Using high-resolution Pb isotopes to unravel the petrogenesis of Sakurajima volcano, Japan

Sakurajima volcano on the rim of Aira caldera erupts daily, threatening the major nearby population centre of Kagoshima. Before 1955, eruptions have typically consisted of intermittent Plinian and effusive activity, but since 1955, frequent Vulcanian explosions have occurred, indicating a change in pre-eruptive processes. High-resolution Pb isotopes are used here to constrain the components, including crustal assimilants, and petrogenetic processes contributing to the composition of both pre and post 1955 magmas. Sakurajima eruptive products have more radiogenic Pb (206Pb/204Pb = 18.40–18.42) than nearby Sumiyoshiike basalts (206Pb/204Pb = 18.24), a proxy for the primitive magma feeding the Sakurajima-Aira system. Sakurajima samples lie along a mixing line between these basalts and locally exposed crustal compositions. Their Pb isotopes are consistent with addition of ~ 5% average crustal melt to the primitive basaltic magma. The narrow range of Pb isotope ratios, despite variable SiO2 contents (56.6–72.3 wt%), suggests that the final erupted magmas are derived from fractional crystallization of a mafic precursory magma displaying relatively consistent levels of crustal contamination. Andesites erupted between 4–3.7 ka and the 1995 eruption are contaminated to a greater extent, indicating that magmas with distinct compositions can feed phases of activity or individual eruptions. Post 1955 andesitic pyroclastics have lower SiO2 and higher MgO than older lavas, yet equivalent Pb isotope ratios. The more mafic composition of post 1955 eruptive products can be attributed to increased throughput of mafic magma to the system.


Introduction
Subduction zone magmas develop by addition of material from the downgoing slab to the mantle wedge (e.g. Armstrong 1971;Hawkesworth et al. 1993;Elliott 2004;Straub et al. 2010), with the resulting melts interacting with crustal lithologies in the overriding plate before eruption along the arc (e.g. Harmon et al. 1984;Hildreth and Moorbath 1988;Davidson et al. 2005;Ducea et al. 2015). Quantifying the proportions and identifying the origin of these slab, mantle and crustal components is important in understanding their roles in arc magma genesis. Isotopic studies provide a method of constraining these proportions and are key to defining where interaction occurs. Broad isotopic compositions of slab, mantle and crust are recognized (e.g. Dupré and Allègre 1983;Taylor and Maclennan 1985;Asmerom and Jacobsen 1993;Plank and Langmuir 1998;Workman and Hart 2005;Straub et al. 2010), but subtle changes in radiogenic isotopes have the potential to provide new information on the development and dynamics of sub-arc magma reservoirs.
Editorial responsibility: A.V. Ivanov; Deputy Executive Editor: J. Tadeucci Electronic supplementary material The online version of this article (https://doi.org/10.1007/s00445-020-1371-0) contains supplementary material, which is available to authorized users. Arc magmas often show temporal variation in radiogenic isotopic composition, both along arc and at individual volcanic centres, which can be related to changing contributions of mantle, slab and crustal components in their petrogenesis (e.g. Elburg and Foden 1998;Gómez-Tuena et al. 2003;Hanyu et al. 2006;Labanieh et al. 2010;Straub et al. 2015;Ishizuka et al. 2015). Recent advances in measurement precision of Pb isotopes allow subtle differences to be determined (e.g. Abouchami et al. 2000;Cassidy et al. 2012;Taylor et al. 2015), offering a means of assessing temporal changes in magmatic processes in greater detail. Despite this, relatively few studies have used highresolution Pb isotopes in arcs to examine short-term processes (within a few thousand years and shorter timespans). Even minor variations in Pb isotopes can reveal the presence and interaction of compositionally distinct magmas beneath arc volcanoes over time (e.g. Cassidy et al. 2012;Kayzar et al. 2014;Ishizuka et al. 2015), which may correlate with surface activity. In this study, we utilize high-resolution Pb isotope analysis, complemented by major and trace element data, to probe short timescale changes in petrogenetic processes and magma-crust interaction at Sakurajima-Aira caldera volcano, of the Ryukyu arc in southern Japan.
Existing isotopic studies of Sakurajima-Aira and the Ryukyu arc suggest that erupted magmas are affected by crustal contamination (Hosono et al. 2008;Shibata et al. 2013a). However, the nature of the assimilant is poorly constrained at Sakurajima-Aira (Shibata et al. 2013a) as is the degree of assimilation and if/how this varies between eruptions. Majority of Sakurajima's eruptive history is characterized by intermittent Plinian and effusive eruptions, with a change in activity style to frequent Vulcanian explosions beginning in 1955 (Nakagawa et al. 2011;Iguchi 2013). This transition is suggestive of a change in processes in the magma plumbing system (Nakagawa et al. 2011).
Several studies have measured major and trace element, and Sr-Nd isotope compositions at Sakurajima, but few have examined in detail the Pb isotopic systematics. These were measured in 14 samples erupted between 29 ka and 2015, as well as for two nearby basalts. Pb isotope data for local crustal rocks is also presented to provide more accurate constraints on the components contributing to the isotopic signature of the samples. Although post 1955 eruptive products have been studied in terms of phenocryst composition (e.g. Nakagawa et al. 2011;Ebihara et al. 2013), there is little geochemical data available for this phase of eruptive activity. This is addressed here by the inclusion of four post 1955 samples. Comparison of the isotopic and elemental data from the post 1955 eruptions with the older volcanism is used for two aims: (1) to constrain crustal assimilation in terms of components, extent of contamination and temporal variation in these parameters and (2) to investigate whether petrogenetic processes occurring prior to more frequent post 1955 eruptions and older eruptions differ.

Geological setting and previous geochemical studies
Sakurajima is an active stratovolcano in southern Kyushu, Japan, located on the southern rim of the Aira caldera within the Kagoshima graben (Fukuyama 1978;Aramaki 1984). The Aira caldera forms part of the northern end of the Ryukyu arc, which exists due to the subduction of the Philippine Sea Plate (PSP) below the Eurasian Plate (Fukuyama 1978), as shown in Fig. 1. Volcanic activity has occurred in the region throughout the Quaternary, with the formation of Aira caldera in a cataclysmic eruption around 29 ka (Aramaki 1984). The Sakurajima edifice began to form around 26 ka (Takahashi et al. 2013) and has been through several phases of eruptive activity.
This study covers three phases of Sakurajima's eruptive history: the Younger Kitadake, Older Minamidake and Younger Minamidake stages. The Younger Kitadake dacite lavas were extruded around 10 ka (Takahashi et al. 2013). The andesitic Miyamoto, Kannonzaki and Arimura lavas were erupted during the Older Minamidake phase from 4 to 3.7 ka (Miki 1999). Notable eruptions in the Younger Minamidake phase include the 1471-1476 Bunmei and 1779-1782 An-ei dacitic lava flows. The most recent major eruption was the 1914-1915 Taisho event which produced two phases of andesitic lava flows. Secondary lava flows with a distinct chemical signature were derived from the flow fronts of the T1 and T2 lavas (Yanagi et al. 1991), forming the T1' and T2' lavas respectively. In 1946, the andesitic Showa lavas were erupted from the newly formed Showa crater (Takahashi et al. 2013). From 1955 to the present day, Vulcanian explosions have continued to occur at the Minamidake and Showa craters (Iguchi 2013). Shibata et al. (2013a) sampled 22 volcano-stratigraphic units from Sakurajima, along with nearby basalts. Their study found no significant differences in the petrology of the eruptive products of Sakurajima, other than compositional variation from andesites to dacites. These were summarized as porphyritic with hyalo-ophitic groundmasses, containing clinopyroxene, orthopyroxene, plagioclase and opaque phenocrysts. Olivine phenocrysts were present in the basalts.  Shibata et al. (2013a). b Sampling sites of basalts and Aira caldera rhyolite relative to Aira caldera and Sakurajima volcano. Also shown are the sampling sites for the crustal materials considered in this study (the Shimanto sediments and Takakumayama granite). Location of Shimanto sediment samples after Hosono et al. (2008). Location of Takakumayama granite after Ishihara and Chappell (2010). Map modified from Hosono et al. (2008). c Sampling sites of the eruptive products of Sakurajima volcano, superimposed on the geological map of the volcano published by Kobayashi et al. (2013). Lava flow units sampled and associated sample names are listed in the legend. Pyroclastic samples collected from eruption fallout at time of eruption Shibata et al. (2013a) measured Sr-Nd-Pb isotope ratios in eight samples from Sakurajima and two basalts. They found that the monogenetic maar volcano (Sumiyoshiike) basalts had Zr/Nb ratios close to those of MORB, and suggested that the mantle wedge below Sakurajima is of MORB-type mantle. Analysis of Sr, Nd and Pb isotopic compositions showed that the samples plot close to a mixing curve between MORB and sediments of the PSP. This led them to conclude that fluids are added to the regional mantle wedge from subducting PSP sediments. They inferred that the Sumiyoshiike basalts represent the primary magma of Sakurajima volcano because they plot almost on the MORB-PSP sediment mixing curve. The Sr-Nd-Pb isotopes of Sakurajima were found to lie off the mixing curve, displaying a more radiogenic signature. Shibata et al. (2013a) suggested this is due to the incorporation of crustal material, such as the locally exposed Shimanto sediments, into Sakurajima's magmas.
Medium-K volcanic rocks from across southern Kyushu form linear arrays between Sr-Nd and Pb-Pb isotopes (Hosono et al. 2008), indicative of mixing between a depleted, less radiogenic component (DC) and an enriched, more radiogenic component (EC). They proposed that Southern Kyushu volcanics acquire their isotopic signature via assimilation of both lower crustal material and Shimanto sediments (which together make up their enriched component) by basaltic mantle melts (the depleted component). They assume the lower crust to be of granodioritic composition based on seismic evidence from Iwasaki et al. (1990). Hosono et al. (2008) also suggest that silicic magmas of Southern Kyushu formed via fractional crystallization of the parent magma following assimilation. This is based on the relationship between isotopic and major element data, where silicic samples lie further away from the DC-EC mixing line. Takahashi et al. (2013) measured major and trace element compositions of rhyolites from the 29 ka Aira caldera eruption together with basaltic andesite mafic inclusions, and Sakurajima andesites and dacites. They observed a wide range of incompatible trace element ratios and suggested that these could be produced by the mixing of mafic magma, represented by the inclusions, and the 29 ka rhyolitic magma. Uto et al. (2005) showed that magmas erupted since 2 ka have 87 Sr/ 86 Sr, Nb/Th and Zr/Th ratios closer to basaltic inclusions than older eruptive products, indicating a recent trend towards more mafic compositions.
Examination of phenocrysts from the Bunmei, An-ei, Taisho and Showa eruptions by Yanagi et al. (1991) showed that the Taisho and Showa lavas contain plagioclase with bimodal An compositions. This led them to propose that prior to the 1914 Taisho and 1946 Showa eruptions, dacitic magma with low-An plagioclase was mixed with high-An plagioclase basaltic magma. Nakagawa et al. (2011) reported the presence of olivine phenocrysts (derived from basaltic magma) in post 1955 eruptive products. This led them to suggest that recent activity has been driven by the injection of basaltic magma, and that this is the cause of increased eruption frequency since 1955.

Methods
The eruptive products of Sakurajima are divided into andesites, dacites and post 1955 andesites, to distinguish the andesitic pyroclastics erupted since 1955 from the older andesitic and dacitic lava flows. Figure 1 shows the sampling sites for the eruptive products of Sakurajima. Locations of local crustal materials (Shimanto sediments and Takakumayama granite), regional basalts (Sumiyoshiike and Shirahama basalts, also used in the isotopic study of Shibata et al. 2013a) and the Aira caldera rhyolite are also shown in Fig. 1. Rock samples were crushed inside polyethylene bags using a fly press. The crushed material was then separated using a Teflon sieve to isolate 0.5-1-mm chips. This fraction was repeatedly rinsed with 18 MΩ water before cleaning in an ultrasonic bath for 10 min. Chips were examined under a binocular microscope to ensure only uncontaminated material progressed to analysis. All measurements were conducted at the School of Ocean and Earth Science, University of Southampton, UK.

Major elements
Selected rock chips were powdered in an agate pestle and mortar, placed in a crucible and dried overnight at 105°C. Loss on ignition was determined via heating in a furnace at 1000°C for 4 h. The cooled powders were then mixed with LiBO 4 (10:1 LiBO 4 :rock powder) and fused into glass beads, as per the methods of Croudace and Williams-Thorpe (1988). Whole rock major element compositions were measured using WD-XRF, using a Philips Mag-Ix-Pro XRF spectrometer calibrated using international rock standards. The uncertainty on the XRF measurements is less than 1% for all elements (see Appendix B for reproducibility using standard BRR1).

Trace elements
An aliquot of clean selected chips was powdered and digested with HNO 3 :HF. Daughter solutions were 4000:1 dilutions and were spiked with a Be-In-Re internal standard prior to measurement. Trace elements were measured by inductively coupled plasma mass spectrometry (ICP-MS) using a Thermo Scientific X-Series II. Measurements were corrected for interferences and blank, then calibrated using a suite of international rock standards (JB-3, JB-1a, JGb-1, BHVO-2, BIR-1, JA-2, reference data from GeoReM database, Jochum et al. 2005) and in-house reference materials. Longterm accuracy relative to reference values is 3-5%. Pb was separated in a class 100 clean laboratory in the University of Southampton using sub-boiled and ultra-pure Pb reagents. 0.4 g of rock chips was leached at 150°C in 4 M HCl for 30 min, then repeatedly rinsed in pure water. Leached chips were then dissolved for more than 24 h in approximately 4 ml of an HF:HNO 3 (4:1) mixture. Pb was then isolated from the matrix using two passes through AG1x8, 200-400 mesh anion exchange resin using 0.5 M HBr and recovery in 10 M HCl. Pb isotopes were measured using a Thermo Neptune multi-collector inductively coupled plasma mass spectrometry (MC-ICP-MS). Samples were corrected for instrumental mass fractionation using the SBL74 double spike . Immediately prior to measurement, the samples were split into a natural fraction and an aliquot spiked with the SBL74. The natural and doubled spiked fractions were run in separate batches separated by cleaning of the sample introduction system. Reproducibility of Pb isotope ratios was estimated based on 39 measurements of SRM NBS 981 run in 2017-2018. This produced average ratios of 206 Pb/ 204 Pb = 16.9402 ± 0.0031 (2σ), 207 Pb/ 204 Pb = 15.4964 ± 0.0025 (2σ) and 208 Pb/ 204 Pb = 36.7118 ± 0.0073 (2σ). Pb blanks are less than 85 pg.

Major and trace elements
Compositions and ages of the samples used for this study are summarized in Appendix A. Whole rock major and trace element concentrations are listed in Table 1, with selected elements plotted on Harker diagrams in Fig. 2 and as a time series in Fig. 3. SiO 2 of Sakurajima ranges from 46.7 wt% in the basalts, through andesite and dacite, to 72.3 wt% in the rhyolite. Figure 2 shows CaO, Al 2 O 3 , Fe 2 O 3 and MgO all decrease linearly with increasing SiO 2 , while Na 2 O and K 2 O increase. TiO 2 contents (Fig. 2) show a general decline with increasing SiO 2 , but are very similar for the andesites, dacites and post 1955 andesites, clustering between 0.65 and 0.81 wt%. P 2 O 5 content is more variable across the range of SiO 2 contents (Fig. 2).
Key incompatible trace element and rare-earth element concentrations, normalized to primitive mantle, are shown in Fig. 4. Similar profiles are observed across the One difference observed between the basalts and the andesites/dacites is that they are less enriched in Cs, Rb and K relative to Ba.

Pb isotopes
High-resolution Pb isotope data for Sakurajima is presented in Table 2. As all samples (other than the Shirahama basalt) were erupted in the last 30,000 years, no age correction was applied to the data. Overall, Sakurajima forms linear arrays in 2 0 7 Pb/ 2 0 4 Pb vs 2 0 6 Pb/ 2 0 4 Pb and 208 Pb/ 204 Pb vs 206 Pb/ 204 Pb (Fig. 5) which plot significantly above the NHRL with Δ7/4Pb and Δ8/4Pb of~12 and 76 respectively. The basalts have less radiogenic Pb  Sun and McDonough (1989) isotope ratios than the more evolved rocks, with the Sumiyoshiike basalt having the lowest ratios of 2 0 6 P b / 2 0 4 P b = 1 8 . 2 4 , 2 0 7 P b / 2 0 4 P b = 1 5 . 5 5 a n d 208 Pb/ 204 Pb = 38.31. Sakurajima and the Aira caldera rhyolites have very similar Pb isotope ratios with 206 Pb/ 204 Pb 18. 40-18.42. When viewed at a finer scale (Fig. 5), the Pb isotope data shows that andesites erupted between 4 and 3.7 ka and in 1995 plot to more radiogenic compositions than the remaining samples.

Crustal assimilation and fractional crystallization
As basalts have not erupted or been recorded from Sakurajima, the composition of the unfractionated magmas is difficult to measure directly. In this case, we utilize basalts erupted in the vicinity of Sakurajima volcano as a proxy for primitive magma compositions feeding the system. The Sumiyoshiike basalts erupted at 7.2 ka (Shibata et al. 2013a) are found less than 4 km outside the Aira caldera (see Fig. 1) and hence may represent eruptions that have bypassed the Sakurajima magma reservoir. An alternative is the Shirahama basalts, which make up a significant component of the volcanism in the area before the formation of Aira caldera, potentially representing mafic magmas from a less mature, more primitive volcanic system. Although these two regional basalts have similar elemental and isotopic compositions, the Shirahama basalt lies outside the age range of Aira-Sakurajima volcanism. On this basis, we have chosen the Sumiyoshiike basalts as the best primitive magma proxy. Their inclusion in a sample set of mafic magmas from Southern Kyushu deemed to have been negligibly affected by shallow crystallization and contamination processes (Shinjo et al. 2000) further support this.
In comparison to the Sumiyoshiike basalts, the eruptive products of Sakurajima (hereon taken to include the Aira caldera rhyolite) have more radiogenic Pb (Fig. 5). This suggests that the Sumiyoshiike basalts and Sakurajima magmas are either derived from mantle sources with different levels of subduction addition or have experienced variable amounts of assimilation of radiogenic crust. To distinguish between these two possibilities, we have examined the Pb isotope systematics of these magmas in relation to mantle, subduction and crustal components in Fig. 6. Subducting sediments and the Philippine Sea ocean crust have been well characterized by drilling and dredging (Hickey-Vargas 1998; Shimoda et al. 1998) while the mantle wedge is constrained from numerous studies across the Philippine Plate and Japan Sea as being a variant of Indian Ocean MORB (e.g. Pouclet et al. 1994;Hickey-Vargas 1998). Upper crust beneath Sakurajima is composed of the Shimanto supergroup sediments, while the lower crust is composed of Miocene granitoids (Hosono et al. 2008). This is supported by seismic studies that indicate a granodioritic lower crustal composition beneath Southern Kyushu (Iwasaki et al. 1990). The Takakumayama granite is a middle Miocene granite exposure located less than 25 km south-east of Sakurajima (Matsui 2006;Ishihara and Chappell 2010), and this is used here to represent the lower crustal composition beneath Sakurajima.
From Figs. 6 and 7, it is apparent that there is an essentially linear trend of progressively more radiogenic compositions through the sequence: basalts, Aira, Sakurajima, Takakumayama granite and Shimanto sediments. This suggests that incorporation of lower crustal granitic or Shimanto sedimentary material into the basaltic magma could result in the Pb isotope compositions of Sakurajima. In contrast, the subducted PSP sediments are not collinear with this sequence and lie to lower Δ7/4Pb and Δ8/4Pb at higher 206 Pb/ 204 Pb (Fig. 6). Figure 7 shows a mixing line between Sumiyoshiike basalt and "average crust" (estimated as 50% Takakumayama granite, 50% average Shimanto sediments) below Sakurajima. Incorporation of crustal materials into magmas can occur via direct assimilation of crustal xenoliths, or incorporation and mixing of a partial melt of the crust (Grove et al. 1988). Due to the lack of xenoliths in Sakurajima's eruptive products, the partial melting model (favoured by Kuritani et al. 2005;Ishizuka et al. 2017; and others) is utilized here. Figure 8 shows that Sakurajima samples plot within error of this mixing line, with their isotopic signature corresponding to the addition of 5-6% average crustal melt to the basaltic magma (see Appendix C for calculation of Pb concentration in crustal melt). Our high-precision analyses allow this mixing trend to be tightly constrained and differences in proportions of mixing components as low as 1% to be resolved. If assimilation of bulk crust is assumed, the amount of crustal material added increases to around 10%.
The Sumiyoshiike basalt is used as the mixing end member because it represents the best available proxy for primitive magmas feeding Sakurajima volcano. However, it could be argued that the intermediate position of the basalts between the mantle wedge and the crustal components in Fig. 6 indicates that the basalts are themselves affected by some degree of crustal contamination. The mixing line in Fig. 7 can be extrapolated further back to other less radiogenic compositions (red trend line on Fig. 6), suggesting that the Pb/ 204 Pb and c, d 208 Pb/ 204 Pb vs 206 Pb/ 204 Pb, for the eruptive products of Sakurajima, Aira caldera and basalts. The Northern Hemisphere Reference Line (NHRL) was calculated using the equations of Hart (1984). Plots b and d are higher resolution versions of a and c respectively. The error ellipse is calculated using starting values of 206 Pb/ 204 Pb = 18.4207, 207 Pb/ 204 Pb = 15.6126 and 208 Pb/ 204 Pb = 38.6774. The ellipse is based on the reproducibility of Pb isotope ratios quoted in the "Methods" section Sumiyoshiike basalts represent primitive magma compositions produced by initial crustal contamination of primary mantle-derived magmas. These primary arc magmas may themselves lie amidst an array formed between subducted sediment, subducted ocean crust and the mantle wedge. Crustal addition estimated for the Sakurajima samples would therefore be accordingly greater if calculated using the less radiogenic mantle-derived magmas.
Mixing lines between basalt and granite, and basalt and average sediments do not overlap with sample error ellipses (Fig. 8), suggesting that addition of only one crustal component does not produce the sample compositions. It is noted that extrapolation of the basalt-average crust mixing line towards more radiogenic compositions causes it to pass through the compositional field defined by the Shimanto sediments (Fig. 8). However, the Shimanto sediments show considerable  Ishizuka et al. (2010). Average Indian Ocean MORB-average composition from compiled data of Meyzen et al. (2007), excluding areas known to be contaminated by mantle plumes. Shimanto sediments- . Takakumayama granite-average composition of samples from Shin (2008). Indian Ocean MORB field (green) and Pacific and Atlantic MORB field (yellow) based on data from Meyzen et al. (2007). Thin red trend line shows collinearity of basalts, Sakurajima-Aira samples and crustal compositions. Plots c and d show Δ7/4 Pb vs isotopic heterogeneity, suggesting that their average may best represent a bulk composition. Explaining the Sakurajima isotope systematics by only adding Shimanto sediments would require a sediment composition different to any existing samples. Hence, "average crust" is considered the more appropriate mixing end member, with the resultant mixing line overlapping with almost all Sakurajima sample error ellipses in both 206 Pb/ 204 Pb-207 Pb/ 204 Pb and 206 Pb/ 204 Pb-208 Pb/ 204 Pb space (Fig. 8).
This assimilation model is consistent with Hosono et al. (2008), who proposed that the Sr-Nd-Pb isotopic signatures of volcanic rocks across Southern Kyushu result from contamination of mafic magma with lower crustal granodioritic and upper crustal sedimentary materials. Significant crustal contamination does not occur in the northern Izu arc (Taylor and Nesbitt 1998) and is not considered a major contributor to the Pb isotope signature of the North-East Japan arc (Hanyu et al. 2006). The higher Δ7/4Pb and Δ8/4Pb of Sakurajima and other Kyushu volcanoes compared to the North Izu and North-East Japan arcs (Fig. 6) is likely to be a product of a crustal signature superimposed on the supra-subduction primary magmas in this region. Figure 9 shows that, in general terms, Sakurajima has relatively consistent Pb isotope ratios (~18.4) through a range of SiO 2 and MgO. This is consistent with this suite of rocks being related by fractional crystallization from an isotopically similar precursory magma. The fractionation vectors on these plots project back to intersect the basalt-average crust mixing line at~5% crustal melt. As the Sumiyoshiike basalts have less radiogenic Pb than this intersection, it implies that they have experienced less crustal contamination than the precursory magma. In other words, the magma feeding directly into the upper crustal Sakurajima-Aira reservoir was likely to have been already contaminated deeper in the crust. This is again consistent with the model for Southern Kyushu volcanics proposed by Hosono et al. (2008), who suggested that higher silica rocks in the region formed via fractional crystallization from a contaminated parent magma. Figure 8 examines in more detail the Pb isotopic variations within Sakurajima. At this high resolution, the dacite samples are isotopically indistinguishable from the 1914, 1946 and post 1955 andesites in terms of their overlapping uncertainty ellipses and essentially equivalent average isotope ratios. With the exception of marginally higher 208 Pb/ 204 Pb, the Aira caldera rhyolite also has an equivalent isotope composition. There is however an isotopic distinction between these samples and the 4-3.7 ka andesites, particularly apparent in 206 Pb/ 204 Pb-207 Pb/ 204 Pb (Fig. 8) where there is a clean separation between the uncertainty ellipses of these groups. The 1995 eruption has a more radiogenic signature than the other post 1955 magmatism and the 4-3.7 ka andesites, yet still plots on or within error of the basalt-crust mixing line. The overall elemental composition of the 1995 eruption closely resembles the 1978 and 2000 eruptions, but its Pb isotopes indicate a greater amount of crustal contamination.
It is also noted that the composition of eruptions has become progressively more mafic through the last~500 years, in accordance with the findings of Uto et al. (2005) and  Takahashi et al. (2013). Figure 3 shows a dacitic composition at~0.5 ka with 1.5% MgO, through andesites with 2-3% MgO, to the recent eruptions with > 3% MgO. Potentially, this could reflect an increase in the magma throughput in the shallow fractionating reservoir, whereby the more frequent incursion of mafic material progressively weights the overall composition. During this progressive change in major element composition, the Pb isotopes generally are invariant with 206 Pb/ 204 Pb~18.40, with the exception of the 1995 eruption with 206 Pb/ 204 Pb~18.42 (Fig. 10). This exception suggests that magma supply for individual eruptions can be drawn from a reservoir that has experienced greater crustal contamination.
Elemental-isotopic systematics resulting from assimilationcrystallization in this system are examined in higher resolution in Fig. 9. There appears to be a common fractionation trajectory represented by the samples with 206 Pb/ 204 Pb~18.40 that stems from a~5% crustal melt addition to the initial basalt. The 4-3.7 ka andesites and 1995 andesite fit with a contribution of ≥ 6% crust. An alternative explanation to variable quantities of crust in the initial magma is a variation in the proportions of different crustal components. For example, uneven contributions from the upper crustal Shimanto sediments and the lower crustal granites/granodiorites would impart a different Pb isotope signature on the contaminated magma. These magmas could also have assimilated an unknown crustal material of more radiogenic composition, though this cannot be confirmed without better constraining the crustal composition below the volcanic system. Regardless of the contamination systematics, this re-affirms that the supply to individual eruptions, or phases of volcanism, can be sourced from distinct magmas contaminated to different extents. Such distinct magma compositions are only distinguished using this high-precision Pb isotope data and may otherwise go undetected.
Sakurajima incompatible trace element concentrations (e.g. Ba, Rb, La) define a wide range and generally increase with increasing SiO 2 , again consistent with fractional crystallization (Appendix D). The lack of correlation between Pb isotopes and incompatible trace element contents (Appendix D) could be explained by the significant compositional variation in the Shimanto sediments (e.g. 257-689 ppm Ba, 75-232 ppm Rb). This trace element heterogeneity defines a much wider range than the isotopic heterogeneity (Appendix D). Assimilation of sediments with relatively similar isotopic yet highly variable trace element compositions could produce Plots e and f demonstrate the processes generating the end members of the mixing that produces the magmas erupted in the twentieth century (and possibly earlier eruptions). e Contaminated mafic precursory magma undergoes fractional crystallization, producing a silicic magma with the same Pb isotope ratio, stored in the Aira caldera reservoir. f Twentieth century: new mafic precursory magma mixes with the silicic magma producing the andesitic magmas of the samples. Mixing end members are the mafic precursory magma, composed of approximately 95% Sumiyoshiike basalt, 5% average crustal melt and silicic/dacitic magma, inferred to have a composition similar to sample YMA 1, the most recently erupted dacite isotopically similar mafic precursory magmas with a range of trace element contents that are then further modified by fractional crystallization. Alternatively, the fact that the samples plot on fractionation trend away f r o m th e b a s a l t -a v e r a g e c r u s t m i x i n g l i n e o n 2 0 6 Pb/ 2 0 4 Pb vs incompatible trace element plots (Appendix D) suggests fractional crystallization alone from precursory magmas with consistent trace element compositions could produce the observed variation. Hence, trace elements may not correlate with extent of crustal contamination, as demonstrated by the 4-3.7 ka and 1995 andesites.
A potential cause of Pb isotopic variation at Sakurajima other than crustal contamination is mantle heterogeneity. Although mantle-derived magmas can show considerable compositional variation, as indicated by the Indian and Pacific MORB fields in Fig. 6 (Doucet et al. 2004) to act as a suitable component.
The mantle wedge should be the bulk contributor to primitive arc magmas which enter the lithosphere beneath Sakurajima. However, in terms of Pb, the contribution of the MORB-like mantle prior to the addition of fluids from the slab may be reduced or minor (e.g. Straub et al. 2010) given that Ce/Pb is very low in arc magmas (~1-8) relative to typical mantle ratios (~25). This slab fluid, and the Pb enrichment, originates from the subducting sediments and ocean lithosphere. As such, primitive arc magma is likely to have Pb isotopes that are essentially derived from the melting of a MORB-like mantle that has been hybridized by these slab components (Taylor and Nesbitt 1998).
In the case of Sakurajima, the slab component generating the hybrid mantle is likely to be a PSP sediment-PSP ocean crust mixture (Shinjo et al. 2000). Pb isotope compositional variation within the PSP ocean crust is shown in Fig. 6 and is essentially coincident with Indian MORB. PSP sediments have more radiogenic compositions with 206 Pb/ 204 Pb~18.7 and lower Δ8/4 relative to Sakurajima. Given these compositions, any fluid-mantle hybridization which integrates PSP sediment, PSP ocean crust and the Indian MORB mantle wedge will not result in the Sakurajima Pb isotope trend.
t could be argued that addition of a more radiogenic subducted sediment with a composition similar to the crustal components used here could explain the isotopic variation without the need for assimilation. PSP sediments have been extensively sampled and measured (Hickey-Vargas 1998;Shimoda et al. 1998), but it is possible that a hitherto unsampled, and isotopically appropriate sedimentary unit exists within the encroaching slab. However, as the regional basaltic lavas are less radiogenic ( 206 Pb/ 204 Pb~18.25) than the andesites and dacites ( 206 Pb/ 204 Pb~18.4), it is likely that the isotopic variation is generated during crystal fractionation and assimilation in the crust rather than a mantle wedge process.
It is possible that the primary mantle-derived arc magmas may have experienced varying levels of PSP sediment addition. For example, a greater PSP sediment component could give rise to more radiogenic primitive magmas that go on to assimilate crust. In this scenario, the composition of precursory magmas feeding the Sakurajima reservoir would be expected to be more radiogenic, which could offer an alternative explanation to variable levels of crustal addition in producing the higher Pb isotope ratios of the 4-3.7 ka andesites and 1995 eruption. Despite this, the proximity of the sample compositions to the basalt-average crust mixing line, and lack of erupted evidence for more radiogenic primitive magmas, suggests that greater crustal contamination is a more realistic explanation.

Eruptive products since 1955
Activity at Sakurajima since 1955 is characterized by frequent Vulcanian explosions (Iguchi 2013). This contrasts with past intermittent larger scale eruptions, suggesting that different processes may be occurring in order to produce this change (1) Addition of fluids from the subducting PSP to the mantle wedge causes partial melting. Primitive basaltic magmas form from primary melts.
(2) As high-temperature basaltic magma ascends, it induces melting of crustal material. These crustal melts are assimilated by the basaltic magma, in both the lower and upper crusts. (3) Basaltic magma reaches the Aira caldera reservoir with a modified isotopic composition following contamination with crustal melts. This mafic precursory magma then undergoes fractional crystallization, producing silicic magma. More contaminated precursory magmas give rise to more contaminated silicic magmas. Silicic magmas are stored in the Aira caldera reservoir. (4) Injection of mafic magma into stored silicic magmas (and associated mixing) triggers migration from Aira caldera to the shallow Sakurajima reservoir and subsequent eruption. Depths of the brittle-ductile transition and base of the lower crust from Iwasaki et al. (1990) and Kneller et al. (2008), respectively in style of activity. Studies of phenocrysts provide evidence for mixing between basaltic and more silicic magmas for the Taisho (1914) and Showa (1946) eruptions (Yanagi et al. 1991;Nakagawa et al. 2011), along with post 1955 eruptions (Nakagawa et al. 2011;Ebihara et al. 2013). This can be reconciled with the Pb isotope data if the basaltic end member of the mixing is taken as the mafic precursory magma that has assimilated crustal materials. Silicic magmas derived from this precursory magma by fractional crystallization would have very similar isotopic ratios to the precursory magma (Fig. 9). The products of mixing of two such magmas would therefore have the same/similar isotopic ratio as the end members, with major element compositions between those of the end members (Fig. 9), lying on an identical trend to that of fractional crystallization. The presence of mafic inclusions in the An-ei dacites suggests that injection and mixing of basaltic magma into more silicic magma also occurred in 1779 (Yanagi et al. 1991). Such studies have not examined older eruptions; hence, the extent to which these involved mixing cannot be constrained.
The lower SiO 2 and higher MgO of the post 1955 andesites indicate a greater basaltic contribution to the final erupted mixed magma relative to the 1914 and 1946 eruptions. This can only be determined from major and trace element data and is not distinguishable isotopically, since the post 1955 mafic end member appears to be contaminated with crust to the same extent as precursory magmas feeding older eruptions. However, the isotopic data does identify that, in the case of the 1995 eruption, at least one of the mixing end member magmas was more crustally contaminated. The incompatible trace element concentrations (e.g. Ba, Rb, Pb) of this sample are lower than other post 1955 eruptions, closer to the basalts. This suggests that this eruption likely involved a significant mafic component (possibly one less enriched in incompatible trace elements than Sumiyoshiike basalt), reducing incompatible trace element and silica content, mixed with a more contaminated silicic magma (derived from an older, more contaminated mafic precursory magma), producing the more radiogenic isotopic composition.
The only discernible change in petrogenetic processes since 1955 is an increased mafic component in erupted mixed magmas, likely due to increased throughput of mafic magma to the Aira caldera reservoir. This supports the conclusions of Nakagawa et al. (2011), who suggest that recent Vulcanian eruptions at Sakurajima are triggered by frequent injection of basaltic magma, which has caused the change in style of activity. Crustal contamination processes prior to crystallization and/or mixing in the Aira caldera reservoir appear to have remained consistent across the post 1955 transition. Individual eruptions and phases of activity involving more radiogenic magmas have occurred both pre and post 1955, suggesting that differences in extent of crustal contamination are not linked to the change in style of activity. Figure 11 shows a schematic model for the petrogenesis of Sakurajima-Aira magmas, based on the findings detailed above combined with geophysical observations. Seismic data has demonstrated the existence of a shallow magma reservoir below Sakurajima (3-6 km depth) and deeper magma reservoir below Aira caldera (approximately 10 km depth) (Iguchi 2013). Primitive basaltic magma assimilates both lower and upper crustal materials, prior to reaching the Aira caldera magma reservoir. Hence, crustal assimilation occurs at depths greater than 10 km in the magma plumbing system. The contaminated mafic precursory magma then undergoes fractional crystallization, producing silicic magmas that are stored in the reservoir. Variation in the extent to which precursory magma is contaminated gives rise to silicic magmas of variable Pb isotope compositions (Fig. 11). Fresh injection of mafic magma into stored silicic magmas likely triggers eruptions (Nakagawa et al. 2011;Ebihara et al. 2013). Studies of vertical ground deformation and observed migration of A-type earthquake hypocentres to shallower depths provide evidence for magma migration to the shallow reservoir below Sakurajima during eruptive episodes (Iguchi 2013). Increased mafic magma throughput since 1955 has resulted in eruption of mixed magmas with an increasing basaltic component, producing the recent trend towards lower silica eruptive products.

Petrogenetic model for Sakurajima volcano
The Aira caldera rhyolite, dacites erupted at 10 ka, 1471 and 1779, and post 1914 eruptive products all share a similarly contaminated precursor magma and/or mixing end members. More contaminated magmas were involved in the petrogenesis of the andesites erupted between 4 and 3.7 ka and in 1995.

Conclusions
This study uses high-resolution Pb isotopes to assess crustal assimilation and temporal changes in magma petrogenesis in the Sakurajima-Aira volcanic system. The Sakurajima-Aira caldera volcano has significantly more radiogenic Pb isotopes and higher Δ7/4Pb and Δ8/4Pb than local basaltic eruptions. This can be attributed to assimilation of lower and upper crustal materials, represented in terms of Pb isotope compositions by the locally exposed Takakumayama granite and Shimanto sediments respectively, by these primitive basaltic magmas. Assimilation occurs prior to differentiation and mixing processes at depths greater than 10 km.
The Pb isotope ratios of the Sakurajima-Aira system eruptive products show little variation with increasing SiO 2 and decreasing MgO content, suggesting they are derived from fractional crystallization of a similarly contaminated precursor magma. This precursor is consistent with addition of 5-6% average crustal melt to a primitive basaltic magma. When the high-resolution Pb isotopes are examined in detail, it is recognized that andesites erupted between 4-3.7 ka and the 1995 eruption contain a greater proportion of crustal contamination. This suggests that distinct magma compositions were available to feed particular eruptions or eruptive phases. Only from such high-precision data can these variations in the extent of crustal contamination be detected. This demonstrates the value of high-resolution Pb isotope analysis in assessing short-term changes in magma-crust interaction.
Final erupted magmas post 1955, along with those from 1914 and 1946 (and possibly older eruptions), were produced by mixing of basaltic (contaminated precursory magma) and stored silicic magmas with essentially identical isotopic ratios. Andesitic pyroclastics erupted since 1955 typically have lower SiO 2 and higher MgO than older lava flows, but (mostly) display equivalent Pb isotope ratios. This can be attributed to a greater throughput of contaminated mafic magma to the Aira caldera reservoir since 1955, resulting in eruption of mixed magmas with a greater mafic component. The extent of crustal contamination before and after 1955 has remained consistent; hence, it is unlikely to be a contributing factor to the transition in activity style.
Application of high-resolution Pb isotopes to other arc volcanoes may reveal subtle variations in the magmas feeding eruptions that previously went undetected by elemental and lower resolution isotopic analyses. Combining this technique with major element and phenocryst data as in this study also provides a tool with which to explore whether there exists a relationship between crustal contamination and surface activity in volcanic systems.