occurred as a result of magma mixing, whereas the groundmass mineralogy actually remained unaffected, exhibiting a systematic change in pyroxene phase because of successive temperature decreases in both series. Following on from the work of Kuno, (1959, , 1960) and Kushiro, (1969), he proposed that the groundmass pyroxene phases evolved along a continuous reaction series from augite + pigeonite to augite + pigeonite ± orthopyroxene and to orthopyroxene ± pigeonite, associated with increasing silica in the residual liquid composition. The eventual demise of augite in the groundmass was attributed to an expansion of the primary hypersthene phase field relative to augite, associated with increasing volatile concentration during progressive crystallization.

Uncertainties, however, arise with both of these classification schemes, as the stability and occurrence of orthopyroxene is dependent not only on the chemical composition of the magma, but also on the magmatic temperature and cooling rate immediately before quenching (Fujinawa, 1988), leading to discrepancies in the classification of samples using Kuno's groundmass pyroxene scheme. This problem was initially tackled by Kawano et al., (1961) who, while working on a series of rocks from NE Honshu, discovered that some samples which had previously been classified as belonging to the tholeiitic series, fell along geochemical trends similar to the calc-alkaline series, i.e. the magmas had crystallized before orthopyroxene had stabilized in the groundmass, allowing metastable pigeonite to remain in the melt, as a result of rapid cooling rates. To resolve this classification problem, Kawano et al., (1961) referred to such samples as `subordinate' tholeiites.

The type and extent of magma mixing in a rock series can also result in mis-classification when using Sakuyama's scheme, with crypto-scale mixing between tholeiitic samples producing reversely zoned phenocrysts while retaining a tholeiitic geochemical signature. Classification problems of this type were encountered by Kay & Kay, (1985), when using zoned phenocrysts to classify tholeiitic and calc-alkaline rock series from the Aleutians. Using Kuno's traditional classification scheme, they sub-divided the two original series into four, to include additional transitional tholeiitic and calc-alkaline series that were characterized by intermediate FeO*/MgO ratios. In this work, they inferred that the transitional calc-alkaline series (containing primary phenocrysts of amphibole and/or orthopyroxene) evolved by mixing a high-alumina basalt with a more evolved magma, whereas the transitional tholeiitic series (containing orthopyroxene and/or pigeonite), exhibited evidence for upper-crustal magma mixing, to produce reversely zoned phenocrysts. Thus, rocks that would have been designated as R-type, and hence calc-alkaline, were reclassified as belonging to the transitional tholeiitic series.

Initially, each sample from Aso was classified as eithertholeiitic or calc-alkaline based on a combination of petrographic observations using the schemes devised by Kuno, (1959) and Sakuyama, (1981). These classifications were subsequently confirmed geochemically [using the FeO*/MgO vs SiO₂ wt % system devised by Miyashiro, (1974)]. Where discrepancies arose between the petrological and geochemical classification schemes, the geochemical classification was adopted in preference to assigning transitional sub-classification groupings. Using these data, each of the eruptive units (Aso 1-4) was then assigned a dominant magmatic series signature, based on the predominance of tholeiitic or calc-alkaline samples.

Geochemical classification of the magma series in Aso 1-4

On the Miyashiro diagram (Fig. 2a), a progressive decrease in FeO*/MgO at SiO₂ values >62 wt % can be observed for Aso, changing the dominant magmatic series from tholeiitic to calc-alkaline over time. When Miyashiro devised this classification scheme, he discriminated between the two magmatic series by comparing the gradient defined by a set of samples with an established dividing line between the tholeiitic and calc-alkaline fields. On this basis, he proposed that rock suites with a tholeiitic affinity would be characterized by a steeper gradient on the FeO*/MgO vs SiO₂ wt % diagram than those with a calc-alkaline affinity. Thus, based on variations in the general compositional trends within Aso 1-4, Aso 1 and 2 can be classified as having a general tholeiitic affinity (even though the least silicic samples from Aso 1 fall within the calc-alkaline field as defined in the FeO*/MgO vs SiO₂ wt % diagram). In contrast, samples from Aso 3 and 4 define trends that are characterized by shallower gradients than the dividing line, and therefore have a calc-alkaline affinity (even though the least silicic samples from Aso 3 and 4 fall in the tholeiitic field). According to these geochemical trends, a general shift in the dominant magmatic series from tholeiitic to calc-alkaline is evident between Aso 1 and 2 and Aso 3 and 4, as the system evolved with time.

If the four eruptive units (Aso 1-4) from the APFD are considered as individual trends in Fig. 2a, which originate from a similar starting magma composition, each unit can be seen to be characterized by a different gradient that decreases with time from Aso 1 to 4 (Fig. 2b). Alternatively, Aso 1-4 can also be viewed as a single evolutionary unit in Fig. 2a, with the data fanning out at >= 62 wt % SiO₂, and Aso 1-4 characterized by different gradients above this SiO₂ content (Fig. 2c). Based on these observations, two hypothetical controls on magma series evolution can be proposed, where: (1) the progressive decrease in gradient exhibited by Aso 1-4, and hence the shift in dominant magma series from tholeiitic to calc-alkaline (model A, Fig. 2b), is the result of shallow-level magmatic differentiation processes in a single intracrustal magma chamber, reflecting differences in the amounts of crustal assimilation and/or changes in the physical conditions [e.g. changes in pressure, f(O₂)] over time; and/or (2) the fanning of the data at ~62 wt % SiO₂ (model B, Fig. 2c) has been produced by a change in the controlling fractionating mineral assemblage (e.g. from plagioclase + pyroxene to plagioclase + pyroxene + amphibole) between (and/or within) the four eruptive units, causing a shift in magma series with time. This shift in the controlling fractionating mineral assemblage could potentially reflect a two-stage magma chamber system, with magma in the deeper chamber subjectto plagioclase + pyroxene dominated fractionation, whereas in the shallower chamber, the physico-magmatic conditions controlling the system [e.g. pressure, P(H₂O), f(O₂,), temperature] have changed sufficiently to allow amphibole to join the dominant fractionating mineral assemblage. The different trends at SiO₂ >62 wt % exhibited by Aso 1-4 could therefore reflect different residence times (and the effects therein) within the upper and lower chambers.

If the tholeiitic and calc-alkaline magma series originated from a common parental magma, the least evolved samples from the four eruptive units should be isotopically homogeneous; this will be discussed later in a subsequent section.

Rock types

A total of 46 samples were selected from the APFD for petrological and geochemical analyses, with these encompassing the compositional and mineralogical ranges exhibited by each of the main sub-units from Aso 1 to 4. Samples with an `Aso' prefix were collected by the author (Hunter, 1993); all other samples were supplied by T. Koyaguchi (University of Kumamoto), along with a detailed description of their location, depositional form (e.g. welded ash, tuff, lava, etc.) and type of sample taken (e.g. whole rock, magmatic clast, glass shard, etc.)

All samples of Aso 1 are welded ignimbrites, whereas those from Aso 2-4 consist primarily of magmatic clasts (pumice or scoria clasts), in an uncompacted to welded ash-rich matrix. The welded ignimbrite samples from Aso 1 were checked carefully for signs of depositional mixing and sorting, and the presence of xenolithic or xenocrystic fragments, before analysis. Samples exhibiting any of these features were rejected from the study to ensure that all analysed samples represented as closely as possible the true, pre-eruptive magmatic signature of the different subunits. Because of the high degree of welding in Aso 1, it is conceivable that discrete amounts of xenolithic material may have been present in some of the analysed samples. Variations in mineralogical composition, phenocryst abundance and whole-rock textures of Aso 1-4 are listed in Table 5.

Mineral chemistry

To examine the chemical variations between the phenocryst phases in the different units, core and rim compositions were measured on multiple grains of all mineralogies of interest in each eruptive unit. This was primarily carried out to test how well the mineralogical classification used to define the tholeiitic and calc-alkaline series agreed with the geochemical classification. Throughout the APFD, the main phenocryst phases are plagioclase, clinopyroxene, orthopyroxene and titanomagnetite, with the relative abundance and mineral chemistry of these phases varying with whole-rock composition and magmatic series (Tables 5 and 6). In the more evolved samples from Aso 3, up to 5% hornblende may occur in the groundmass and as rims aroundorthopyroxene phenocrysts. Amphibole forms part of the main phenocryst assemblage in samples from Aso 4. The following sections briefly describe each of the main phenocryst phases found throughout the APFD.

Analytical conditions

All analyses were made at The Open University, UK, with whole-rock major and selected trace elementcompositions (Rb, Sr, Y, Zr, Nb, Ba) determinedby X-ray fluorescence (XRF) spectrometry on an ARL8420 + wavelength-dispersive spectrometer. Major element data were continuously assessed against internal standards, with the analytical precision for trace element analysis better than 2% (2[sgr]) and 5% (2[sgr]) at the 100 and 10 ppm levels, respectively. Rare earth element (REE) plus some trace element data (Th, Ta) were obtained by instrumental neutron activation analysis (INAA), following the procedure of Potts et al., (1981, , 1985). The data were continuously assessed, and have been adjusted against internal standards; reproducibility was better than 3·5% (2[sgr]; excluding Ta, for which 2[sgr] error was 5·6%). ⁸⁷Sr/⁸⁶Sr and ¹⁴³Nd/¹⁴⁴Nd analyses were performed on unleached whole-rock powders, using a Finnigan MAT 261 multi-collector mass spectrometer, with an appropriate international standard run with each batch of samples. Procedural blanks were analysed on a VG Isomass 54E solid source mass spectrometer. For this study, the Sr standard (NBS 987) gave a mean ⁸⁷Sr/⁸⁶Sr of 0·710257 ± 20, and the Nd standard [Johnson-Mattey (J&M)] gave a mean ¹⁴³Nd/¹⁴⁴Nd of 0·511890 ± 20. Major and trace element and isotopic analyses for samples from Aso are listed in Table 7.

Major element variations

The trends on the major element variation diagrams (Fig. 3) can be interpreted either in terms of individual eruptive units (e.g. Aso 1-4), or as one overall evolutionary trend for the caldera-forming stage. The simplest interpretation for the array of data (either as Aso 1-4 or the APFD) is that it is governed by fractional crystallization, the amount and controls on which vary with the evolution of each eruptive unit (i.e. increasing SiO₂ in Aso 1-4), as well as with time (i.e. petrochemical variations between the four eruptive units).

Figure 3. Major elements (TiO₂, Al₂O₃, Fe₂O₃, MgO, CaO, Na₂O, K₂O and P₂O₅ wt %) vs SiO₂ wt % at Aso. Symbols as in Fig. 2.

Taken as a single trend, the data for the APFD exhibit an inflection in slope at 61 wt % SiO₂ on several major oxide diagrams (e.g. TiO₂, Al₂O₃, P₂O₅; Fig. 3). TiO₂ increases up to 61 wt % SiO₂ (Aso 1 and 2 only), after which the concentration decreases, as a result of the onset of titanomagnetite fractionation. Because of the lack of data between 57 and 62 wt % SiO₂ for Aso 3 and 4, it is unclear whether these units have undergone a similar fractionation trend to Aso 1 and 2. However, if TiO₂ wt % is plotted against mg-number instead of SiO₂ wt % (Fig. 4), Aso 1 and 2 still exhibit an inflection in their trends at mg-number = 51, whereas in Aso 3 and 4, TiO₂ wt % decreases steadily with decreasing mg-number. Thus, the dominantly calc-alkaline trends of Aso 3 and 4 do not appear to follow the same fractionation path as the tholeiitic units, Aso 1 and 2.

Figure 4. mg-number vs TiO₂ wt % for Aso 1-4, illustrating the different evolutionary trends undergone by the tholeiitic and calc-alkaline series.

The tholeiitic series (Aso 1 and 2) clearly exhibits inflections at 61 wt % SiO₂ in its trend for Al₂O₃ and Fe₂O₃ (Fig. 3), whereas the calc-alkaline series (i.e. Aso 3 and 4) is characterized by linear decreasing trends against increasing SiO₂ wt % for these oxides. In MgO, CaO, Na₂O and K₂O vs SiO₂ wt %, both magmatic series exhibit similar linear trends, with MgO and CaO decreasing whereas Na₂O and K₂O increase with increasing SiO₂ wt % (Fig. 3). In general, the trends exhibited by the major oxides can be explained in terms of continued fractionation of varying amounts of plagioclase + pyroxene ± titanomagnetite ± amphibole in Aso 1-4, whereas the inflection in the P₂O₅ trend (Fig. 3) is the result of apatite fractionation.

In addition to the effects of fractional crystallization, the decrease in concentration of K₂O wt % (± TiO₂ and Fe₂O₃) at similar levels of magmatic differentiation (i.e. constant SiO₂ values) between Aso 1 and 4 (Fig. 3) suggests that other intracrustal processes may have also affected the evolutionary path of the APFD over time.

Trace element variations

Separate trends for Aso 1 and 2 and Aso 3 and 4 are also apparent in trace element vs SiO₂ diagrams (Fig. 5), with element concentrations either decreasing (e.g. Rb, Zr and Nb) or increasing (e.g. Sr) above 60 wt % SiO₂ between the eruptive units, Aso 1-4.

Figure 5. Trace elements (Rb, Sr, Ba, Zr, Y and Nb ppm) vs SiO₂ wt % at Aso. Symbols as in Fig. 2.

In general, Aso 1 and 2 (dominantly tholeiitic) are relatively enriched in all alkali earth elements (except Sr) at similar SiO₂ values, compared with the dominantly calc-alkaline eruptive units (Aso 3 and 4; Fig. 5). Rb and Ba concentrations increase with increasing SiO₂ in all four eruptive units, whereas Sr decreases with increasing SiO₂ (Fig. 5). The difference in slope in Rb, Ba and Sr abundances in Aso 1-4 implies either that the predominantly tholeiitic units (Aso 1 and 2) are characterized by different D_x values compared with the calc-alkaline units (Aso 3 and 4), or that this is a reflection of different mechanisms of silica enrichment with constant D_x values. All high field strength elements (HFSE, e.g. Zr, Nb, Y; Fig. 5) increase in concentration with increasing SiO₂, with the trends for both series characterized by similar gradients. The strong inflection exhibited in Y ppm at ~61 wt % SiO₂ in the tholeiitic and calc-alkaline series can be attributed to apatite (± clinopyroxene) fractionation, in which Y is strongly compatible.

Rare earth element variations

All REE data have been normalized to C1 chondrite abundances, and are plotted in Fig. 6. Comparisons between Jakes & Gill's, (1970) standard island arc tholeiitic REE pattern (shaded area) and the least evolved samples from Aso 1-4 indicate that Aso volcanic rocks are more enriched in light REE (LREE) and medium REE (MREE) than typical island arc tholeiites (Fig. 6). When compared with each other, the REE trends for tholeiitic and calc-alkaline samples from Aso 1-4 are relatively homogeneous, suggestive of a common parental magma for the two magmatic series. Variations in the relative concentrations of the REEs between the four eruptive units (i.e. as the caldera-forming stage evolved), as well as between the APFD and standard island arc tholeiitic basalt (Jakes & Gill, 1970) can be attributed to simple fractionation processes, which have produced an increase in abundance in all REEs at Aso.

Figure 6. REE profiles for the least evolved tholeiitic and calc-alkaline samples from Aso 1-4. Shaded field represents typical island arc tholeiite (Jakes & Gill, 1970). All samples have been normalized to C1 chondrite (Sun & McDonough, 1989).

Isotopic variations

At Aso, ⁸⁷Sr/⁸⁶Sr ranges from 0·704020 to 0·704466, with the majority of samples lying between 0·70404 and 0·70420 (Fig. 7a, b). The four samples from Aso 1 with anomalously high ⁸⁷Sr/⁸⁶Sr ratios (i.e. >0·7043) are also variably enriched in Sr ppm but depleted in K₂O wt %, Rb and Ba ppm, compared with expected major and trace element abundances (see Figs 3 and 5 and Table 7 for comparisons). Such shifts in geochemical signature are characteristic of post-depositional alteration (e.g. weathering) and as such, these samples can be excluded from the ensuing petrogenetic models. Disregarding the four isotopically altered samples from Aso 1, this unit still exhibits the greatest range in ⁸⁷Sr/⁸⁶Sr (0·704096-0·704219), extending well outside the 2[sgr] error, whereas in each of the subsequent eruptive units (Aso 2-4), ⁸⁷Sr/⁸⁶Sr is more constant, varying by 0·000057-0·000095. [As with Aso 1, the anomalously high ⁸⁷Sr/⁸⁶Sr (accompanied by shifts in K₂O wt %, Rb, Ba and Sr ppm) exhibited by one sample from Aso 4 can also be attributed to post-depositional alteration.]

Figure 7. (a) ⁸⁷Sr/⁸⁶Sr vs SiO₂ wt %; (b) ⁸⁷Sr/⁸⁶Sr against time, illustrating the general small decrease in isotopic ratio from Aso 1 to 4. Error bar in (a) and (b) represents a 2[sgr] error of ±0·00002. (c) ¹⁴³Nd/¹⁴⁴Nd vs SiO₂ wt %. (d) ¹⁴³Nd/¹⁴⁴Nd against time. There is no apparent correlation between changes in the Nd-isotopic ratio and magmatic series or time. Error bar represents a 2[sgr] error of ±0·00001. Symbols for (a)-(d) as in Fig. 2.

Therefore, excluding the five altered samples from Aso 1 and 4, the young age of the rocks and short time period involved between each eruptive unit, prohibits the observed range in ⁸⁷Sr/⁸⁶Sr from being attributed to radiogenic decay. Similarly, the time scale involved also prevents the variation in ⁸⁷Sr/⁸⁶Sr from being attributed to mixing between different magmatic batches that have evolved in a closed system from a common parental source, as these would be isotopically homogeneous. Therefore, the small variation in ⁸⁷Sr/⁸⁶Sr from 0·70402 to 0·70420 must be attributable to contamination of the magma source by an isotopically distinct material (e.g. crust), with the small but successive decrease in average ⁸⁷Sr/⁸⁶Sr from Aso 1 to 4 (Fig. 7b), indicating that the amount of contamination that occurred in the chamber decreased progressively with time.

Statistically, the isotopic variation exhibited by Aso 2 and 3 (±0·000027) is only slightly greater than the 2[sgr] analytical uncertainty (±0·000020), and the range in ⁸⁷Sr/⁸⁶Sr in Aso 4 is almost equivalent to the 2[sgr] error. As such, any petrogenetic conclusions based on isotopic variations exhibited by these units should be interpreted with caution.

¹⁴³Nd/¹⁴⁴Nd exhibits no systematic variation with whole-rock composition (Fig. 7c), magmatic series or time ( Fig. 7d), with values falling between 0·512778 and 0·512694. In each of the four eruptive units, the variations in ¹⁴³Nd/¹⁴⁴Nd extend outside the 2[sgr] error range (±0·000014), with the tholeiitic units (Aso 1 and 2) exhibiting slightly greater variation (0·512778-0·512694) than the calc-alkaline units (Aso 3 and 4; 0·512777-0·512717). Although ¹⁴³Nd/¹⁴⁴Nd varies erratically with time (Fig. 7d), a subtle increase in the average isotopic ratio (>2[sgr] error) is discernible for the APFD with time.

When trends for ⁸⁷Sr/⁸⁶Sr and ¹⁴³Nd/¹⁴⁴Nd against time are compared (Fig. 7b and d), it can be seen that the small but constant decrease in average ⁸⁷Sr/⁸⁶Sr is mirrored by a similarly small increase in the average Nd-isotopic ratio. This suggests that both isotopic systems may have been affected in similar ways by the same intracrustal processes throughout the evolution of the APFD, i.e. the erratic changing pattern and subtle increase in ¹⁴³Nd/¹⁴⁴Nd from Aso 1 to 4 can be attributed to varying amounts of crustal contamination.

On a plot of ¹⁴³Nd/¹⁴⁴Nd vs ⁸⁷Sr/⁸⁶Sr (Fig. 8a), all four eruptive units in the APFD appear to originate from an isotopically similar starting composition, with ⁸⁷Sr/⁸⁶Sr ~0·704080 and ¹⁴³Nd/¹⁴⁴Nd ~0·512768. When the trends for Aso 1-4 are examined closely in Fig. 8a, a progressive increase in the gradient for data from each eruptive unit is clear, with this trending towards infinity in Aso 4, with the total isotopic range exhibited by ¹⁴³Nd/¹⁴⁴Nd and ⁸⁷Sr/⁸⁶Sr also decreasing with time. Within Aso 1, the effects of post-depositional alteration are apparent, shifting ⁸⁷Sr/⁸⁶Sr to higher than expected values, compared with other samples at similar low ¹⁴³Nd/¹⁴⁴Nd values.

Figure 8. (a) ¹⁴³Nd/¹⁴⁴Nd vs ⁸⁷Sr/⁸⁶Sr for Aso 1-4, illustrating the progressive increase in gradient with time. Error bar represents a 2[sgr] error of ±0·00002 for ⁸⁷Sr/⁸⁶Sr and ±0·00001 for ¹⁴³Nd/¹⁴⁴Nd. (b) ¹⁴³Nd/¹⁴⁴Nd vs ⁸⁷Sr/⁸⁶Sr for Aso 1-4 compared with a general array for the whole of Japan. [Field of mid-ocean ridge basalt (MORB) shown for reference. The data for the Japan and MORB fields are from Faure, (1986) and references therein.] Error bar represents a 2[sgr] error of ±0·00002 for ⁸⁷Sr/⁸⁶Sr and ±0·00001 for ¹⁴³Nd/¹⁴⁴Nd. Symbols as in Fig. 2.

Comparison between the APFD and the whole of Japan in a plot of ¹⁴³Nd/¹⁴⁴Nd vs ⁸⁷Sr/⁸⁶Sr (Fig. 8b) reveals that Aso 1-4 fall at the radiogenic end of the Japanese array, close to Bulk Earth values. Although Aso 1-4 appear to be characterized by a steeper gradient than the whole of Japan, trending towards lower ¹⁴³Nd/¹⁴⁴Nd than expected (Fig. 8b), when compared with other volcanic systems in Japan, data from each volcano form a similar trend to that exhibited by Aso, producing a series of oblique trends down the mantle array. As before, the effect of post-depositional alteration on some samples from Aso 1 is clear, producing a shift towards higher than expected ⁸⁷Sr/⁸⁶Sr at a given ¹⁴³Nd/¹⁴⁴Nd value, compared with the general Aso array.

Mineral-whole-rock equilibrium relationships

Mineral-whole-rock compositional relationships can be used to establish whether a magma evolved in a closed or open system. In a closed system, the whole-rock compositions will be in equilibrium with the phenocryst cores, whereas the groundmass melt will be in equilibrium with the rims. (In this study, phenocryst rim-whole-rock groundmass equilibrium has not been tested for, as the glass was generally too microlitic to be analysed by electron microprobe techniques.)

Equilibrium between the whole-rock composition and phenocryst cores can be tested using diagrams of FeO/MgO_px vs FeO/MgO_WR for pyroxene (Fig. 9), and CaO/Na₂O_plag vs CaO/Na₂O_WR for plagioclase (Fig. 10). If equilibrium exists, the observed relationships should be consistent with experimentally determined equilibrium exchange coefficients, i.e. K_D values (Grove et al., 1982; Grove & Bryan, 1983; Kay & Kay, 1985; Sisson & Grove, 1993). [In Fig. 9, to compare iron from the pyroxenes and whole rocks in a similar state, whole-rock FeO*/MgO ratios were determined after the Fe₂O₃ had been recalculated as a ratio of Fe₂O₃:FeO (with this varying according to whole-rock chemistry from 0·9Fe³⁺:0·1Fe²⁺ for basalts, to 0·8Fe³⁺:0·2Fe²⁺ for dacitic to rhyolitic samples). This ratio of Fe³⁺:Fe²⁺ was then converted to ferrous iron oxide (i.e. FeO* wt %) by multiplying the ferric iron component by 0·9.]

Figure 9. (a) FeO/MgO_px vs whole-rock FeO/MgO_WR for (a) clinopyroxene (cpx) and (b) orthopyroxene (opx) compared with the expected equilibrium K_D lines (Nakamura & Kushiro, 1970a, 1970b) for each phase. Analyses that fall below the lines are too Mg rich to coexist in equilibrium with the whole rock, whereas those that fall above the lines are too Fe rich. Symbols as in Fig. 2.

Figure 10. CaO/Na₂O_plag vs CaO/Na₂O_WR for plagioclase analyses. A range of equilibrium K_D values are shown for anhydrous (K_D = 1·1) to water-saturated (K_D = 5·5) melts (Sisson & Grove, 1993). Ranges in CaO/Na₂O_plag at constant whole-rock values indicate that mixing between differently evolved melts has occurred. Symbols as in Fig. 2.

Pyroxene-bulk-rock relationships

In the FeO/MgO_px vs FeO/MgO_WR figure (Fig. 9), a number of analyses fall above and below the K_D equilibrium lines for clinopyroxene and orthopyroxene, lying outside the 1[sgr] error range (±0·02). Analyses that fall above the equilibrium lines indicate that the pyroxene cores are too iron rich for the whole-rock composition, whereas analyses that plot below the line represent phenocryst cores that are too magnesian to coexist in equilibrium with the whole-rock liquid composition. In the latter case, the magnesian cores probably represent early formed phases that have remained in suspension and have failed to react with the surrounding liquid, i.e. they represent xenocrysts. If this is indeed the case, the pyroxene crystals should display various disequilibrium textures such as rim resorption, mantling by different or similar phases, normal, reverse or oscillatory zoning. Such textures are prevalent in samples from Aso.

Phenocryst core compositions that are too iron rich for the whole-rock liquid composition probably represent either phenocrysts that have cooled rapidly and failed to equilibrate with the surrounding liquid, or crystals that have originated from more evolved units in the system. If the latter case is correct, the phenocrysts will be characterized by reverse zoning ± rim resorption, textures which are readily found in all calc-alkaline samples from Aso. The simplest method of producing these disequilibrium assemblages is by mixing differently evolved magma batches in the chamber. The importance of magma mixing during the evolution of the APFD is further emphasized in Fig. 9 by the large ranges in core FeO*/MgO values measured in a selection of pyroxene phenocrysts from individual rocks.

Equilibrium between coexisting phenocrysts

Equilibrium between coexisting phenocrysts can be used to examine whether different phases in a sample are in equilibrium with each other, and thus formed under similar conditions before eruption. Microprobe analyses were carried out on the cores and rims of hypersthene, augite and pigeonite phenocrysts and microcrysts that occurred contiguously throughout the sample. Mg-Fe²⁺ partitioning between Ca-rich (augite) and Ca-poor(hypersthene ± pigeonite) pyroxenes was used to test for equilibrium between the phases, with the K_D^Mg-Fe calculated using Nakamura & Kushiro's, (1970a, 1970b) general equation (see Fig. 11 caption).

Figure 11. Correlation of X_Mg for contiguous clinopyroxene and orthopyroxene phenocryst and microcryst pairs in Aso 1-4. K_D values for the pyroxene pairs can be calculated using the equation K_D= X_Mg^Ca-poor 1-X_Mg^Ca-poor· 1-X_Mg^Ca-rich X_Mg^Ca-rich(Nakamura & Kushiro, 1970b), where X_Mg represents the molecular proportions of Mg/(Mg + Fe). By comparing these values with standard equilibrium K_D values (Nakamura & Kushiro, 1970b), it can be seen that the contiguous pairs in tholeiitic and calc-alkaline samples from Aso 1-4 are in a state of quasi-equilibrium with an apparent average K_D value of 0·94. Symbols as in Fig. 2.

At equilibrium, the expected K_D^Mg-Fevalues vary from 0·75 to 1·08 (Nakamura & Kushiro, 1970a, 1970b) for the magmatic temperature (i.e. 1250-950°C), calculated by two-pyroxene thermometry using the procedure of Lindsley, (1983) and f(O₂) ( >= -8·4) ranges exhibited by samples from Aso. [f(O₂) was calculated from Fe³⁺/Fe²⁺ relationships in pyroxenes, using the procedures of Wood & Banno, (1973) and Wells, (1977).] Table 8 indicates that within each eruptive unit, the orthopyroxene and clinopyroxene pairs are indeed in a state of quasi-equilibrium (K_D = 0·71-1·29), with a best fit least-squares regression line implying an average apparent K_D value of 0·94 ( Fig. 11). This average value falls in the range proposed by Nakamura & Kushiro, (1970b) for equilibrium relations between high- and low-Ca pyroxene phases.

Table 8. Comparison of K_D^Mg-Fe for orthopyroxene-clinopyroxene pairs at Aso.

Combined plagioclase and pyroxene phenocryst relationships

Sakuyama, (1981) was the first to examine the relationship between plagioclase and mafic phenocryst phases within volcanic rocks, correlating compositional ranges and zoning patterns of plagioclase with the associated mafic phenocrysts. He discovered two groups consisting of: (1) plagioclase with a narrow compositional range, coexisting with normally zoned pyroxene phenocrysts (referred to as N-type), where pigeonite was the only Ca-poor groundmass pyroxene (i.e. tholeiitic), and (2) plagioclase with widely variable compositions, associated with reversely ± normally zoned pyroxene phenocrysts (R-type), with orthopyroxene as the predominant Ca-poor groundmass pyroxene phase (i.e. calc-alkaline). The two rock series were interpreted by Sakuyama as corresponding to magmas that had undergone mixing (R-type), or had not (N-type).

All calc-alkaline samples (excluding Aso 1.1B) from the APFD can be classified as R-type magmas and are characterized by a range of mixing textures, including: (1) `crypto-mixing' (i.e. the rock appears virtually homogeneous, with mixing visible only in the groundmass because of differences in texture or density of glass), and (2) bulk mixing, producing streaky grey and white pumice clasts. (The discrepancy in petrological versus geochemical classification of Aso 1.1B may be the result of syn-eruption mixing processes.) Although many of the tholeiitic samples correspond to Sakuyama's N-type magmas, some samples which were previously categorized as tholeiitic by other mineralogical and geochemical methods contain both N- and R-type phenocryst zonations, requiring a third group to be added to Sakuyama's classification scheme. This group (Table 6: Class 2) consists of plagioclase phenocrysts that exhibit a narrow compositional range, and are associated with reversely (± normally) zoned pyroxene phenocrysts, i.e. a transitional tholeiitic series. Mixing in these tholeiitic samples is limited to bulk mixing on a microscopic to macroscopic scale, forming streaky pumice.

The necessity for this additional group can be explained by the range of zonation relationships exhibited by coexisting plagioclase and pyroxene phenocrysts. The microscopic to macroscopic mixing textures in the streaky pumice can be produced by mixing two compositionally distinct tholeiitic magmas, which will result in the formation of R-type zonations although retaining an overall geochemical tholeiitic signature. Alternatively, rock samples that are characterized by bimodal plagioclase core compositions imply that `crypto-mixing' of two magmas (usually categorized as tholeiitic by geochemical means) has occurred. This produces an apparently homogeneous whole rock, containing two groups of compositionally heterogeneous phenocrysts, producing a transitional tholeiitic series rock according to petrochemical observations.

Thus, examination of the mineral chemistry and textural relationships between coexisting phenocrysts and whole rocks at Aso indicates that the system has undergone periods of magma mixing. This has resulted in quasi-equilibrium to disequilibrium relationships between whole-rock-phenocryst and phenocryst-phenocryst compositions. Although the tholeiitic and calc-alkaline series at Aso cannot be distinguished by differences in their constituent mineral compositions (as these overlap), the two series can be distinguished by variations in the compositional range and zoning patterns in the plagioclase and associated mafic phenocrysts (Sakuyama, 1981). This divides all samples into three interrelated groups-tholeiitic (Aso 1, 2 and 3), transitional tholeiitic (Aso 1 and 3) and calc-alkaline (Aso 1 and 4).

Fractional crystallization and combined crustal assimilation (AFC)

Least-squares fractional crystallization modelling has been carried out for Aso 1-4 using a fractionatingassemblage of plagioclase ± clinopyroxene ±orthopyroxene ± hornblende ± titanomagnetite (Table 9a). The actual assemblage used varied with whole-rock SiO₂ content within the four eruptive units, as well as with time between the eruptive units.

Table 9a. Comparison of the mineral assemblages, F values and [Sigma]r² used to model variations in major and trace element abundances in Aso 1-4 by fractional crystallization.

Table 9b. Comparison of the mineral assemblages, F values, r values and percentage of crust used to model variations in trace element abundances and isotopic ratios in Aso 1-4 by combined assimilation and fractional crystallization (AFC).

A plausible solution for each of the four eruptive units in the APFD was obtained based on major element modelling, with the extent of fractionation varying between 19·5 and 47·5% [i.e. F = 0·805-0·525, where [Sigma]r² = 0·06-0·09 (Table 9a)]. Within each of the eruptive units, the fractional crystallization path was modelled in two steps breaking at ~61 wt % SiO₂. [In Aso 4, major element fractional crystallization was modelled in four steps because of the dearth of samples between 52 and 62 wt % SiO₂. The bulk of data between 62 and 69 wt % SiO₂ was modelled in two steps (similar to Aso 1-3), whereas the compositional gap between the least evolved sample 4KC-03 and the rest of the data was modelled in two steps using two intermediate values, by taking the average of 4KC-03 and 4KC-01 (to form Intermediate 1), and then the average of this value (Intermediate 1) and 4KC-01 (to form Intermediate 2).]

From Table 9a, the total amount of fractionation required to reproduce the four eruptive units can be calculated, varying from ~58 to 63% in Aso 1-3, increasing to ~69% in Aso 4. At the same time, the dominant fractionating assemblage changed from plagioclase + augite ± hypersthene in Aso 1 and 2 (i.e. the tholeiitic series) to plagioclase ± augite + hypersthene ± hornblende in Aso 3 and 4 (i.e. the calc-alkaline series). Therefore, based on least-squares major element modelling of the data from Aso 1-4, the magmatic evolution within each of the eruptive units is apparently controlled by fractional crystallization, whereas between the four eruptive units, changes to the fractionating assemblage have resulted in different evolutionary paths for the tholeiitic (Aso 1 and 2) and calc-alkaline (Aso 3 and 4) magma series.

Although the major element trends for each eruptive unit can be modelled by fractional crystallization (Table 9a), variation in trace element abundances in Aso 1 and 3 cannot be reproduced successfully by this method. Shifts in the trace element abundances (and isotopic ratios) in these two eruptive units require the addition of a crustal contaminant into the system along with fractional crystallization to reproduce the observed trends, i.e. combined assimilation and fractional crystallization is required (Table 9a and b).

Simultaneous assimilation and fractional crystallization (AFC; DePaolo, 1981) is an extremely useful quantitative model for testing whether various trace element concentrations (and isotopic ratios) in two magmas are related, solving the relationship between the assimilation rate-fractional crystallization rate, the fraction of magmatic liquid remaining and the bulk partition coefficient for any trace element, i.e.

(1)

where C_m^o, C_m and C_a represent the concentration of the trace element in the original magma, the contaminated magma and the assimilant, respectively; F represents the fraction of melt remaining; r is the assimilation ratio divided by the fractional crystallization rate; D is the bulk solid-liquid partition coefficient; and z = (r + D - 1)/(r - 1). Equation (1) is valid only where r [=/=] 1 (i.e. the assimilation rate does not equal the fractional crystallization rate) and the term D remains constant. r is the most important term in this equation as it determines the rate at which the contaminant is assimilated into the system, with the implication that the heat required to melt the assimilant is provided by the latent heat of crystallization. From this, models with r < 0·5 will be dominated by fractional crystallization.

To ensure that the resultant models are petrologically viable, two constraints were applied. First, all D values had to be positive (i.e. >0), falling within a particular range applicable to the known mineralogical assemblage in the system being examined. Second, the mineral proportions used in the model correspond to those observed from petrological analyses. Once an acceptable r value had been established, the amount of material assimilated (F_a) was estimated as a fraction of the initial magma mass (D'Orazio, 1993), i.e.

(2)

If the assumption that crustal contamination does influence the magmatic evolution of the APFD is correct, then the small variation observed in ⁸⁷Sr/⁸⁶Sr and ¹⁴³Nd/¹⁴⁴Nd in Aso 1-4 requires that the contaminant must either be isotopically similar to the parental magmatic signature and/or that the amount of material assimilated is minimal (i.e. r 0·5). Borehole studies carried out around the northern edge of Aso caldera (Ono et al., 1981), have shown that the basement consists of Cretaceous granites as well as small Miocene volcanic and plutonic complexes, typical of basement complexes throughout the HVZ in Kyushu and SW Honshu (Hashimoto, 1991). Work carried out by Ando et al., (1987) and Ando & Shibata, (1988) on a selection of Geological Survey of Japan (GSJ) reference standard samples from this region has shown that the basement in the HVZ has an ⁸⁷Sr/⁸⁶Sr signature of 0·7043-0·7055, and a more general estimate of ⁸⁷Sr/⁸⁶Sr <0·7060 for the bulk crust in Central Kyushu has been established by Hashimoto, (1991). From this, it can be stated that in general terms the basement in the HVZ is isotopically similar to the APFD.

Using borehole data on the lithological character and composition of the basement in Central Kyushu (Ono et al., 1981) as a guide, the GSJ reference sample JG-3 [a granite from the HVZ, SW Honshu (Ando et al., 1987; Ando & Shibata, 1988)] was selected as representative of the composition of the crustal contaminant below Aso. Using the same mineral assemblages as for major element modelling, the trace element trends for Aso 1 and 3 were successfully reproduced by assimilating between 17·3 and 8·1 wt % granitic crust (r = 0·2-0·1, F = 0·525-0·707; Fig. 12). Within and between each of these eruptive units, the amount of contaminant required to reproduce the shift in trace elements and ⁸⁷Sr/⁸⁶Sr decreased with continued fractionation (Table 9a and b).

Figure 12. ⁸⁷Sr/⁸⁶Sr vs Rb/Sr. The observed trends displayed by Aso 1-4 have been reproduced using the AFC parameters in Table 9b, assimilating between 1 and 10% crust, represented by the GSJ reference sample JG-3. (The high ⁸⁷Sr/⁸⁶Sr ratios displayed by four samples from Aso 1 have been attributed to post-depositional alteration and, as such, these samples have been excluded from the AFC modelling.)

Although the variations in trace element concentrations in Aso 2 were reproduced successfully by fractional crystallization, a small amount of crustal material is still required to be added to the system during this stage, to replicate the observed shifts in ⁸⁷Sr/⁸⁶Sr and ¹⁴³Nd/¹⁴⁴Nd (Figs 7 and 8), which extend outside the 2[sgr] error range. Reproduction of the isotopic variations can be achieved by adding between 5·3 and 3·3 wt % crust (r = 0·1, F <= 0·7; Figs 7 and 12; Table 9b) to the magma, without significantly altering trace element concentrations from either the observed trend or that modelled by fractional crystallization. Similarly, the small variation in ⁸⁷Sr/⁸⁶Sr exhibited by samples from Aso 4 was reproduced by assimilating <2 wt % crust during any one evolutionary step (r = 0·05; F = 0·590-0·805), again without affecting the trace element concentrations significantly (Fig. 12; Table 9b).

Consequently, variations in trace element and isotopic ratios indicate that the amount of crustal material assimilated by the four successive eruptive units decreased with time. Decreasing amounts of crustal assimilation (Table 9b) associated with shorter residence times (Table 1) in later magmatic batches may have enhanced the observed progressive depletion of major [K₂O (± TiO₂, Fe₂O₃) wt %; Fig. 3] and trace elements (Rb, Ba, Zr and Nb ppm; Fig. 5) and changing isotopic ratios ( Figs 7 and 8) between the four eruptive units. This also corresponds to a decrease in the period of quiescence between eruptions from 200-100 kyr (Aso 1 and 2) to 20 kyr (Aso 3 and 4) (Aramaki, 1984), as well as a systematic increase in the volume of erupted material (Table 1) and the amount of fractional crystallization (Table 9a) that occurred from Aso 1 to 4. Thus during the caldera-forming stage, the influence of crustal contamination was diluted with time by increasing volumes of magma in the chamber and by decreasing magmatic residence times in the chamber.

Reduction in crustal assimilation over time may relate to changes in the physical conditions of the chamber and the surrounding country rock as the caldera-forming stage evolved. First, successive periods of fractional crystallization could potentially line the chamber walls and conduits with layers of crystals that either had grown in situ or had been deposited by gravitation settling. Over time, these cumulates would act as an insulating barrier between the magma and surrounding country rock, preventing and/or limiting further wall-rock assimilation. Second, continued interaction between the magmatic system and the country rock would result in the progressive removal of melts from the wall-rocks, decreasing its fertility and hence the amount of crust assimilated over time. The decrease in amount of contaminant assimilated over time complies with the observed decreasing shifts in isotopic and trace element concentrations in the APFD (Figs 5, 7 and 8; Table 7).

Therefore, the switch in dominant magmatic series from tholeiitic to calc-alkaline between Aso 1 and 2 and Aso 3 and 4 can be attributed to variations in (1) the controlling fractionation assemblage, (2) the amount of fractionation, (3) the effects of magma mixing between variably evolved melts, and (4) the amount and efficiency of crustal contamination over time. Of these four processes, variations in the fractionation assemblage and the extent of magma mixing have had the greatest control on determining the geochemical and petrological signature of the dominant magma series at any one time in the system.