YUKI SUZUKI 1 * AND SETSUYA NAKADA 2

Transcription

1 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 PAGES1543^ doi: /petrology/egm029 Remobilization of Highly Crystalline Felsic Magma by Injection of Mafic Magma: Constraints from the Middle Sixth Century Eruption at HarunaVolcano, Honshu, Japan YUKI SUZUKI 1 * AND SETSUYA NAKADA 2 1 INSTITUTE OF MINERALOGY, PETROLOGY AND ECONOMIC GEOLOGY, GRADUATE SCHOOL OF SCIENCE, TOHOKU UNIVERSITY, AOBA-KU, SENDAI, , JAPAN 2 EARTHQUAKE RESEARCH INSTITUTE, UNIVERSITY OF TOKYO, 1-1-1, YAYOI, BUNKYO-KU, TOKYO, , JAPAN RECEIVED JULY 21, 2006; ACCEPTED MAY 15, 2007 ADVANCE ACCESS PUBLICATION JUNE 22, 2007 The latest eruption of Haruna volcano at Futatsudake took place in the middle of the sixth century, starting with a Plinian fall, followed by pyroclastic flows, and ending with lava dome formation. Gray pumices found in the first Plinian phase (lower fall) and the dome lavas are the products of mixing between felsic (andesitic) magma having 50 vol. % phenocrysts and mafic magma.the mafic magma was aphyric in the initial phase, whereas it was relatively phyric during the final phase.the aphyric magma is chemically equivalent to the melt part of the phyric mafic magma and probably resulted from the separation of phenocrysts at their storage depth of 15 km. The major part of the felsic magma erupted as white pumice, without mixing and heating prior to the eruption, after the mixed magma (gray pumice) and heated felsic magma (white pumice) of the lower fall deposit. Although the mafic magma was injected into the felsic magma reservoir (at 7 km depth), part of the product (lower fall ejecta) preceded eruption of the felsic reservoir magma, as a consequence of upward dragging by the convecting reservoir of felsic magma. The mafic magma injection made the nearly rigid felsic magma erupt, letting low-viscosity mixed and heated magmas open the conduit and vent. Indeed the lower fall white pumices preserve a record of syneruptive slow ascent of magma to 2 km depth, probably associated with conduit formation. KEY WORDS: high-crystallinity felsic magma; magma plumbing system; multistage magma mixing; upward dragging of injected magma; vent opening by low-viscosity magma INTRODUCTION Knowledge of the triggering mechanisms of eruptions is important for the forecasting of future eruptions. The injection of new magma into a magma reservoir is one of the major mechanisms. Theoretical aspects of the triggering process include the increase in pressure as a result of (1) an increase in the magma reservoir resulting from the simple volumetric addition of injected magma (Blake, 1984), and (2) the intrinsic increase in volume owing to the exsolution of volatile phases from the melt. Exsolution can be generated either by cooling-induced crystallization of the injected high-temperature magma (Folch & Martí, 1998) or by heating of the low-temperature reservoir magma (Sparks et al., 1977). Furthermore, petrological studies on the ejecta have focused on magma mixing that results from the injection. To link the magma injection and eruption, some recent petrological studies have investigated the timescale from mixing to ejection (e.g. Nakamura, 1995; Venezky & Rutherford, 1999). Also, petrological studies have proposed a new triggering process for the case of highly crystalline felsic magma eruption: the injected low-viscosity mafic magma remobilizes the highly viscous felsic magma, reducing the magma viscosity as a result of mixing (e.g. Pallister et al., 1996; Venezky & Rutherford, 1997; Murphy et al., 2000; Takeuchi & Nakamura, 2001). *Corresponding author. Telphone: þ Fax: þ suzukiyk@ganko.tohoku.ac.jp ß The Author Published by Oxford University Press. All rights reserved. For Permissions, please journals.permissions@ oxfordjournals.org

2 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 1. Distribution of ejecta from the Futatsudake eruptions and earlier ejecta of Haruna volcano (after Oshima (1986) and Soda (1989)). Earlier ejecta include those of all stages (Stage 1^5) in Haruna activity. Inset shows the location of Haruna volcano in the NE Japan arc. Samples of the middle sixth century eruption were collected at Owazawa and Ohinata. Only fall deposits are observed at Ohinata. Py-flow, pyroclastic flow; VF (inset), volcanic front. Although petrological study of ejecta is a powerful tool for studying the triggering mechanisms of eruptions, it has certain drawbacks. There are only a few studies that consider the migration of magma around the reservoir (e.g. Venezky & Rutherford, 1997; Cottrell et al., 1999). Although studies of groundmass microlites can provide information about magma ascent at shallow levels in the conduit (e.g. Hammer et al., 1999; Nakada & Motomura, 1999; Suzuki et al., 2007), they do not address migration around the reservoir. When an eruption is triggered by magma injection, the migration of magma around the reservoir is associated with chemical modification of the magmas, and is thus important for understanding the whole picture of the eruption triggering processes. The most important factors that characterize and control the migration of the magma are the depth, shape and dimension of the magma storage system. However, it is sometimes difficult to resolve these storage systems, because of the still poor spatial resolution of geophysical imaging methods and incomplete knowledge of the endmember magmas that are mixed (especially the mafic magma, which comes from a deeper level than the felsic magma). The poor resolution of the magma storage systems makes it difficult to evaluate magma migration around the reservoir. To ascertain the locations of magma storage we need to determine the bulk and melt compositions of the end-member magmas, for example, using mass balance of phenocrysts for the mixing of phyric and aphyric magmas (Nakamura, 1995), and the compositions of melt inclusions (Cottrell et al., 1999). If magma storage conditions can be better ascertained, this would help in understanding the interaction of different magmas at the time of injection, through the estimation of their physical properties. At the same time, detailed study of phenocryst zoning could help us to reveal the progress of magma migration and mixing. If no modern observation record exists for a volcano, eruption triggering processes and the depths of magma storage reservoirs, which are obtained from petrological approaches, are particularly important in predicting the future eruptions based on geophysical observations. With this motivation and background, we have investigated the mid sixth century Futatsudake eruption of the Haruna volcano in central Honshu (Fig. 1). Although no historical documentation is available, the eruption sequence is well known (Oshima, 1983; Soda, 1989, 1996). In this study we show that a highly crystalline (50 vol. %) felsic (andesitic) magma and two mafic magmas (aphyric and phyric) were involved in the eruption. Although the mafic magmas did not erupt without mixing, we reveal that the aphyric mafic magma is chemically equivalent to the melt fraction of the phyric mafic magma and that the two magmas are from a related storage system. We propose that injection of the mafic magma into the felsic reservoir triggered the eruption by forming low-viscosity magmas, such as heated felsic magma and mixed magmas. Based on separate periods of mafic magma ejection (in the first and last phases of the eruption), we conclude that upward dragging of the injected magma can come to an end 1544

3 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN (see Snyder & Tait, 1996). With regard to magma migration around the reservoir, it is shown that the first erupted magma was associated with conduit formation and ascended slowly. We propose further, using data from detailed compositional zoning in phenocrysts, a multistage mixing model for the formation of homogeneous mixed magma with a minor contribution from mafic magma. BACKGROUND OF FUTATSUDAKE VOLCANISM Haruna volcano is located in the southern end of the northeast Japan arc (Fig. 1). The activity of this volcano (4300 ka) can be divided into five stages (Oshima, 1983, 1986). The ejecta is andesitic during Stages 1 and 2, andesitic^dacitic in Stage 3, and dacitic during Stages 4 and 5. Based on the SiO 2^FeO * /MgO diagram (Miyashiro, 1974), Watanabe & Takahashi (1995) indicated that rocks from Stage 1 activity can be classified as belonging to the tholeiitic series whereas those from Stages 2 to 5 belong to the calc-alkali series. During the first two stages, the formation and destruction of the stratovolcano were repeated around the present summit (Fig. 1), forming the base of the present volcanic edifice. During Stage 3 magma intruded into the flank of the stratovolcano. Stage 4 is characterized by two caldera-forming eruptions ( years bp for the most recent; Oshima, 1986) at the present summit. During Stage 5, more than five lava domes (501km 3 ) were formed inside the caldera and on the eastern flank. The latest activity occurred at the place where the present Futatsudake lava domes are located (Fig. 1). Three eruptive phases have been identified archaeologically, during the fifth century and in the early and middle sixth century (Machida et al., 1984; Sakaguchi, 1986, 1993). Soda (1989) proposed that the early sixth century event started with a phreatomagmatic eruption, followed by the eruption of pyroclastic flows (Fig. 1; 01km 3 in total). The middle sixth century event started with a Plinian eruption, followed by eruption of pyroclastic flows, and ended with lava dome formation. The pyroclastic fall and flow deposits (without any erosion hiatus) are directly covered by stratified fine ash layers from phreatomagmatic eruptions at the time of lava dome emplacement (Soda, 1993, 1996). This indicates that the time elapsed between the first two phases and lava dome emplacement was short. The volumes of the fall and flow deposits for the middle sixth century event are 13km 3 and 03km 3, respectively (Soda,1996). SAMPLES Localities Pumice blocks of the Plinian and pyroclastic flow phases were sampled at Owazawa and Ohinata (Figs 1 and 2). The samples were classified as white pumice, gray pumice and banded pumice (a mixture of the magmas forming the white and gray pumices). In Owazawa, the pyroclastic fall deposit was divided into the lower, middle and upper fall deposits, each of which is distinctive in terms of the type and size of the pumice. The pyroclastic flow deposit has been divided into the lower flow, top 30^60 cm and top 0^30 cm. The lower fall unit includes gray and banded pumices along with white pumices. In Ohinata, the pyroclastic fall deposit has been divided into five units (Fig. 2). The gray and banded pumices are limited to the lowermost unit (unit 1; Fig. 2), similar to Owazawa. The samples from Ohinata were used only for bulk-rock analyses. The lava dome samples were collected from three peaks (Odake, Medake and Magodake). Petrography Crystals in pumice and dome lava include phenocryst phases and microlites in the groundmass. The phenocryst phase can be classified as phenocryst (4300 mm across) and microphenocryst (100^300 mm across) (Fig. 3). The size difference between microphenocrysts and microlites is evident only in the white pumices and white parts of banded pumice (microlites of 520 mm across). Amphibole exists, but not as a microlite (Table 1). The common phenocryst phases for all ejecta are plagioclase (52 mm), orthopyroxene (52 mm), amphibole (55 mm), magnetite (51mm) and ilmenite (5700 mm) (Table 1; Fig. 3). Sometimes they form multi-phase aggregates (Fig. 3c), and contain inclusions of other minerals and glass. Olivine occurs as a phenocryst (52 mm) and microphenocryst in the dome lava (Table 1; Fig. 4), and is found as a microphenocryst (100 mm across) in the gray pumice and the gray part of the banded pumice (Table 1; Fig. 4). Clinopyroxene is found as a phenocryst (5500 mm; Fig. 3d) and as a microphenocryst in the dome lava, and as a microphenocryst (5300 mm; Fig. 3b) in the white pumice and the white part of the banded pumice in the lower fall deposit (Table 1). Olivine and clinopyroxene contain inclusions of Fe^Ti oxides. Although phenocrysts and microphenocrysts are commonly euhedral and free from reaction rims, this is not the case for the following phases (Table 1). Ilmenite shows resorption in gray pumice and the gray part of banded pumice. Olivine has an orthopyroxene reaction rim (Fig. 4). Some plagioclase phenocrysts and microphenocrysts in the dome lavas and the lower fall (white, gray and banded) pumices have dusty zones (100 mm wide). Some orthopyroxene grains in the dome lava, gray pumice and gray part of banded pumice have clinopyroxene reaction rims (20 mm). Some of the grains of amphibole are mantled by aggregations of pyroxene, plagioclase, Fe^Ti oxide and glass (up to 100 mm wide), as seen in pumices from the lower fall, upper fall and flow top 30^60 cm. All amphibole grains are mantled by similar rim intergrowths in the dome lava. 1545

4 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 2. Columnar sections showing eruptive sequence and bulk-rock SiO 2 contents of juvenile materials. Distance from Futatsudake is shown next to the site name. For the Owazawa site, SiO 2 contents for white pumices are shown with different symbols, depending on the number of pumices used for a single analysis. Fig. 3. Photomicrographs of representative ejecta from the middle sixth century eruption at Futatsudake (plane-polarized light). All scale bars represent 1mm. (a) Banded pumice from the lower fall deposit; (b) white pumice from the lower fall deposit; (c) white pumice from flow top 30^60 cm; (d) dome lava (Odake). In (a), the dark gray part (G) is adjacent to the white part (W), and the boundary between the two is sharp. Cpx, clinopyroxene; Am, amphibole; Opx, orthopyroxene; Pl, plagioclase; Ox, Fe^Ti oxide. (Note clinopyroxene with different sizes from microphenocrysts (MPh) in (b) and phenocrysts in (d).) 1546

5 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Table 1: Phenocryst and groundmass phases Phenocryst and microphenocryst Ol Cpx Am Opx Pl Mt Il Groundmass Dome lava P (Opx) P P (aggregate) P (Cpx*) P (dusty*) P P (exsolution) Gl þ Plþ Cpxþ Opxþ Ox Flow Top 0 30 cm P P P P P Gl Top cm P (aggregate*) P P P P Gl Pl Py Ox Lower P P P P P Gl Pl Py Fall Upper P (aggregate*) P P P P G þ Pl þ Py þ Ox Middle P P P P P Gl Pl Py Ox Lower (whitey) P MPh P (aggregate*) P P (dusty*) P P Gl Lower (grayy) P MPh(Opx) P (aggregate*) P (Cpx*) P (dusty*) P P (resorbed) Gl þ Pl þ Cpx þ Opx þ Ox Parentheses show reaction rim and disequilibrium texture; P, present;, absent; MPh, only microphenocryst; Ol, olivine; Cpx, clinopyroxene; Am, amphibole; Opx, orthopyroxene; Pl, plagioclase; Mt, magnetite; Il, ilmenite; Gl, glass; Py, pyroxene; Ox, Fe Ti oxides; aggregate, Pl þ Py þ Ox þ Gl; dusty, dusty zone. *Not found in all crystals. y Includes white and gray parts in banded pumices. Fig. 4. Backscattered electron image of typical olivine (Ol) in ejecta fromthe middle sixth century eruption at Futatsudake. Scale bars represent100 mm. (a) Phenocryst in dome lava (Odake); (b) microphenocryst olivine in gray pumice from the lower fall. (Note the size difference between (a) and (b), and euhedral outline of the crystal in (a) and (b) irrespective of the orthopyroxene (Opx) reaction rim (R.R.).) ANALYTICAL METHODS Major and trace elements in minerals and glass were analyzed with a JXA-8800R EPMA at the Earthquake Research Institute (ERI), University of Tokyo. The beam diameter was set focused for minerals and at 10 mm for glass. Major elements were analyzed using an accelerating voltage of 15 kvand a beam current of 12 na. The measuring time was 10 s for the peak and the two backgrounds, respectively, for every element. In glass analyses, Na, K and Si were measured in the first analytical cycle and Al in the second cycle, to minimize the evaporation of alkalis as a result of continuous beam exposure (Devine et al., 1995). For NiO in olivine, the accelerating voltage and counting time were set at 20 kv and 120 s, respectively. Microlites with a large enough size were analyzed by EPMA. At least, two samples of each type from the individual eruptive units were analyzed. Water contents in melt inclusions in phenocrysts were determined using the EPMA difference method (Devine et al., 1995; Morgan & London, 1996), using the JXM-8800 EPMA intohoku University. Hydrous glass standards were used as reference, together with the unknown. Following the optimal conditions for hydrous glass analyses (Morgan & London, 1996), the beam diameter was set at 15 mm, the beam current at 2 na and the measuring time at 30 s (for both the peak and the two backgrounds). The rest of the analytical conditions were similar to those used for the mineral and glass analyses. Vesiculated inclusions were not used for water content estimation. Whole-rock compositions were analyzed by X-ray fluorescence (XRF) (PW2400) at ERI, using glass beads with 1547

6 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Table 2: Representative whole-rock analyses for pumice and dome lava samples. All oxide values normalized to 100% and total iron as FeO Eruption phase: Lower-Fa Middle-Fa Upper-Fa FlT FlT 0 30 Lava dome Sample no.: Gray-1 Gray-4 Gray-5 Banded-3 Banded-6 White-2 White* White-1 y White-3 White-5 White-1 Odake-1 Medake-4 Magodake-2 Major elementsz (wt %) SiO TiO Al 2 O FeO MnO MgO CaO Na 2 O K 2 O P 2 O Trace elements (ppm) Nb Zr Y Sr Rb Zn Cu Ni Co Cr V Sc Ba All data for fall and flow deposits are from the Owazawa site (Fig. 2). (For eruption phases and pumice types (gray, banded, white), see text.) Fa, fall; FlT, flow top. *Mixture of several pumice fragments. y Used as a representative bulk composition of white pumice in Tables 5 and 7. z All oxide values are normalized to 100% and total iron is given as FeO. 10 parts flux to one part sample. When pumice grains are not large enough for X-ray analysis, multiple grains having similar characteristics and from the same horizon were used (Fig. 2). We could not separate each band of the banded pumices and hence obtained their bulk compositions (Table 2). Four to seven dome lava samples were analyzed for each peak of Futatsudake. ANALYTICAL RESULTS Whole-rock chemistry The SiO 2 content of the white pumice is nearly constant, independent of the eruptive unit (600^616; Fig. 2). Gray and banded pumices have lower SiO 2 content (575^604 wt %) than the white pumice. The SiO 2 content of the banded pumice (575^604 wt %) is similar to that of the gray pumice (579^585 wt %) or higher. The SiO 2 content of the dome lava (597^604 wt %) is lower than that of the white pumice, but partly overlaps with that of the banded pumice (Fig. 5). All samples show a linear trend in most Harker diagrams (Fig. 5). In other diagrams (e.g. MgO^SiO 2,K 2 O^Cr), however, the trend for gray and banded pumices is different from that of the dome lava. Mineral chemistry Compositions of the major phenocryst phases are reported in Electronic Appendices 1^6, available as Supplementary Data at In the 1548

7 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Fig. 5. Variation of SiO 2,TiO 2 and CaO vs MgO (wt %), and Zr and Cr (ppm) vs K 2 O (wt %) of whole-rock data. following discussion, Rim represents the outermost part of a crystal adjacent to the groundmass or reaction rim. The part inside the rim is referred to here as the core, but we introduce different terms for orthopyroxene and dusty plagioclase that shows complex zoning. Cores of olivine, clinopyroxene, magnetite and ilmenite are homogeneous except near the rims, so the core data are from crystal centers in the thin sections. Clinopyroxene Phenocrysts and microphenocrysts in the dome lava have cores of Mg-number 80^75 and Wo [Ca/ (Mg þ Fe þ Mn þ Ca)] 46^40, and rims of Mg-number 74^70 and Wo 45^38, showing normal zoning (Fig. 7). Clinopyroxene in the overgrowth rim on orthopyroxene phenocrysts in gray pumices and gray parts of banded pumices has Mg-number 71^74 and Wo 35^41. Olivine Olivine crystals all show normal zoning in Mg-number [100 Mg/(Mg þ Fe)] (Fig. 6a). The olivine microphenocrysts in the gray pumice and gray parts of banded pumice of the lower fall deposit have a core composition lower in Mg-number (78^76) than the phenocrysts and microphenocrysts in the dome lava (Fig. 6a). In the Mgnumber vs NiO diagram, the composition of olivine in dome lava overlaps with that of the olivine microphenocrysts from the lower fall deposit (Fig. 6b). Orthopyroxene Two types of orthopyroxene can be identified based on the zoning profile (Fig. 8). Type 1 has an Mg-number peak in the rim, whereas Type 2 has the peak inside the rim. Regardless of the type, each phenocryst has a homogeneous part ( core ) accounting for most of the area. The cores of both types are almost identical in composition. For Type 2, we define the part between the core and the rim as inner rim. Occurrence of Type 2 is limited to the lower fall pumice and the dome lava (Figs 8 and 9). 1549

8 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 6. Chemistry of olivine in gray pumice, gray part of banded pumice (BP), and dome lava. (a) Mg-number [Mg/(Mg þ Fe)] of cores vs Mg-number of rims; (b) Mg-number vs NiO. Legend in (a) also applies to (b). Data for (b) were obtainedthrough line-scan analyses for representative grains. Fig. 7. Variation of Mg-number in cores and rims of clinopyroxene phenocrysts and microphenocrysts in dome lavas. The core composition range (Mg-number 624^657 and Wo 08^18) is uniform throughout the eruption. Rims of Type 1 have Mg-number 653^675 and Mg-number 653^678 for the gray and white parts of pumices (including banded pumices), respectively, for lower fall, and Mg-number 660^732 for the dome lava. These values are higher than those for others (Mg-number 638^660). Inner rims of Type 2 have higher Mg-number than the rims of Type 1 in each eruption phase. Orthopyroxene microlites in the groundmass overlap chemically with the phenocryst rims (Fig. 9). Amphibole Most amphibole phenocrysts are hornblende (Fig. 10). They have oscillatory zoning in Na þ K (in A site) and Al(IV), both of which change in a sub-parallel fashion. The rims are relatively low in Na þ K (Fig. 10) and Al(IV). No systematic chemical changes were found between eruption units. Plagioclase For a dusty phenocryst, we define parts inside the dusty zone as core, and the parts between core and the rim as inner rim (Figs 11 and 12). Cores of all the plagioclase phenocrysts show oscillatory zoning, ranging from 55 to 90 in An mol % (100 Ca/(Ca þ Na)] (Figs 11 and 12). The FeO and MgO contents are relatively homogeneous with 02^04 wt % and 005 wt %, respectively, except near the rims of some clear phenocrysts in lavas and lower fall pumices. Analytical uncertainties (relative percent) in FeO and MgO are c. 20%at05 wt % and c. 40% at 005 wt %, respectively. The rims of clear phenocrysts vary in their chemistry within the range of their cores, except for the lower fall pumice and dome lava. Cores of both clear and dusty plagioclase phenocrysts in the lower fall pumice and dome lava mostly have lower FeO and MgO contents than in the rim and inner rim (Figs 11a, b and 12). Plagioclase microlites in the groundmass overlap chemically with the rim and inner rim of plagioclase phenocrysts (Fig. 12). The thickness of the final stage of growth is commonly larger in microphenocrysts than in phenocrysts because of the different surface areas. The wider domain makes it possible to determine precisely, by microprobe analysis, the compositions of the parts equilibrated just before the eruption. The An contents of microphenocryst rims are compared in Fig. 13. Microprobe backscattered electron images show that the microphenocryst rims are different chemically from the groundmass microlites in these 1550

9 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Fig. 8. Results of line-scan profiles across representative orthopyroxene phenocrysts. (For explanation of phenocryst types, see the text.) Cpx indicates reaction rim (no data shown). Although the core centers are not shown in these profiles, the compositions are almost uniform throughout the core parts. Fig. 9. Mg-number vs Wo [Ca/(Mg þ Fe þ Mn þ Ca)] mol% plot for orthopyroxene phenocryst (Ph) and groundmass microlites. (For explanation of types and definition of parts in phenocryst, see the text.) BP, banded pumice. 1551

10 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 10. Si vs Na þ K (in A site) diagrams for amphiboles in representative samples. Dotted line indicates the boundary between tschermakite (Ts) and hornblende (se nsu strı cto) (Hb), after Deer et al. (1992). Data for dome lavas do not include those of oxyhornblende. We analysed crystals without breakdown rims (Table 1) for samples excluding the dome lava. Rim composition is not available for the dome lava because of breakdown (Table 1). The compositional range in the middle fall deposit and lower flow is slightly wider than in the lower fall and is probably due to the larger size of the phenocrysts. There are no large phenocrysts in the lower fall deposit because of fragmentation. samples, even if microlites exist. Distribution of An content does not change significantly from sample to sample. FAL-2 and FAL-5 from the lower fall deposit have the lowest An contents for both maximum and minimum values of the whole distribution (An 50 and An 70, respectively). Magnetite The cores of magnetite phenocrysts are chemically homogeneous, with x-usp 017^020 (where x-usp is the fraction of the ulvo«spinel molecule) and Mg/Mn 3^8, except for some grains in the lower fall gray parts of the banded pumices with high x-usp (Fig. 14). Phenocryst rims show no difference in x-usp and Mg/Mn from their cores, except for the lower fall pumices and lavas. In the white parts of pumices (including banded pumices) of the lower fall and dome lavas, some of the phenocryst rims have higher x-usp, but have similar Mg/Mn, in comparison with the cores. In the gray pumice and gray part of the banded pumice, the rims of some phenocrysts have higher x-usp and Mg/Mn. The cores of the microphenocrysts seem to have crystallized simultaneously with the outer zones of the phenocrysts. For example, zoned margins (rim and near rim) of phenocrysts chemically resemble the cores of the microphenocrysts. Ilmenite Cores of ilmenite phenocrysts are chemically homogeneous, with x-ilm 055^060 (where x-ilm is the fraction of ilmentite in the rhombohedral phase) and Mg/Mn 6^9 (Fig. 15). The rims of some phenocrysts in the lower fall unit (white, gray and banded pumices) have higher x-ilm and Mg/Mn (11) than the cores. However, such an increase is not found in the other eruptive phases. The cores of the microphenocrysts are similar in composition to the phenocryst cores. Fe^Ti oxide thermometry and fo 2 barometry Because ilmenite is resorbed (Table 1), the ilmenite rims in the gray part of the pumice (including banded pumice) were not used. The equilibrium between the ilm-mt pairs used for calculation was checked in terms of Mg/Mn (Bacon & Hirshmann, 1988). The algorithm of Andersen & Lindsley (1988) yields temperatures of 820^8508C and log fo 2 of 103 to 109 for the core pairs from all samples and the rim pairs without compositional changes from the cores. The algorithm yields a temperature of 8808C and log fo 2 of 99 to 108 for the rim pairs from the white pumice fraction (including banded pumice) of the lower fall deposit. Groundmass composition We employed two methods to obtain the composition of the groundmass. For crystal-free groundmass samples (white pumice; Table 1) with relatively large glass domains (10 mm across) the analysis of the glass part was treated as the groundmass composition. For crystal-bearing groundmass (Table 1), we averaged analytical data for up to 200 point analyses. The latter method was limited to lavas (three representatives), as the highly vesiculated groundmass of pumice prevents random analyses. The estimated SiO 2 content of the groundmass for the white pumice is in the range of 774^784 wt % (Table 3); no systematic change between eruption units was found. 1552

11 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Fig. 11. Line profiles for representative plagioclase phenocrysts. Each line profile is shown by an arrow in the adjacent backscattered electron image. Contents of An, FeO and MgO are shown on the left and right vertical axes, respectively. Shaded area in (a) indicates the inner rim and rim of dusty plagioclase. Although the core centers are not shown in these profiles, the compositions are almost uniform throughout the core parts. (For explanation of phenocryst types (clear, dusty) and definition of parts in phenocrysts, see the text.) Pl, plagioclase; DZ, dusty zone. 1553

12 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 12. An (mol %) vs FeO and MgO (wt %) diagrams for plagioclase phases. FA, fall; LFA, lower fall; GP, gray pumice; WP, white pumice; BP, banded pumice; G, gray part; W, white part; FL, flow; DL, dome lava; Ph, phenocryst. (For explanation of phenocryst types (clear, dusty) and definition of parts in phenocrysts, see the text.) The groundmass of the dome lava is less fractionated (SiO 2 of 72^73 wt %) than that of the white pumice. Phenocryst phase volume For highly vesiculated samples it was necessary to section many samples to obtain statistically reasonable estimates of Fig. 13. An (mol %) histograms for plagioclase microphenocryst rims. The data in each histogram are from a single sample of white pumice. Samples with asterisks include plagioclase microlites in the groundmass. The dimensions of the vertical axes are the same in all the histograms. FA, fall; LFA, lower fall; FL, flow. However, the microprobe backscattering images show that the phenocryst rims are different chemically from groundmass microlites. the phenocryst mode. Accordingly, we employed chemical mass balance only for the white pumice (with high vesicularity) in all samples. The middle fall white pumice we used as representative (Table 5) because it is free from groundmass microlites, and its groundmass glass is considered equivalent to the melt part of the reservoir magma. The total volume of the phenocrysts (including microphenocrysts) decreases from the white pumice (c. 50%), to the dome lava (370^488%) and to the gray pumice (305^354%) (Tables 4 and 5). Ratios of individual phenocrysts to total phenocrysts barely change throughout the samples (e.g. plagioclase is consistently the most abundant phase). Olivine and clinopyroxene account for 06vol.% in total for the dome lava (Table 4). DISCUSSION Mixed and heated magmas at the beginning and end of the eruption Constraints from mineralogy The lower fall pumice and the dome lava have characteristics that cannot be explained by simple equilibrium crystallization from one to the other. It is important to note here that the gray and white parts of the banded pumice 1554

13 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Fig. 14. Histograms for x-usp and Mg/Mn contents of magnetite phenocrysts (Ph) and microphenocrysts (MPh). C, core; R, rim; FA, fall; LFA, lower fall; GP, gray pumice; WP, white pumice; BP, banded pumice; G, gray part; W, white part; FL, flow; N, number. (Note the different x-usp axis values for the dome lava.) are identical to the white pumice and gray pumice of the lower fall deposit. We stress the following properties: (1) coexistence of normally and reversely zoned mafic phases (olivine, clinopyroxene, and orthopyroxene) in the gray parts of the pumice (including banded pumice) and the dome lava (Figs 6^9); Fig. 15. Range of x-ilm compositions for ilmenite phenocrysts (Ph) and microphenocrysts (MPh). Data are not available for samples including dome lavas because of their exsolution (Table 1) and scarce amount of crystals. FA, fall; FL, flow.white part of pumice of the lower fall includes white pumice and white part of banded pumice. Similarly, gray part of pumice includes gray pumice and gray part of banded pumice. Data for banded pumice are connected with dotted lines. Table 3: Representative groundmass compositions White pumice Dome lava Lower fall Middle fall Flow top 0 30 cm Odake Magodake Sample GMS* Gl GMS* Gly GMS* Gl GMS GMS SiO (025)z 7835 (027) 7744 (038) TiO (003) 024 (003) 023 (005) Al 2 O (013) 1282 (021) 1288 (024) FeO* 171 (011) 159 (009) 158 (008) MnO 007 (004) 006 (003) 009 (002) MgO 045 (001) 038 (003) 042 (002) CaO 237 (002) 244 (010) 246 (007) Na 2 O 321 (010) 295 (013) 359 (008) K 2 O 131 (007) 118 (003) 132 (004) Original 9967 (064) 9657 (211) 9884 (151) total n All oxide values are normalized to 100% and total iron is given as FeO. n, number of analyses for glass; GMS, groundmass; Gl, glass. *Crystal-free groundmass. y Used in Tables 5 and 8. z Standard deviation. Not shown for dome lava, because data averaged to yield the composition in this table are from various GMS phases (see text for details). 1555

14 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Table 4: Phenocryst phase volume proportions in gray pumice and dome lavas Occurrence: Lower fall gray pumice Dome lava Sample: No. 1 No. 4 No. 5 Odake-1 Odake-6 Medake-3 Magodake-4 Bulk composition: 585, 36* 581, , , , , , 34 vol. % Olivine Clinopyroxene Plagioclase 258 (729) 231 (736) 218 (715) 332 (703) 399 (823) 275 (749) 363 (793) Amphibole 50 (141) 38 (121) 32 (105) 84 (177) 43 (88) 45 (124) 40 (87) Orthopyroxene 39 (110) 36 (115) 43 (141) 41 (86) 29 (60) 38 (104) 42 (92) Fe Ti oxides 07 (20) 09 (29) 12 (39) 16 (34) 14 (28) 08 (23) 13 (29) Total phenocryst Groundmass Phenocryst phases includes both phenocrysts and microphenocrysts. Volume percent values are based on point counts up to about Data in parentheses are volume percent plagioclase, amphibole, orthopyroxene and Fe Ti oxides, as a proportion of the total phenocryst population. *SiO 2 and MgO contents (wt %), in order. Table 5: Phenocryst content in white pumice Case A (Pl An 66 ) Case B (Pl An 88 ) wt % vol. % wt % vol. % Pl (2700) (719) (708) Am (3200) (185) (214) Opx (3500) (57) (41) Mt (5200) (31) (34) Il (4700) (08) (02) Total phenocryst GMS-Gl (2400) Mass-balance (mixing) calculations were used to estimate proportions of phenocryst phases and groundmass for white pumice (middle fall). Bulk-rock and groundmass compositions are listed in Tables 2 and 3, and the core compositions of phenocrysts used are the averaged values for all eruptive units except for An 66 and An 88 of plagioclase. Density values (kg/m 3 ) in parenthesis in the first column are from Deer et al. (1992) and Hall (1987). An, An mol %; Pl, plagioclase; Am, amphibole; Opx, orthopyroxene; Mt, magnetite; Il, ilmenite; GMS, groundmass; Gl, glass. Volume percents among phenocrysts are shown in parentheses in the forth and seventh columns. (2) reversely zoned orthopyroxenes in the white parts of the pumice (including the banded pumice) (Figs 8 and 9); (3) chemical disequilibrium between olivine and orthopyroxene in the gray parts of the pumice (including the banded pumice) and the dome lava (Fig. 16a); (4) enrichment in FeO and MgO at and near the rims of plagioclase (Figs 11a, b and 12); (5) Mg/Mn increasing at and near the rims of both magnetite (gray parts of pumice including banded pumice; Fig. 14) and ilmenite (gray and white parts of pumice including banded pumice). Compositional zoning, as in (1), (2), (4) and (5), was formed just before eruption, because it is found near the rims. The characteristics (2), (4) and (5) occur when crystals encounter less evolved melt. The most plausible process for explaining these characteristics would be magma mixing. The onset of amphibole and magnetite precipitation (Gill, 1981) or an increase in fo 2 may produce reverse zoning of mafic minerals. However, in such cases, the lines of evidence (4) and (5) favor magma mixing instead. FeO in plagioclase can increase as a result of an increase in fo 2 (Hattori & Sato, 1996). However, a simultaneous increase of MgO does not support a change in fo 2. Formation of a dusty zone in plagioclase requires an encounter with more calcic melt, an increase in temperature (e.g. Nakamura & Shimakita, 1998), or perhaps both, in accordance with the increases in FeO and MgO. The increase in Mg/Mn supports a temperature increase as a result of mixing with high-temperature mafic magma. Dissolution of ilmenite in the gray pumice and the gray part of banded pumice (Table 1) could be related to the mixing process. Normally zoned phenocrysts of olivine and clinopyroxene were derived from the high-temperature magma, whereas plagioclase and Fe^Ti oxides are from the low-temperature magma (Fig. 17). Microphenocryst olivines in the gray pumice and the gray part of the banded pumice are not 1556

15 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN Fig. 16. Mg^Fe distribution between cores of olivine and pyroxene. (a) olivine Mg-number vs orthopyroxene Mg-number; (b) olivine Mg-number vs clinopyroxene Mg-number. Data are plotted for both phenocrysts and microphenocrysts. Each diagram shows the average with the compositional range. Ranges of equilibrium were calculated from K D [(Fe/Mg)crystal/(Fe/Mg)melt] of 027^033 for olivine (Roeder & Emslie, 1970; Ulmer, 1989), 022^031 for clinopyroxene (Sisson & Grove, 1993a, 1993b) and 0245^0323 for orthopyroxene (Beattie, 1993). Fig. 17. Schematic diagram showing the range of magmas involved in the Futatsudake eruption, together with their phenocryst assemblages. White indicates the white part of pumice including banded pumice; Gray indicates the gray part of pumice including banded pumice; Ol, olivine; Cpx, clinopyroxene; Opx, orthopyroxene; Am, amphibole; Mt, magnetite; Il, ilmenite; Pl, plagioclase. crystals that nucleated upon and after mixing, because they have reaction rims of orthopyroxene. The fact that the microphenocryst olivine in the gray part of the pumice (100 mm) is much smaller in size than olivine in the dome lava (Fig. 4) may represent a shorter growth time prior to mixing; the mafic magma with olivine microphenocrysts (100 mm) having been aphyric until just before mixing. Amphibole is from the low-temperature magma, because of the association with aggregates of plagioclase and Fe^Ti oxides. However, their rims lack a record of heating, as indicated by increases in Na þ K and Al(IV) (Blundy & Holland, 1990; Scaillet & Evans, 1999; Rutherford & Devine, 2003), in the lower fall white pumice (Fig. 10). This may be due to the amphibole becoming unstable and no longer growing after the mixing event (Table 1). The above phenocryst assemblages in the end-member magmas are also supported by chemical equilibrium or phase stability between the phenocrysts. That is, olivine and clinopyroxene in the dome lava show equilibrium Mg^Fe distributions (Fig. 16b). The white pumice and the white part of the banded pumice from the lower fall deposit contains clinopyroxene microphenocrysts (Table 1) that probably nucleated during or after mixing, because they have no chemical zoning or textures indicative of mixing. Even in the gray pumice and the gray part of the banded pumice, plagioclase was not present in the high-temperature end-member (Fig. 17). If so, some plagioclase microphenocryst cores should have higher MgO and FeO than the plagioclase rims crystallized from the mixed magma (absent in Fig. 12). White pumice from the middle fall deposit to the flow top 0^30 cm probably maintains the original characteristics of the low-temperature end-member magma (Fig. 17), considering the phenocryst assemblage and the chemical compositions. Thus, the magma erupted in the middle stage represents the low-temperature end-member of the mixed magmas. In the pumice, a slight reverse zoning of orthopyroxene (Figs 8 and 9) is observed, whereas enrichments of FeO, MgO and Mg/Mn in the phenocryst rims of plagioclase and magnetite are not found (Figs 11, 12 and 14). The following two scenarios may explain this. The first is that the reverse zoning of orthopyroxene resulted from an increase in fo 2. The loss of H 2 gas resulting from devolatilization causes oxidation (e.g. Czamanske & Wones, 1973), which is likely to occur during syneruptive degassing processes. However, an increase of f O 2 should increase the FeO content of plagioclase (Hattori & Sato, 1996), which is not found in the present case (Figs 11 and 12). The other possibility may be that the reverse zoning in orthopyroxene records a previous heating event not shown by the magnetite because of rapid diffusion re-equilibration. In such a scenario, it is possible that reverse zoning of orthopyroxene was preserved, whereas plagioclase rims that once had been enriched in FeO and MgO disappeared because of diffusion. At the temperature of the felsic magma (8508C), the diffusion coefficient of Mg in plagioclase with An 70 [ ¼ m 2 /s, calculated using the data of Costa et al. (2003)] is found to be larger than 1557

16 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 the Fe 2þ^Mg interdiffusion coefficient in orthopyroxene ( m 2 /s; Tomiya & Takahashi, 2005). We shall defer discussion on the possible heating of the felsic magma erupted in the middle stages of the eruption until the section Possible pre-heating of the felsic reservoir magma. Constraints from groundmass and bulk-rock compositions The linear bulk-rock trends for the gray pumices and dome lavas (Fig. 5) support their origin as binary-mixing products. However, the trends for the lavas and gray pumices intersect on the high K 2 O side of the K 2 O^Cr diagram, indicating the same low-temperature (high K 2 O) mixing end-member but different high-temperature end-members (phyric and aphyric mafic magmas; Fig. 17). Although the white pumice is considered in terms of bulk chemistry to be the felsic end-member of mixing (Fig. 5), the white pumice of the lower fall deposit has some characteristics of a mixed magma; for example, the properties (2), (4) and (5) in the previous section. The white parts of the banded pumices have the same mineralogical characteristics as the lower fall white pumices. For these units, we suggest heating in a chemically closed system instead of mixing (Fig. 17). The heating can make the melt composition less evolved as a result of dissolution of earlier formed crystals, which leads to the formation of the crystal textures and compositional zoning observed in mixed magmas. Neighboring mixed magma (gray pumices and gray parts of banded pumices) or its mafic end-member should have heated the felsic magma. Irrespective of the heating that may result in the partial dissolution of phenocrysts, the white pumices from the lower fall deposits have similar groundmass compositions to the white pumices of the middle eruption (Table 3). This can be explained by the possibility that the lower fall felsic magma experienced slow syneruptive ascent and resultant crystallization at shallow level in the conduit (discussed below). In major element variation diagrams the banded pumices plot on a mixing line between white and gray pumices (Fig. 5). This is consistent with their mineralogical characteristics; the gray and white parts of the banded pumices resemble the white pumice and gray pumice of the lower fall, respectively, and the resultant banded pumice is equivalent to a mixture of the white and gray pumice. Heterogeneous reactions of phenocrysts upon heating and mixing in the lower fall deposits and dome lavas Patterns of compositional zoning and the degree of compositional change are variable in each product, and are even discernible at the hand-specimen scale. Examples are Types 1 and 2 orthopyroxenes (Figs 8 and 9), clear and dusty plagioclases (Figs 11 and 12) and magnetites with variable x-usp and Mg/Mn values (Fig. 14). These indicate heterogeneous physicochemical changes and/or variable times from such changes to the magma quenching. Because of the extent of magma mingling or mixing, this heterogeneity should be anticipated and it might reflect the proximity of the contrasting magmas. The coexistence of variable phenocrysts in a hand specimen implies effective stirring in each mixed and heated magma. Timescale from mixing and heating to final ejection for the lower fall deposit We provide a rough estimate of mixing timescales by using the zoning pattern of magnetites in the white pumices and white parts of banded pumices from the lower fall deposit. The zoning profile in the magnetite grains (200 mm across) would be homogenized in 15^12 years at 800^9008C (Tomiya & Takahashi, 2005). Magnetite microphenocrysts in the white pumices and in the white parts of the banded pumices from the lower fall deposit have relatively homogeneous cores (x-usp 017^020; Fig. 14), surrounded by reversely zoned rims that are as thin as 55 mm. The core compositions are similar to those of the white pumices from the middle eruption, whose magma represents the low-temperature end-member magma in the mixing (Fig. 17). The cores are believed to record the condition before the heating of the lower fall magma. The preservation of the core composition represents a timescale from heating (mixing) to eruption that is shorter than 15^20 years. This indicates that the heating and mixing are caused by an event unrelated to an eruption that took place in early sixth century at Haruna, 50 years before the eruption studied here. Devine et al. (2003) showed experimentally that magnetites in the Soufrie' re Hills magma became zoned within 2^10 days after temperature changed from 835^850 to 860^8808C. This condition may provide a minimum timescale for the white pumice and the white part of banded pumice. Characteristics of the mafic end-member magmas Chemical and genetic relationships between the mafic magmas The linear trend in Fo^NiO of olivines (Fig.6b) suggests that the aphyric and phyric mafic magmas (Fig. 17) have a close genetic relationship. However, the switch from aphyric to phyric magma during an eruption does not simply result from the temporal evolution of aphyric mafic magma as a result of crystallization. If so, small olivine crystals in the gray pumices and gray parts of banded pumices should have the same core composition as the phenocrysts in the dome lavas (not observed in Fig. 6a) and mixed magmas, formed from the common felsic end-member magma (Fig. 17), should have the same bulk-rock compositional trend (not observed in Fig. 5). It is instead considered that the aphyric mafic magma is chemically equivalent to the 1558

17 SUZUKI & NAKADA HARUNA ERUPTION, JAPAN melt part of the phyric mafic magma.the cores of the small olivine microphenocrysts in the gray pumice and the gray part of the banded pumice are chemically close to the rim of the olivine phenocrysts in the dome lava (Fig. 6a). This indicates that small olivines in the gray pumices and the gray parts of the banded pumices nucleated from a melt of similar composition to that which crystallized the rim of the phenocrysts. Bulk composition of the aphyric mafic magma The bulk chemistry of the aphyric mafic magma was estimated using phenocryst volumes and the major oxide compositions of the mixed magma (gray pumice) and felsic magma (white pumice). Assuming 0 vol. % phenocrysts in the aphyric magma, the estimated compositions are shown intable 6 (A and B; c.52wt%sio 2 ). The chemistry resembles that of basalt erupted during the early stage (Stage 1) of Haruna activity (Table 6), which has chemical characteristics common to this volcano. This estimation is supported by observed equilibrium between the melt (aphyric bulk) and the cores of olivine microphenocrysts in the gray pumice and the gray part of the banded pumice. K D for the Fe^Mg distribution [(Fe/Mg) Ol /(Fe/ Mg) melt ] is 031^036 for the two compositions of aphyric mafic magma (A and B in Table 6). We used 01 for Fe 3þ /(Fe 2þ þ Fe 3þ ) in the aphyric mafic magma at quartz^fayalite^magnetite (QFM) and nickel^nickel oxide (NNO) oxygen buffer conditions, based on estimations from Sack et al. (1980). For equilibrium olivine^melt pairs, the K D can be for a wide range of melt compositions, temperatures and water contents (e.g. Roeder & Emslie, 1970; Ulmer, 1989). Phenocryst content and bulk composition of the phyric mafic magma Mass-balance calculations for the formation of the dome lava through the addition of aphyric magma (melt) and olivine Table 7: Mass-balance calculation to form average dome lava composition Result of mass balance (wt %) and clinopyroxene phenocrysts (derived from the phyric mafic magma) to the white pumice magma yields their mixing ratios (Table 7). The ratios of felsic magma : aphyric mafic magma (melt) : olivine : clinopyroxene are about Table 6: Estimated bulk-rock compositions for mafic magmas (a) Aphyric (b) Phyric (c) Basalt in early Haruna activity A B A B Major elements* (wt %) SiO TiO Al 2 O FeO MnO MgO CaO Na 2 O K 2 O P 2 O Estimation methods are described in the text. (a) Averaged chemical composition of aphyric mafic magma was calculated by extension to 0% of phenocrysts from the white pumice (Middle-Fa White-1 in Table 2) with different phenocryst contents, A and B (495 and 510 vol. %) in Table 5, through three gray pumice samples (Gray-1, -4 and -5 in Table 2) with phenocryst contents in Table 4. (b) Chemistry of phyric mafic magma was estimated in the mass balance shown in Table 7. (c) Ejecta of Stage 1 (Oshima, 1983). *All oxide values are normalized to 100% and total iron is given as FeO. Vol. % of phases in phyric mafic magma F.M. Aphyric M.M. Crystal in phyric M.M. R 2 A B Ol Cpx Melt Ol Cpx Case A Case B Results of mass-balance calculations; mixed magma ¼ felsic magma (F.M.) þ phyric magma, where phyric magma ¼ aphyric mafic magma (M.M.) þ olivine (Ol) þ clinopyroxene (Cpx). The average composition of dome lavas and a representative white pumice sample, Middle-Fa White-1 (Table 2), were used as mixed and felsic magmas, respectively. Aphyric mafic magmas of A and B in Table 6 were tested (Cases A and B, respectively). Average compositions of olivine and clinopyroxene were used. Densities used for melt, olivine and clinopyroxene (in kg/m 3 ) were 2650, 3400 and 3300, respectively (Hall, 1987). 1559

18 JOURNAL OF PETROLOGY VOLUME 48 NUMBER 8 AUGUST 2007 Fig. 18. Mixed magmas (dome lava, gray pumice) and the mafic end-members of mixing in MgO^SiO 2 (wt %) and K 2 O(wt %) ^Cr (ppm) variation diagrams. Each regression line determined for the mixed magmas (gray pumices or dome lavas) and the felsic end-member (white pumice, not shown in this figure) connects mafic and felsic end-member magmas. Phyric mafic magma is compositionally equivalent to a mixture of aphyric mafic magma and phenocrysts (Table 7). (a) A and B are given intable 6. (b) Mafic magmas are plotted, based on estimated K 2 O contents (Table 6, average of A and B). 85^86:12^13:03:03 (inwt %).Theweight per cent ratios of olivine and clinopyroxene are roughly consistent with their phenocryst volumes in the dome lava (Table 4), when considering their relative densities. In the calculation, phenocrysts of felsic magma origin account for 42^44 vol. % of the mixture, consistent with the total phenocryst volumes in the dome lavas (37^48 vol. %; Table 4). The calculated proportion of phenocrysts in the phyric mafic magma is about 6 wt % (Table 7). These results allowed us to estimate the bulk composition of the phyric mafic magma (Table 6). Figure 18 shows the relationships between mixed magmas (gray and banded pumices, and dome lavas) and the mafic end-members (aphyric and phyric), in which the phyric and aphyric mafic magmas lie at the ends of the mixing lines. The enrichment of MgO and Cr content in the phyric mafic magma compared with the aphyric mafic magma can be explained by the concentration of olivine and clinopyroxene in the former, in both of which Cr is highly compatible (Gill, 1981). Storage conditions of the magmas Storage condition of the mafic magmas Because olivine and clinopyroxene were stable in the phyric mafic magma, the stability conditions were estimated using the MELTS algorithm (Ghiorso & Sack, 1995), using the composition of the melt fraction (aphyric mafic magma; Table 6) with different water contents. The estimated liquidus surface shown in Fig. 19 indicates near-liquidus coexistence of olivine and clinopyroxene at Fig. 19. Liquidus surface of aphyric magma (Table 6, A), estimated using MELTS (Ghiorso & Sack, 1995) as a function of pressure and H 2 O content. Redox conditions appropriate to QFM were assumed because the oxygen fugacity of typical island arc magma is in the range QFM^NNO þ 2 (Carmichael & Ghiorso, 1990). Although a decrease in fo 2 expands the field of olivine relative to orthopyroxene, the stabilities of clinopyroxene and plagioclase are independent of fo 2 (e.g. Berndt et al., 2005). Therefore, we may have overestimated the olivine stability field if the fo 2 is more oxidizing than assumed. Pl, plagioclase; Px, pyroxene; Cpx, clinopyroxene; Opx, orthopyroxene; Ol, olivine. The shaded zone shows olivine^clinopyroxene coexistence on the liquidus. 1560